COPPOLINO

Series Editor: Kristin B. Zimmerman, SEM

The Integrated Test Analysis Process for

THE INTEGRATED TEST ANALYSIS PROCESS FOR STRUCTURAL DYNAMIC SYSTEMS

Structural Dynamic Systems
Robert N. Coppolino, Measurement Analysis Corporation

Over the past 60 years, the U.S. aerospace community has developed, refined, and
standardized an integrated approach to structural dynamic model verification and validation.
One name for this overall approach is the Integrated Test Analysis Process (ITAP) for
structural dynamic systems. ITAP consists of seven sequential tasks, namely: (1) definition
of test article finite element models; (2) systematic modal test planning; (3) measured data
acquisition; (4) measured data analysis; (5) experimental modal analysis; (6) systematic
test-analysis correlation; and (7) reconciliation of finite element models and modal test
data. Steps 1, 2, and 7 rely strictly on mathematical model disciplines, and steps 3 and 4
rely on laboratory disciplines and techniques. Current industry practice of steps 5 and 6
calls for interaction of mathematical model and laboratory disciplines, which compromises
the objectivity of both modeling and laboratory disciplines. This book addresses technical
content, strategies, and key relevant experiences related to all steps of ITAP, except for
measured data acquisition which is the specialized domain of highly experienced laboratory
professionals who contend with mechanical and electrical practicalities of instrumentation,
excitation hardware, and data collection systems.

About SYNTHESIS
This volume is a printed version of a work that appears in the Synthesis Digital Library of
Engineering and Computer Science. Synthesis books provide concise, original presentations of

MORGAN & CLAYPOOL

important research and development topics, published quickly, in digital and print formats.
The Integrated Test Analysis
Process for Structural
Dynamic Systems
 Synthesis SEM Lectures on
Experimental Mechanics
Editor
Kristin Zimmerman, SEM
Synthesis SEM Lectures on Experimental Mechanics follow the technical divisions and the
direction of The Society for Experimental Mechanics (SEM). The SEM is composed of
international members of academia, government, and industry who are committed to
interdisciplinary application, research and development, education and active promotion of
experimental methods to: (a) increase the knowledge of physical phenomena; (b) further the
understanding of the behavior of materials, structures and systems; and (c) provide the necessary
physical basis and verification for analytical and computational approaches to the development of
engineering solutions. The members of SEM encompass a unique group of experimentalists,
development engineers, design engineers, test engineers and technicians, and research and
development scientists from industry and educational institutions working in materials; modeling
and analysis; strain measurement and structural testing.

The Integrated Test Analysis Process for Structural Dynamic Systems

Robert N. Coppolino
2019

Structure-Property Relationships under Extreme Dynamic Environments: Shock

Recovery Experiments
Cyril L. Williams
2019

Mechanics of Materials Laboratory Course

Ghatu Subhash and Shannon Ridgeway
2018

The Old and New... A Narrative on the History of the Society for Experimental Mechanics
Cesar A. Sciammarella and Kristin B. Zimmerman
2018

Hole-Drilling Method for Measuring Residual Stresses

Gary S. Schajer and Philip S. Whitehead
2018
Copyright © 2020 by Morgan & Claypool

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in
any form or by any means—electronic, mechanical, photocopy, recording, or any other except for brief quotations
in printed reviews, without the prior permission of the publisher.

The Integrated Test Analysis Process for Structural Dynamic Systems

Robert N. Coppolino
www.morganclaypool.com

ISBN: 9781681736815 paperback

ISBN: 9781681736822 ebook
ISBN: 9781681736839 hardcover

DOI 10.2200/S00955ED1V01Y201910SEM005

A Publication in the Morgan & Claypool Publishers series

SYNTHESIS SEM LECTURES ON EXPERIMENTAL MECHANICS

Lecture #5
Series Editor: Kristin Zimmerman, SEM
Series ISSN
Print 2577-6053 Electronic 2577-6088
The Integrated Test Analysis
Process for Structural
Dynamic Systems

Robert N. Coppolino
Measurement Analysis Corporation

SYNTHESIS SEM LECTURES ON EXPERIMENTAL MECHANICS #5

M
&C Morgan & cLaypool publishers
ABSTRACT
Over the past 60 years, the U.S. aerospace community has developed, refined, and standard-
ized an integrated approach to structural dynamic model verification and validation. One name
for this overall approach is the Integrated Test Analysis Process (ITAP) for structural dynamic
systems. ITAP consists of seven sequential tasks, namely: (1) definition of test article finite
element models; (2) systematic modal test planning; (3) measured data acquisition; (4) mea-
sured data analysis; (5) experimental modal analysis; (6) systematic test-analysis correlation; and
(7) reconciliation of finite element models and modal test data. Steps 1, 2, and 7 rely strictly
on mathematical model disciplines, and steps 3 and 4 rely on laboratory disciplines and tech-
niques. Current industry practice of steps 5 and 6 calls for interaction of mathematical model
and laboratory disciplines, which compromises the objectivity of both modeling and labora-
tory disciplines. This book addresses technical content, strategies, and key relevant experiences
related to all steps of ITAP, except for measured data acquisition which is the specialized do-
main of highly experienced laboratory professionals who contend with mechanical and electrical
practicalities of instrumentation, excitation hardware, and data collection systems.

KEYWORDS
model, test plan, data analysis, modal analysis, correlation, reconciliation
 vii

Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii

1 Overview of the Integrated Test Analysis Process (ITAP) . . . . . . . . . . . . . . . . . . 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 System Dynamic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Modal Test Plan and Test-Analysis Correlation . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.4 Measured Data Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.5 Measured Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.6 Experimental Modal Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.7 Model Updating (Reconciliation) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.8 General References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.9 Key Illustrative Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.10 Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Definition of Test Article Finite Element Models . . . . . . . . . . . . . . . . . . . . . . . . 7

2.1 Part 1: Variational Foundations of Modern Structural Dynamics . . . . . . . . . . . 7
2.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.2 Economy in Nature and Basic Variational Formulations . . . . . . . . . . . . 8
2.1.3 Mathematical Physics and Hamilton’s Principle . . . . . . . . . . . . . . . . . . 9
2.1.4 The Contributions of Ritz, Galerkin, and Trefftz . . . . . . . . . . . . . . . . . 9
2.1.5 The Ritz Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.6 Galerkin’s Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.7 Trefftz’s Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.8 Automated Formulations in Structural Dynamics . . . . . . . . . . . . . . . . 11
2.1.9 The Finite Element Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.10 Assembly of Linear Structural System Models . . . . . . . . . . . . . . . . . . 12
2.1.11 System Models with Localized Nonlinearities . . . . . . . . . . . . . . . . . . 12
2.1.12 Matrix Structural Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.13 Free Vibration and Modal Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.14 Uncoupled Structural Dynamic Equations . . . . . . . . . . . . . . . . . . . . . 13
2.1.15 Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
viii
2.2 Part 2: Guidelines for Systematic Model Development . . . . . . . . . . . . . . . . . . 14
2.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.2 Linear Single-Degree-of-Freedom Equations . . . . . . . . . . . . . . . . . . . 14
2.2.3 Linear Multi-Degree-of-Freedom Equations . . . . . . . . . . . . . . . . . . . 15
2.2.4 Normal Modes of Undamped MDOF Systems . . . . . . . . . . . . . . . . . 16
2.2.5 Response to Impulsive and Transient Force Excitations . . . . . . . . . . . 17
2.2.6 Response Spectrum and Shock Spectrum . . . . . . . . . . . . . . . . . . . . . . 18
2.2.7 Illustrative Example: Half-Sine Pulse Force . . . . . . . . . . . . . . . . . . . . 18
2.2.8 Illustrative Example: 1940 El Centro CA Ground Motion Data . . . . 19
2.2.9 Definition of Relevant Structural Dynamic Models . . . . . . . . . . . . . . 19
2.2.10 Illustrative Example: Aluminum Launch Vehicle and Spacecraft
Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2.11 Illustrative Example: Aluminum Launch Vehicle Feedline . . . . . . . . . 22
2.2.12 Modal Density and the Effectiveness of Finite Element Models . . . . 23
2.2.13 Illustrative Example: Fluid-Filled Circular Cylindrical Shell . . . . . . . 23
2.2.14 Free Surface Slosh Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2.15 Interface Flexibility in Structural Assemblies—A Retrospective . . . . 30
2.2.16 Damping in Structural Assemblies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.2.17 Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.3 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3 Systematic Modal Test Planning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3.1 Part 1: Understanding Modal Dynamic Characteristics . . . . . . . . . . . . . . . . . . 37
3.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.1.2 Illustrative Example: ISS P5 Short Spacer . . . . . . . . . . . . . . . . . . . . . . 38
3.1.3 Illustrative Example: Axisymmetric Shell Finite Element Model . . . 40
3.1.4 Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2 Part 2: Target Mode Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.2.2 Structure Excited by Boundary Accelerations and Applied
External Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.2.3 Modal Effective Mass and Target Mode Selection . . . . . . . . . . . . . . . 49
3.2.4 Illustrative Example: ISS P5 Short Spacer . . . . . . . . . . . . . . . . . . . . . . 50
3.2.5 Modal Gains and Target Mode Selection . . . . . . . . . . . . . . . . . . . . . . 51
3.2.6 The Mode Acceleration Method and Target Mode Selection . . . . . . . 51
3.2.7 Illustrative Example: Axisymmetric Shell Finite Element Model . . . 53
3.2.8 Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
 ix
3.3 Part 3: Response DOF Selection for Mapping Experimental Modes . . . . . . . 58
3.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.3.2 Selection of DOFs for Modal Mapping (the RKE Method) . . . . . . . 59
3.3.3 Illustrative Example: ISS P5 Short Spacer . . . . . . . . . . . . . . . . . . . . . . 60
3.3.4 Allocation of Modal Excitation Resources . . . . . . . . . . . . . . . . . . . . . . 63
3.3.5 Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

4 Measured Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

4.1 Part 1: Preliminary Measured Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.1.2 Key Preliminary Measured Data Analysis Functions, Mean,
Variance, and Standard Deviation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.1.3 Normalized Probability Density and Ideal Gaussian Distribution . . . 66
4.1.4 Total Normalized Probability Function . . . . . . . . . . . . . . . . . . . . . . . . 67
4.1.5 Autospectrum or Power Spectral Density Function . . . . . . . . . . . . . . 67
4.1.6 Cross-Spectrum or Cross-Power Spectral Density Function . . . . . . . 69
4.1.7 The Spectrogram Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.1.8 Illustrative Example: Sinusoidal Time History with Background
Random Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.1.9 Illustrative Example: SDOF Linear System Response to Random
Excitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.1.10 Illustrative Example: ISPE Modal Test . . . . . . . . . . . . . . . . . . . . . . . . 71
4.1.11 Illustrative Example: Wire Rope Test Article . . . . . . . . . . . . . . . . . . . 72
4.1.12 Illustrative Example: ISS P5 Modal Test . . . . . . . . . . . . . . . . . . . . . . . 72
4.1.13 Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.2 Part 2: Frequency Response Function Estimates from Measured Data . . . . . . 75
4.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.2.2 MI/MO Frequency Response Equations for Systems with
Algebraic Nonlinearities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.2.3 SI/SO, MI/SO and MI/MO Frequency Response Function
Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.2.4 Illustrative Example: ISPE Modal Test . . . . . . . . . . . . . . . . . . . . . . . . 79
4.2.5 Illustrative Example: Wire Rope MI/SO Test Data Analysis . . . . . . . 81
4.2.6 Illustrative Example: ISS P5 Modal Test . . . . . . . . . . . . . . . . . . . . . . . 81
4.2.7 Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.3 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
x
5 Experimental Modal Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.1 Part 1: Preliminary Experimental Modal Analysis . . . . . . . . . . . . . . . . . . . . . . 85
5.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.1.2 Frequency Response Characteristics of Linear SDOF Systems . . . . . 85
5.1.3 Frequency Response Characteristics of Linear MDOF Systems . . . . 88
5.1.4 Illustrative Example: ISS P5 Modal Test . . . . . . . . . . . . . . . . . . . . . . . 89
5.1.5 Illustrative Example: ISPE Modal Test . . . . . . . . . . . . . . . . . . . . . . . . 89
5.1.6 Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.2 Part 2: The Simultaneous Frequency Domain Method . . . . . . . . . . . . . . . . . . 92
5.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.2.2 Effective Dynamic Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.2.3 The SFD Method Prior to 2018 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.2.4 Illustrative Example: ISS P5 Modal Test . . . . . . . . . . . . . . . . . . . . . . . 95
5.2.5 SFD 2018: A Fresh Look at Experimental Modal Analysis . . . . . . . . 99
5.2.6 Illustrative Example: ISPE Modal Test . . . . . . . . . . . . . . . . . . . . . . . 100
5.2.7 Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.3 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

6 Systematic Test Analysis Correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

6.1 Part 1: Conventional Mass Weighted Correlation of Test and FEM Modes 107
6.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.1.2 Derivation of Mass Weighted Test-Analysis Correlation Metrics . . 107
6.1.3 Illustrative Example: ISS P5 Modal Test . . . . . . . . . . . . . . . . . . . . . . 108
6.1.4 Illustrative Example: ISPE Modal Test . . . . . . . . . . . . . . . . . . . . . . . 110
6.1.5 Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.2 Part 2: Correlation of Test and FEM Modes Using Left-Hand Eigenvectors 112
6.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
6.2.2 Modal Test Data Resulting from Estimated State-Space Models . . 112
6.2.3 Theoretical System Modes in State-Space Form . . . . . . . . . . . . . . . . 113
6.2.4 Correlation of Experimental and Theoretical Modes (Complex
Least Squares) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
6.2.5 Illustrative Example: ISPE Modal Test . . . . . . . . . . . . . . . . . . . . . . . 115
6.2.6 Roadmap for a Highly Improved Integrated Test Analysis Process . 116
6.2.7 Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
6.3 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
 xi
7 Reconciliation of Finite Element Models and Modal Test Data . . . . . . . . . . . 123
7.1 Part 1: Finite Element Model Modal Sensitivity . . . . . . . . . . . . . . . . . . . . . . 123
7.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
7.1.2 Sensitivity Analysis Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
7.1.3 Residual Vector Augmentation for Local Alterations . . . . . . . . . . . . 125
7.1.4 Residual Mode Augmentation (RMA) for Dispersed Alterations . . 126
7.1.5 RMA Solution Qualities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
7.1.6 Illustrative Example: ISPE Modal Test . . . . . . . . . . . . . . . . . . . . . . . 129
7.1.7 Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
7.2 Part 2: Test-Analysis Reconciliation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
7.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
7.2.2 Modal Cost Function .C / and Optimal Test-Analysis
Reconciliation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
7.2.3 Illustrative Example: Rectangular Plate Supported by Corner Posts 134
7.2.4 Illustrative Example: ISS P5 Modal Test . . . . . . . . . . . . . . . . . . . . . . 137
7.2.5 Illustrative Example: ISPE Modal Test . . . . . . . . . . . . . . . . . . . . . . . 138
7.2.6 Illustrative Example: Wire Rope Test Article . . . . . . . . . . . . . . . . . . 140
7.2.7 Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
7.3 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

Author’s Biography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

 xiii

Preface
The Integrated Test Analysis Process (ITAP) for structural dynamic systems, presented in this
book, offers in-depth expositions of six key process steps (recognizing the fact that measured
data acquisition is the domain of highly specialized professionals, possibly requiring a completely
separate book to properly cover that discipline).
Definition of an appropriate test article finite element model (FEM) involves (a) deter-
mination of the anticipated operational system’s dynamic environments (specifically their fre-
quency bandwidth and intensity envelopes). In the case of launch vehicles and spacecraft, NASA
and USAF Space Command organizations specify typical bandwidths associated with dynamic
environments (generally at or below 70 Hz for environments that do not include acoustic and
shock load phenomena). Once the bandwidth of the dynamic environment is established, spatial
resolution requirements for the test article FEM’s structural components are defined based on
upper frequency bound to wavelength relationships. As the FEM must be suited for adjustment
of uncertain features, it must include definition of joints and component interfaces in a man-
ner consistent with engineering drawings. The wealth of theoretical and experimental resources,
especially for shell-type structures, offers opportunities to intelligently restrict attention to key
sensitivity parameters.
Development of an effective modal test plan requires careful study of the test article’s FEM
predicted vibration modes. Specifically, close attention should be paid to analytical modal kinetic
energy distributions in addition to modal frequencies and geometric mode shapes. Appreciation
of the fact that modal kinetic energy is mathematically an “unpacking” of the mass-weighted or-
thogonality matrix points to the importance of modal kinetic energy to assist initial selection of
response measurement (accelerometer) allocation. A common challenge associated with shell-
type launch vehicle and spacecraft structures is the “many modes” problem that arises from the
fact that many overall shell breathing modes occur in the same frequency band as overall body
bending, torsion, and axial modes. Difficulties associated with this “many modes” problem may
be alleviated by a well-informed target mode selection process based on modal decomposition of
predicted flight dynamic events. Simplistic modal effective mass criteria, which employ target
mode selection for restricted situations involving base excitation environments, are not relevant
for more general situations. A definitive approach for allocation of response measurement (ac-
celerometer) allocation is offered by the residual kinetic energy (RKE) method, which is widely
employed in the U.S. aerospace community.
Preliminary measured data analysis employing a variety of metrics including probability
density functions, autospectra, time history and associated spectrograms, and shock spectra is an
invaluable prerequisite for detailed data analysis. Issues associated with anomalous data channels
xiv PREFACE
and unexpected (e.g., nonlinear) behavior can be noted and dealt with prior to engagement in
detailed data analysis. Multiple Input/Multiple Output (MI/MO) spectral analysis procedures
are the primary cornerstone for detailed measured data analysis. A cumulative coherence tech-
nique, based on Cholesky (triangular) decomposition of partially correlated excitation sources,
provides a systematic tool for (1) assessment of the role (prominence) of individual excitations
and (2) localization and characterization of nonlinear aspects of dynamic response (when promi-
nent). The product of measured data analysis is estimated frequency response functions (FRFs)
and accompanying coherence functions.
Experimental modal analysis is a discipline that benefits from techniques developed dur-
ing the analog era (prior to 1971) and the digital era (after 1971). Intuitive graphical procedures
for preliminary experimental modal analysis owe much of their content to analog era technol-
ogy as well as newer procedures that highlight overall modal content (generally termed modal
indicator functions); this represents the last opportunity for correction of problematic FRF data
prior to detailed experimental modal analysis. A wide range of experimental modal analysis tech-
niques have been developed during the post 1971 digital era. The techniques fall into two dis-
tinct categories, namely: (1) curve fitting procedures and (2) effective dynamic system estimation
procedures. Simultaneous Frequency Domain (SFD) techniques belong to the latter category.
Recent challenges encountered in NASA MSFC’s Integrated Spacecraft and Payload Element
(ISPE) modal test in 2016, associated with the “many modes” problem led to development of
the SFD-2018 technique. This latest SFD innovation possesses a variety of features that alleviate
the “many modes” challenge. Specifically, SFD-2018 validates estimated complex modes by a
decoupling operation that defines single mode (SDOF equivalent) FRFs. This SFD-2018 oper-
ation is reminiscent of multi-shaker tuning, single mode isolation techniques developed during
the analog era, without requiring multi-shaker tuning. In addition, since SFD-2018 automati-
cally computes left-hand eigenvectors of an estimated state-space plant, the product of left- and
right-hand eigenvector matrices automatically produces a mathematically perfect orthogonality
matrix without reliance on a possibly flawed Test Analysis Mass (TAM) matrix. This feature of
SFD-2018 alleviates common difficulties associated with satisfaction of both NASA STD-5002
and USAF Space Command SMC-S-004 test mode orthogonality criteria.
TAM mass matrix dependent test mode orthogonality and test-analysis cross-
orthogonality criteria specified in NASA STD-5002 and USAF Space Command SMC-S-004
are widely used in the U.S. aerospace community. The most commonly employed strategy for
systematic test analysis correlation involves employment of “real” experimental modes that are
defined based on real mode curve fitting and/or approximate test modes constructed from the
real component of estimated complex test modes. In most situations, this strategy is deemed
appropriate. The NASA/MSFC ISPE modal test appears to present severe challenges to the
commonly employed strategy. In response to this difficulty, a new complex test mode-based
test-analysis cross-orthogonality procedure, which is independent of the TAM mass matrix,
was developed. This provides further alleviation of difficulties presented by commonly employed
 PREFACE xv
cross-orthogonality criteria. Further analysis of ISPE test data suggests that ambiguities may
occur as a result of employment of real test mode approximations. Specifically, complex mode
orthogonality “unpacking” operations that produce test mode orthogonality distributions that
are not necessarily in agreement with the commonly employed “real” test mode strategy. This
difficulty led to introduction of an alternative “roadmap for a highly improved integrated test
analysis process.”
Real test mode-based reconciliation of FEMs and modal test data depends upon accurate,
efficient parametric sensitivity analysis of the test article FEM and employment of robust modal
cost functions. Augmentation of baseline model modes with a set of residual modes (called
the residual mode augmentation (RMA) method) has been shown to eliminate unsatisfactory
compromises inherent in a popular technique called structural dynamic modification (SDM).
Employment of a balanced modal cost function was first successfully applied for test analysis
reconciliation as part of the ISS P5 modal test conducted at NASA MSFC in 2001. It was
determined at that time that Nelder–Meade Simplex optimization did not perform satisfactorily,
while a Monte Carlo search strategy offered a robust search option. An exercise demonstrating
hysteretic nonlinear system identification employing minimization of a time history based error
norm is included as a final application. Nonlinearity and complex test modes are possibly the
next challenges to be addressed in the continuing adventure called the integrated test analysis
process for structural dynamic systems.

Robert N. Coppolino
October 2019
 1

CHAPTER 1

Overview of the Integrated

Test Analysis Process (ITAP)
1.1 INTRODUCTION
Over the past 60 years, the U.S. aerospace community has developed, refined, and standard-
ized an integrated approach to structural dynamic model verification and validation. One name
for this overall approach is the Integrated Test Analysis Process (ITAP) for structural dynamic
systems, which is summarized in Figure 1.1.

System Dynamic Model

Measured Data
Modal Test Plan
Acquisition

Measured Data
Analysis

Test-Analysis Experimental Modal

Correlation Analysis

Model Updating
(Reconciliation)

Figure 1.1: Overview of the integrated test analysis process.

The yellow-colored steps are associated with analytical-centric disciplines, while the aqua-
colored steps are associated with laboratory-centric disciplines. The mixed-colored steps involve
a blend of analytical and laboratory disciplines. Each step in the process is the product of a variety
of authoritative, widely recognized sources. This book is intended to serve readers who have some
familiarity with the foundations of structural dynamics and modal testing. A comprehensive
2 1. OVERVIEW OF THE INTEGRATED TEST ANALYSIS PROCESS (ITAP)
treatment of the background subject matter for the seven steps is well beyond the scope of
this book. In lieu of a thorough exposition of the seven ITAP steps, primary sources of their
foundations are denoted herein.

1.2 SYSTEM DYNAMIC MODEL

System dynamics modeling rests on fundamental mathematical concepts in physics, particularly
variational principles, structural mechanics, structural dynamics, matrix structural analysis, and
the finite element method (FEM). Key references in these subject matters may be found in the
following classic and modern publications.
1. R. B. Lindsay and H. Margenau, Foundations of Physics, Dover Publications, 1957.
2. Y. C. Fung, Foundations of Solid Mechanics, Prentice Hall, 1965.
3. S. P. Timoshenko, History of Strength of Materials, McGrawHill, 1953.
4. R. R. Craig and A. J. Kurdila, Fundamentals of Structural Dynamics, 2nd ed., Wiley, 2006.
5. H. Norman Abramson, The Dynamic Behavior of Liquids in Moving Containers, NASA-
SP-106, 1966.
6. R. H. MacNeal, Finite Elements: Their Design and Performance, Marcel–Dekker, 1994.
7. O. C. Zienkiewicz, R. L. Taylor, and J. Z. Zhu, The Finite Element Method, Its Basis and
Fundamentals, 6th ed., Elsevier, 2005.
8. L. Cremer, M. Heckl, and E. E. Ungar, Structure Borne Sound, Springer-Verlag, 1973.
Among the above key references, the work of Abramson is particularly noteworthy in that
a thorough exposition of fluid-structure interaction (hydroelasticity) is not treated in the other
key references. Hydroelasticity is particularly important in aerospace applications, as roughly
80% of a launch vehicle’s lift-off mass is liquid propellants (unless the propellants are solid).
Also noteworthy is Structure Borne Sound, which singularly treats the empirical foundations of
damping in structures (subject matter generally ignored in the literature).

1.3 MODAL TEST PLAN AND TEST-ANALYSIS

CORRELATION
The modal test plan and test-analysis correlation, particularly in aerospace applications, are re-
sults of key contributions documented in many technical papers since 1960. Among the variety
of background materials, the following references are noteworthy.
9. D. R. Martinez and A. Keith Miller, Ed., Combined Experimental/Analytical Modeling of
Dynamic Structural Systems, ASME-AMD-67, 1985.
 1.4. MEASURED DATA ACQUISITION 3
10. Load Analysis of Spacecraft and Payloads, NASA-STD-5002, 1996.

11. U.S. Air Force Space Command, Independent Structural Loads Analysis, SMC-S-004,
2008.

The 1985 book edited by Martinez and Miller documents the thought process at that
time, which focused on hybrid analytical-experimental dynamic models. The “hybrid” approach
focused on direct incorporation of experimental modal data (in lieu of reconciliation of experi-
mental and analytical information), resulting in stringent requirements for modal test planning
and test-analysis correlation that are incorporated in NASA-STD-5002 and SMC-S-004.

1.4 MEASURED DATA ACQUISITION

Measured data acquisition is a technical discipline and practice that is the domain of highly
experienced specialists. While some reference materials are included in background literature
in Measured Data Analysis and General Sources, the most comprehensive guidance is to be
found in the community of experienced specialists.

1.5 MEASURED DATA ANALYSIS

The foundations of measured data analysis, in this book, are documented in the following books.

12. J. S. Bendat and A. G. Piersol, Random Data Analysis and Measurement Procedures, 4th ed.,
Wiley, 2010.

13. J. S. Bendat, Nonlinear Systems Techniques and Applications, 2nd ed., Wiley, 1998.

It should be noted that the sign convention employed by Bendat and Piersol for complex
numbers and phase angle differ from most prevailing literature. The sign convention in the
present book conforms to prevailing practice. Additional key references are listed in General
Sources.

1.6 EXPERIMENTAL MODAL ANALYSIS

Experimental modal analysis has a rich historical background that is documented in three key
sources, namely:

14. D. L. Brown and R. J. Allemang, The Modern Era of Experimental Modal Analysis, Sound
and Vibration Magazine, January 2007.

15. R. E. D. Bishop and G. M. L. Gladwell, An Investigation into the Theory of Resonance

Testing, Philosophical Transactions, Royal Society of London, Series A, 225(A-1055):241–280,
1963.
4 1. OVERVIEW OF THE INTEGRATED TEST ANALYSIS PROCESS (ITAP)
16. D. J. Ewins, Modal Testing Theory and Practice, Wiley, 1984.
That being said, the primary emphasis in this book’s chapter on experimental modal anal-
ysis may be found in the following publications.
17. R. N. Coppolino, A Simultaneous Frequency Domain Technique for Estimation of Modal Pa-
rameters from Measured Data, SAE Paper 811046, 1981.
18. R. N. Coppolino, Efficient and Enhanced Options for Experimental Mode Identification,
IMAC 21, 2003.
19. R. N. Coppolino, Experimental Mode Verification (EMV) using Left-Hand Eigenvectors,
IMAC 37, 2019.
While the simultaneous frequency domain technique (SFD) is featured in this book, no
implication as to its superiority over the variety of excellent experimental modal analysis tech-
niques is implied.

1.7 MODEL UPDATING (RECONCILIATION)

The general subject of model updating (test-analysis reconciliation) is intimately tied to ana-
lytical approximation of structural dynamic model parametric sensitivity. It is most important
to first cite techniques that are ultimately unreliable due to (a) the presence of closely spaced
or repeated modes, (b) large uncertainties in structural joint (stiffness) parameters, and (c) sub-
stantial errors associated with modal truncation; the following published works describe those
techniques in this category:
20. R. I. Fox and M. P. Kapoor, Rates of Change of Eigenvalues and Eigenvectors, AIAA
Journal, 6, 1968.
21. R. B. Nelson, Simplified Calculation of Eigenvector Derivatives, AIAA Journal, 14, 1976.
22. P. Avitabile, Twenty Years of Structural Dynamic Modification-A Review, IMAC 20, 2002.
The technique that appears to efficiently and effectively circumvent deficiencies (a)–(c) is
documented in the following.
23. R. Coppolino, Methodologies for Verification and Validation of Space Launch System (SLS)
Structural Dynamic Models, NASA CR 2018-219800, 2018.
The reconciliation process is completed via incorporation of an optimization (error norm
minimization) procedure. Three leading strategies designed to accomplish reconciliation are
found in the following.
24. T. K. Hasselman, D. C. Zimmerman, and D. L. Herendeen, An Integrated FEA Software
Capability for Dynamic Model Validation and Verification, AIAA 40th SDM Conference,
1999 (Bayesian Statistics).
 1.8. GENERAL REFERENCES 5
25. J. A. Nelder and R. Mead, A Simplex Method for Function Minimization, Computer
Journal, 7(4):308–313, 1965 (Gradient Search).
26. N. Metropolis, The Beginning of the Monte Carlo Method, Los Alamos Science (1987 Spe-
cial Issue dedicated to Stanislaw Ulam), pp. 125–130 (Random Search).

1.8 GENERAL REFERENCES

Two additional key references that offer expositions related to most aspects of the Integrated
Test Analysis Process for structural dynamic systems are as follows.
27. A. G. Piersol and T. L. Paez, Eds., Harris’ Shock and Vibration Handbook, 6th ed.,
McGraw-Hill, 2010.
28. R. Allemang and P. Avitabile, Eds., Handbook of Experimental Structural Dynamics,
Springer-Verlag, 2021.

1.9 KEY ILLUSTRATIVE EXAMPLES

Five key illustrative examples are employed throughout this book.
1. Abramson Model: An empty and fluid-filled circular cylindrical shell (modeled in closed
form) serves to introduce (a) the “many modes” problem, (b) important parametric sensi-
tivities, and (c) the profound effect of fluid mass on the frequency and content of structural
dynamic modes.
2. ISS P5 Modal Test: End-to-end application of the integrated test analysis project com-
pleted in 2001 details (a) modal test planning, (b) measured data analysis, (c) experimental
modal analysis, (d) test-analysis correlation, and test-analysis reconciliation, and (e) tech-
niques for identification of nonlinear test article behavior.
3. Axisymmetric Shell Structure: A FEM is used to describe approaches for (a) understand-
ing and characterizing modal dynamic characteristics, (b) selection of modal test target
modes, and (c) evaluation of the effects of imperfections and local features (modal sensi-
tivity) on otherwise axisymmetric systems.
4. ISPE Modal Test: This NASA/MSFC modal test program was completed in 2016. Inde-
pendent review activities consisted of (a) development of a sensitivity analysis convergence
test, (b) development of advanced experimental modal analysis methodology, and (c) for-
mulation of a roadmap for a highly improved integrated test analysis process.
5. Wire Rope Test: Testing of a shock and vibration isolator test article, conducted as part
of a U.S. Army Phase II SBIR ending in 2008, serves as an example for (a) measured data
analysis and (b) system identification for a highly nonlinear dynamic system.
6 1. OVERVIEW OF THE INTEGRATED TEST ANALYSIS PROCESS (ITAP)
1.10 CLOSURE
This book is intended to provide details of the Integrated Test Analysis Process, principally from
the perspective of the author’s experience over the past 52 years. It presupposes the reader’s basic
knowledge of structural dynamics technology (engineering course work, hopefully supplemented
by some graduate research and professional experience).
The previously cited references serve as essential recommended background materials. This
book, to be sure, is a personal essay. It is certainly not intended to overshadow the excellent con-
tributions of many individuals and organizations engaged in the field of analytical and experi-
mental structural dynamics. The content of this book does not follow the usual logical sequence
that is typical of foundational textbooks on structural dynamics and random data analysis. In-
stead, subject matters relevant to each of the seven key ITAP steps are presented in a manner
that supports the content of each sequential chapter.
 7

CHAPTER 2

Definition of Test Article

Finite Element Models
2.1 PART 1: VARIATIONAL FOUNDATIONS OF MODERN
STRUCTURAL DYNAMICS

2.1.1 INTRODUCTION
The development of mathematical physics, of which structural dynamics is a branch, owes its
present state to two general viewpoints. The first viewpoint due to Newton [1] sees nature fol-
lowing postulates describing the dynamic equilibrium of interacting bodies. The second view-
point, due to d’Alembert [2], Hamilton [3], and Lagrange [4], sees nature following postulates
describing efficient organization of energies (variational principles). The variational viewpoint,
specifically Hamilton’s principle, has guided the development of partial differential equations
and natural boundary conditions for technical structural theories [5, 6].
Difficulties encountered in the quest for exact solutions of partial differential equations,
subjected to natural boundary conditions, led to the introduction of approximate techniques
based on Hamilton’s principle. Ritz [7] employed assumed shape functions and generalized co-
ordinates to deduce mass and stiffness matrices, which are foundational to the FEM [8] and Ma-
trix Structural Analysis [9]. Galerkin [10] similarly applied assumed functions to a variational
integral associated with a system’s partial differential equations; his formulation, while influential
in development of the FEM, has also been applied to solve nonlinear dynamic problems [11].
Trefftz [12] introduced a novel variational technique focusing on approximate satisfaction of
natural boundary conditions, which are a byproduct of Hamilton’s principle; his employment
of shape functions that automatically satisfy the system’s partial differential equations forms a
basis of the boundary element method [13].
Hamilton’s principle and the Ritz method, in particular, have been instrumental in the
development of both the FEM and Matrix Structural Analysis. The basic building blocks of the
FEM are most often developed on the basis of assumed boundary displacement referenced and
optional interior displacement shape polynomials, which are used to define element mass and
stiffness matrices based on the Ritz method. Assembly and analysis of structural system models
falls in the category of Matrix Structural Analysis, which for a time before general acceptance
of the FEM employed a force (degree of freedom) method.
8 2. DEFINITION OF TEST ARTICLE FINITE ELEMENT MODELS
Free vibration analysis of a matrix structural system is typically performed on an undamped
system, for which the modes or real eigenvectors satisfy a mass weighted orthogonality relation-
ship. Free vibration analysis of a damped structural system is rarely performed due to difficulties
associated with theoretical damping matrices that do not relate to empirical modal damping.
An alternative to the explicit damping matrix, namely a complex structural damping constant,
is often employed in aeroelastic analysis [14]. The more widely used description of damping is
based on critical damping ratio factors assigned to a truncated set of undamped system modes.

2.1.2 ECONOMY IN NATURE AND BASIC VARIATIONAL

FORMULATIONS
Philosophers of antiquity, before the advent of modern science initiated by Newton (1686),
believed that nature operated in accordance with a rule of economy as noted by Aristotle (d. 312
BC): “Nature follows the easiest path that requires the least amount of effort.” In the medieval
era, William of Ockham (1347) suggested an economic principle for human reasoning with
his famous saying, “It is futile to employ many principles when it is possible to employ fewer”
(popularly known as Ockham’s razor). It is fascinating that Newton’s second law may be used
as a postulate to deduce variational formulations of mechanics (as theorems). Moreover, when
the variational principle due to Hamilton (1824), is postulated, Newton’s second law follows as
a theorem.
The postulate-theorem interrelationship is summarized in Table 2.1 for a mechanical sys-
tem composed of either (1) a single particle, i.e., single-degree-of-freedom (SDOF) system or
(2) a multi-degree-of-freedom (MDOF) matrix system. Newton’s second law states the foun-
dational postulate, sequentially followed by d’Alembert’s Principle and Hamilton’s Principle, as
“theorems.”

Table 2.1: Hierarchal relationship of theoretical foundations of classical dynamics

SDOF Systems MDOF (Matrix) Systems
Theoretical Foundation
Basic Equations Supporting Information Basic Equations Supporting Information

M !Ü" = !F"
d du
Newton’s 2nd Law (1689) F = — (m — )
dt dt

(F = — (m du !δU" M !Ü" = !δU" !F"

d’Alembert’s Principle d T T
— )) • δu = 0
(1743) dt dt
Classical Statements Integration by Parts Classical Statements Integration by Parts
t2 t2 t2 t2 t2 t2
#(F - — (m du #$δ$– m (—)& + F • δu& • dt - (m —) • δu' = 0 # !δU" !F-MÜ" • dt = {0} #$δ$— !U̇"M !U̇"+!δU"•!F"&•dt - !δU" M !U̇"'=!0"
2
d 1 du du T 1 T T
— )) • δu • dt = 0
t
dt dt t
2 dt dt t t1 t1 2 t1
1 1 1

t2 Kinetic Energy and Virtual Work Kinetic Energy and Virtual Work
#*δT + δW + • dt = 0
t2

T = — m (—
du ) # *δT + δW+ • dt = 0 T = — !U̇" M !U̇" δW = !U̇" !F"
2 T T
Hamilton’s Principle 1 1
δW = F • δu
(1824) t1 2 dt t1 2
Potential Energy, Non-Conservative Work, Lagrangian Potential Energy, Non-Conservative Work, Lagrangian
!F" = - –—– + !FNC"FNC
t2
#*δL + δW + • dt = 0
dV t2 d{V}
# *δL + δW+ • dt = 0
F = - –– + FNC d{u}
du
δW → !δU" !FNC"
t1 δW → FNC δu T
t1
L=T -V L=T -V
 2.1. PART 1: VARIATIONAL FOUNDATIONS OF MODERN STRUCTURAL DYNAMICS 9
2.1.3 MATHEMATICAL PHYSICS AND HAMILTON’S PRINCIPLE
Application of Hamilton’s principle to a dynamic system described as a continuum yields a vol-
ume integral of the type
Zt2 Z
.ıTR ıUR C ıWR /
dR
dt D 0; (2.1)
t1 R

where TR , UR , and ıWR are the kinetic energy, potential (or strain) energy, and virtual work
functions per unit volume, R, respectively. Analysis of any particular dynamic system, described
in terms of displacement variables, u.x; y; z; t /, which may be vectors, results in the following
type of functionals:
Zt2 Z Zt2 Z
.P.D.E./
ıu
dR
dt C .N.B.C./
ıu
dS
dt D 0; (2.2)
t1 R t1 S

“P.D.E.” represents the particular partial differential equation(s) within the system’s volume.
“N.B.C.” represents the natural boundary conditions, which are mathematically and physically
admissible on the system’s boundary surface(s), S . The general process for derivation of a system’s
partial differential equations and natural boundary conditions has provided a consistent basis
for the development of technical structural theories for prismatic bars, beams, rings, plates, and
shells.

2.1.4 THE CONTRIBUTIONS OF RITZ, GALERKIN, AND TREFFTZ

Three outstanding contributions that led to the development of approximate analysis techniques
date back to the early part of the twentieth century. The methods bearing the names of Ritz,
Galerkin, and Trefftz are all consequences of Hamilton’s principle and the assumption of ap-
proximate solution functions.

2.1.5 THE RITZ METHOD

A monumental contribution to approximate analysis was introduced by Ritz (1908) [7], who
described the displacement field in variable separable terms
N
X
u.x; y; z; t / D ‰n .x; y; z/
qn .t /; (2.3)
nD1

where ‰n .x; y; z/ are assumed shape functions and qn .t/ are temporal generalized coordinates
(displacements). In addition, the strain field,
N
X
".x; y; z; t/ D ‰";n .x; y; z/
qn .t /; (2.4)
nD1
10 2. DEFINITION OF TEST ARTICLE FINITE ELEMENT MODELS
is linearly related to the assumed displacement field employing the appropriate partial deriva-
tives. It should be noted that exact closed form solutions of partial differential equations are often
expressed in variable separable form, whenever such solutions are possible. By assuming a series
of functions that satisfy particular boundary conditions (or generally permit solution of natural
boundary conditions), substitution of Equation (2.3) into Hamilton’s principle (Equation 2.1),
the following symmetric matrix equations are deduced:
ŒM  fqg
R C ŒK fqg D Œ€ fQg ; (2.5)
where the positive semi-definite, symmetric mass, and stiffness matrix terms are
Z Z
Mmn D ‘‘”
‰m ‰n dR; Kmn D ‘‘E”
‰";m ‰";n dR: (2.6)
R R

[Note: “” and “E ” are representative of mass density and elastic stiffness material properties.] In
addition, the generalized forcing terms are governed by volume and surface integrals associated
with applied body forces and surface loads, respectively.
The Ritz method, outlined above, was initially employed to approximately solve difficult
problems described by partial differential equations and associated natural boundary conditions.
Ultimately, it was extensively applied in development of the Finite Element Method [8] and
Matrix Structural Analysis [9].

2.1.6 GALERKIN’S METHOD

Galerkin [10] defined an approximate method using the variable separable displacement field
and associated generalized coordinates (Equations (2.3) and (2.4)) by substitution of the as-
sumed functions into Equation (2.2) where the boundary conditions are automatically satisfied
by choice of an appropriate set of spatial functions. The general statement of Galerkin’s method
is Z
.P.D.E./
ıu
dR D 0: (2.7)
R

An appealing aspect of Galerkin’s method is that it can be applied to any set of partial
differential equations (even if a suitable variational formulation is unknown). The method has
been successfully applied in the study of nonlinear dynamic systems [11].

2.1.7 TREFFTZ’S METHOD

Trefftz [12] proposed an approximate method that employs a set of assumed functions that
automatically satisfy the partial differential equations. Therefore, the form of Equation (2.2)
that must be satisfied is Z
.N.B.C./
ıu
dS D 0: (2.8)
S
 2.1. PART 1: VARIATIONAL FOUNDATIONS OF MODERN STRUCTURAL DYNAMICS 11
This method was generally ignored for about three decades until its advantage in the ap-
proximate solution of infinite domain problems was recognized. Trefftz’s method has been in-
strumental in the development of the boundary element method.

2.1.8 AUTOMATED FORMULATIONS IN STRUCTURAL DYNAMICS

Among the methodologies that owe their foundations to Hamilton’s principle and the contri-
butions of Ritz, Galerkin, and Trefftz, two leading techniques are of prominence, namely (a) the
Finite Element Method and (b) systematic Matrix Structural Analysis. Discussion of these two
topics will be limited to linear systems.

2.1.9 THE FINITE ELEMENT METHOD

The Finite Element Method encompasses three mathematical disciplines, namely: (1) definition
of building blocks (finite elements), (2) assembly of structural system models, and (3) solution
of structural system equations. The third discipline is addressed by systematic Matrix Structural
Analysis.
Finite elements are generally derived as a specialized application of the Ritz method. The
displacement field associated with a single element is most generally described in terms of the
following family of shape functions:
Nh Np
X X
u.x; y; z; t/ D ‰h;n .x; y; z/
un .t/ C ‰p;n .x; y; z/
qn .t/
nD1 nD1 (2.9)
or in matrix form,
 ˚
fug D Œ‰h  fuh g C ‰p qp :

The “h” shape functions are referenced to physical displacements at specific grid points
on the element’s (boundary) surface and the “p ” shapes are polynomial functions that have null
value along the element’s boundary surface. Substitution of the above shape family functions
(and their appropriate strain partial derivatives) into Hamilton’s principle (Equation 2.1) results
in positive semi-definite mass and stiffness matrices of the forms
   
Mpp Mph Kpp Kph
ŒM element D ; ŒKelement D : (2.10)
Mhp Mhh element Khp Khh element

It should be emphasized here that definition of well-posed and accurate finite elements
depends upon expert selection of shape functions and numerical integration schemes. The most
commonly employed finite elements in commercial finite element codes is the “h” element which
does not include “p ” generalized coordinates. The more general elements are often called “h-p ”
elements.
12 2. DEFINITION OF TEST ARTICLE FINITE ELEMENT MODELS
2.1.10 ASSEMBLY OF LINEAR STRUCTURAL SYSTEM MODELS
Assembly of a linear system structural dynamic model involves the allocation and superposition
(overlapping) of individual finite elements onto a system degree-of-freedom (DOF) map. This
process defines sparse system mass and stiffness matrices that are positive semi-definite and
symmetric. A system composed of an assembly of “h” elements is defined by “grid” set mass
and stiffness matrices denoted by ŒM  and ŒK, respectively. Additional quantities complete the
ingredients for a structural dynamic model, namely: (1) allocation of excitation forces to system
grid points, (2) formation of an assumed viscous damping matrix (which unfortunately does not
resemble physical reality in most commercial finite element codes), and (3) allocation of local
nonlinear internal forces. The grid set equations for the structural dynamic model are of the for
˚ ˚ ˚

ŒM  UR C ŒB UP C ŒK fU g D Œ€e  fFe g C Œ€N  FN UN ;UP N ; p ;
˚ ˚ (2.11)
fUN g D Œ€N T fU g ; UP N D Œ€N T UP :

Constraints and boundary conditions, collectively described by transformations of the

form,
fug D ŒG fua g ; (2.12)
are applied in a symmetric manner as a consequence of the quadratic forms and integrals defined
by the Ritz method, resulting in the “analysis” set equations and matrices (for a linear system)

ŒMaa  fuR a g C ŒBaa  fuP a g C ŒKaa  fua g D Œ€ae  fFe g ; (2.13)

where the reduced order matrices are

 T   T 
ŒMaa  D Gga Mgg Gga ; ŒBaa  D Gga Bgg Gga ;
 T   T   (2.14)
ŒKaa  D Gga Kgg Gga ; Œ€a  D Gga €g :

2.1.11 SYSTEM MODELS WITH LOCALIZED NONLINEARITIES

Structural assemblies sometimes display locally nonlinear behavior (primarily at structural joints
that connect subassemblies). The “analysis” set equations for such situations are augmented by
localized nonlinear forces, fFN g, allocated in accordance with the nonlinear force distribution
matrix, Œ€aN , are of the form

ŒMaa  fuR a g C ŒBaa  fuP a g C ŒKaa  fua g D Œ€ae  fFe g C Œ€aN  fFN g : (2.15)

The nonlinear forces are defined as functions of localized displacements and velocities that
are defined as  T   T 
fuN g D €aN fua g ; fuP N g D €aN fuP a g : (2.16)
In the most general case, localized nonlinear forces are “hysteretic” functions of localized
displacements, velocities, and (path dependent) parameters, fpg, according to algorithms of the
 2.1. PART 1: VARIATIONAL FOUNDATIONS OF MODERN STRUCTURAL DYNAMICS 13
following general form:
fFN .t/; p.t C dt/g D ff .uN .t/; uP N .t/; p.t/g : (2.17)
For the more restricted cases in which the parameters are not path dependent, the localized
nonlinear forces are “algebraic,” rather than “hysteretic.”

2.1.12 MATRIX STRUCTURAL ANALYSIS

Matrix Structural Analysis predates the Finite Element Method by several decades. Histori-
cally, matrix formulations for structural systems developed along two paths, namely (a) the force
method and (b) the displacement method. The displacement method, based primarily on Hamil-
ton’s principle and the Ritz method, ultimately eclipsed the force method due to the advent of
the finite element method. Matrix structural analysis procedures, applied to damped systems
that include localized nonlinear forces, most often employ an efficient transformation defined
by free vibration modal analysis of an undamped system. [Note: In the following exposition on
modal analysis, the “a” set subscript employed in Equations (2.12)–(2.16) is dropped.]

2.1.13 FREE VIBRATION AND MODAL ANALYSIS

Free vibration of an undamped structural system is described by solutions of the type
fu.t /g D fˆn g
sin .!n t/ or using complex exponentials fu.t/g D fˆn g
e i !n t ; (2.18)
which define the real eigenvalue problem


ŒK fˆn g ŒM  fˆn g n D f0g n D !n2 ; (2.19)
which has as many independent eigenvectors, fˆn g (normal modes) and eigenvalues .n D !n2 /,
as the system matrix order. The collection of all or a truncated subset of normal modes (the
modal matrix) defines the modal transformation
fug D Œˆ fqg : (2.20)
The modal matrix has the following mathematical properties (for unit mass normalized
modes):
ŒˆT ŒM  Œˆ D ŒI  ; ŒˆT ŒK Œˆ D Œ ; (2.21)
where Œ is a diagonal matrix of eigenvalues.

2.1.14 UNCOUPLED STRUCTURAL DYNAMIC EQUATIONS

Application of the modal transformation to Equations (2.15) and (2.16) results in
     
R C ˆT Bˆ fqg
fqg P C Œ fqg D ˆT €e fFe g C ˆT €N fFN g ;
 T   T  (2.22)
fuN g D €N ˆ fqg ; fuP N g D €N ˆ fqg
P :
14 2. DEFINITION OF TEST ARTICLE FINITE ELEMENT MODELS
 
In general, the “modal” damping matrix, ˆT Bˆ , is fully coupled. However, an uncou-
pled approximation (based on a wide variety of experimental results) is quite common, resulting
in the fully uncoupled modal dynamic equations,
   
qR n C 2n !n qP n C !n2 qn D ˆTn €N fFN g C ˆTn €e fFe g : (2.23)

2.1.15 CLOSURE
The theoretical foundations of finite element analysis and matrix structural analysis, which rely
on systematic variational principles, have been outlined. Free vibration modes of undamped, lin-
ear structural dynamic systems have been shown to uncouple general MDOF structural dynamic
matrix equations with local nonlinear features.

2.2 PART 2: GUIDELINES FOR SYSTEMATIC MODEL

DEVELOPMENT
2.2.1 INTRODUCTION
Definition of the relevant structural dynamic model of a vehicle or stationary system begins with
determination of the frequency band associated with anticipated dynamic service environments.
Shock and response spectra [15] offer a general approach for estimation the frequency band.

2.2.2 LINEAR SINGLE-DEGREE-OF-FREEDOM EQUATIONS

An elementary linear structural dynamic or mechanical dynamic system is described in terms
of a SDOF linear mechanical system. The dynamic response, u.t/, of a SDOF system which is
excited by an applied force, F .t /, and/or base (foundation) motion input, u0 .t/, is governed by
the ordinary differential equation (schematic depicted in Figure 2.1),
mu.t/
R C b u.t
P / C ku.t / D F .t / C b uP 0 .t/ C ku0 .t/; (2.24)
where m; b , and k are the constant mass, viscous damping, and elastic stiffness coefficients,
respectively. By defining the relative displacement variable, uR .t/ D u.t/ u0 .t/, and dividing
by the mass, m, Equation (2.1) simplifies to
uR R .t / C 2n !n uP R .t/ C !n2 uR .t / D F .t /=m uR 0 .t/; (2.25)
where the undamped natural frequency (rad/sec) and critical damping ratio, respectively, are
r
k
!n D ; n D b= .2m!n / : (2.26)
m
In addition, the damped natural frequency (rad/sec) is defined as
q
!d D !n 1 n2 : (2.27)
 2.2. PART 2: GUIDELINES FOR SYSTEMATIC MODEL DEVELOPMENT 15
u(t), F(t)
m

k b uo(t)

Figure 2.1: Linear SDOF dynamic system schematic.

2.2.3 LINEAR MULTI-DEGREE-OF-FREEDOM EQUATIONS

Linear structural dynamic systems are generally described in terms of partial differential equa-
tions of mathematical physics or matrix differential equations constructed as assemblies of finite
elements. The typical form for matrix (MDOF) structural dynamic equations is
˚ ˚
ŒM  UR .t / C ŒB UP .t/ C ŒK fU.t /g D Œ€e  fFe .t/g : (2.28)
Symmetric system mass, damping, and stiffness matrices, ŒM ; ŒB, and ŒK, respectively,
are associated with discrete displacement degrees of freedom, fU.t/g. The external loads, fFe .t/g
are allocated to the discrete displacements in accordance with geometric distributions described
by the (linearly independent) columns of the matrix, Œ€e .
By partitioning the displacements into “interior” and “boundary” subsets,
 
Ui
fU g D ; (2.29)
Ub
the MDOF equations are expressed as
     
Mi i Mib UR i Bi i Bi b UP i
C
Mbi Mbb UR b Bbi Bbb UP b
    
Ki i Kib Ui €i e
C D fFe g : (2.30)
Kbi Kbb Ub €be
The transformation of “interior” displacements to “relative” displacements with respect to
the “boundary” displacements is defined as
    0 
Ui Ii i Ki i 1 Kib Ui
D : (2.31)
Ub 0bi Ibb Ub
Symmetric application of this transformation to Equation (2.30) results in
  0    0 
Mi i M 0 ib UR i Bi i B 0 ib UP i
C
0
M bi M bb 0
UR b 0
B bi B bb0
UP b
     
Ki i 00 ib U 0i €i e
C 0 0 D fFe g : (2.32)
0 bi K bb Ub € 0 be
16 2. DEFINITION OF TEST ARTICLE FINITE ELEMENT MODELS
The interior partition of the transformed structural dynamics equations is
˚ ˚ ˚ ˚ ˚
ŒMi i  UR 0 i C ŒBi i  UP 0 i C ŒKi i  U 0 i D Œ€i e  fFe g ŒMi b  UR b ŒBi b  UP b : (2.33)
Note the similarity of form of SDOF (Equation (2.25)) and MDOF (Equation (2.33))
“relative” displacement relationships (with the exception of the MDOF damping term, ŒBi b ).
It should be noted that for a rigid foundation, ŒBi b  vanishes.

2.2.4 NORMAL MODES OF UNDAMPED MDOF SYSTEMS

Free vibration of an undamped, linear structural dynamic system is described by solutions of the
real eigenvalue problem (retaining the notation of Equation (2.33)),


ŒKi i  fˆi n g ŒMi i  fˆi n g n D f0g ; n D !n2 ; (2.34)
where fˆi n g are distinct, orthogonal individual eigenvectors (or mode shapes), and n D !n2 ,
are the corresponding eigenvalues (note: !n are circular natural frequencies). The collection of
all (or a truncated set of ) eigenvectors, Œˆi , defines the real mode displacement transformation,
fUi .t/g D Œˆi  fq.t/g ; (2.35)
which has the following decoupling mathematical properties (for unit mass normalized modes):
Œˆi T ŒMi i  Œˆi  D ŒI  ; Œˆi T ŒKi i  Œˆi  D Œ : (2.36)
Note that the terms of the diagonal matrix, Œ, are the real eigenvalues, n D !n2 .
Application of the real mode displacement transformation to Equation (2.34), results in
the modal equations
 
R C ˆTi Bi i ˆi fqg
fqg P C Œ fqg
 T   T ˚  T ˚
D ˆi €i e fFe g ˆi Mi b UR b ˆi Bib UP b : (2.37)
A common approximation for the “modal” damping  matrix assumes it is “uncoupled” (or
diagonal). Moreover, the right side damping term, ˆTi Bib is often assumed negligible, result-
ing in the uncoupled modal equations, which are compared directly to SDOF Equation (2.25)
as follows:
   T ˚
qR n .t/ C 2n !n qP n .t / C !n2 qn .t/ D ˆTin €i e fFe .t/g ˆi n Mi b UR b .t/ ;
# # # # . (2.38)
uR R .t / C 2n !n uP R .t/ C !n2 uR .t/ D F .t /=m uR 0 .t/:
While many more consequences of the modal transformation will be discussed throughout
this book, the clear relationship between linear SDOF system dynamics and uncoupled modal
dynamics (Equation (2.37)) will be exploited in the following sections for systematic definition
of MDOF dynamic system modeling requirements.
 2.2. PART 2: GUIDELINES FOR SYSTEMATIC MODEL DEVELOPMENT 17
2.2.5 RESPONSE TO IMPULSIVE AND TRANSIENT FORCE
EXCITATIONS
Consider the case of a linear SDOF system excited by applied loads,

mu.t
R / C b u.t/
P C ku.t/ D F .t / or u.t/ P / C !n2 u.t / D F .t/=m:
R C 2 n !n u.t (2.39)

A pure impulsive force, FO , is defined as the finite integral,

Z0C
FO D F .t /dt ; (2.40)
0

over which the force approaches infinity as the time interval approaches zero duration. When
such a loading is applied to a system, which is initially at rest, the velocity response immediately
after the impulse .t D 0C/ is
u.0C/
P D FO =m; (2.41)
and the ensuing free decay response is
q
n !n t
m!n u.t /=FO D h.t/ D e sin.!d t/; where !d D 1 n2 !n : (2.42)

Response of the SDOF system to an impulsive force, FO . / D F ./d  , is the “particular” solu-
tion,

up .t / D 0 for t < 
(2.43)
up .t / D .1=m!n / h.t  /F . /d  for t  :

This function forms the basis of the Duhamel integral formula for SDOF system response
to a general transient force excitation [15], which is the summation or integral of responses to a
continuous train of pulses,
Zt
up .t / D .1=m!n / h.t  /F . /d : (2.44)
0

For the more general case of a linear SDOF system with initial “steady-state” displacement asso-
ciated with an applied force that begins with a finite value, F .0/, an additional “homogeneous”
solution component is required to eliminate a potentially dominant transient. Specifically, the
complete response is
Zt


u.t / D .1=m!n / h.t /F ./d  C F .0/= m!n2
e n !n t
cos .!d t/ : (2.45)
0
18 2. DEFINITION OF TEST ARTICLE FINITE ELEMENT MODELS
2.2.6 RESPONSE SPECTRUM AND SHOCK SPECTRUM
Two closely related “signature” functions for linear SDOF system response to transient excitation
are the response spectrum and shock spectrum [15]. The response spectrum is associated with
response of a unit mass SDOF system excited by an applied force, F .t/, in particular,
u.t/ P / C !n2 u.t/ D F .t/:
R C 2n !n u.t (2.46)
The maximum and minimum response map linear SDOF systems to a specific force
history, F .t /, with selected value of n for the range of natural frequency, 0 < !n < !max
is defined as the response spectrum. The response spectrum is defined for extremes in dis-
placement, velocity and acceleration response. Normalized response spectra are defined (using
F0 D max.jF .t/j/) as follows:



RD .!n ; n ; F .t// D max !n2 u.t /=F0 ; min !n2 u.t /=F0
RV .!n ; n ; F .t// D Œmax .!n u.t/=F
P 0 / ; min .!n u.t
P /=F0 / (2.47)
R /=F0 / ; min .u.t/=F
RA .!n ; n ; F .t// D Œmax .u.t R 0 / :

The shock spectrum is associated with response of a linear SDOF system excited by an
applied base acceleration, in particular,
uR R .t/ C 2 n !n uP R .t/ C !n2 uR .t/ D uR 0 .t/; (2.48)
uR .t/ D u.t / u0 .t/: (2.49)
The maximum and minimum response map SDOF systems to a specific base accelera-
tion history, uR 0 .t /, with selected value of n for the range of natural frequency, 0 < !n < !max
is defined as the shock spectrum. The shock spectrum is defined for extremes in relative dis-
placement, relative velocity and absolute acceleration response. Normalized response spectra are
defined (using UR 0 D max .juR 0 .t/j/) as follows:



SD .!n ; n ; u.t//
R D max !n2 uR .t /=UR 0 ; min !n2 uR .t/=UR 0



R // D max !n uP R .t /=UR 0 ; min !n uP R .t/=UR 0
SV .!n ; n ; u.t (2.50)



R // D max u.t
SA .!n ; n ; u.t R /=UR 0 ; min u.t/=
R UR 0 :

2.2.7 ILLUSTRATIVE EXAMPLE: HALF-SINE PULSE FORCE

Consider the half-sine pulse force excitation, shown in Figure 2.2. The normalized displacement
response spectrum (assuming n D 0:02) for this excitation is also shown in Figure 2.2.
A more commonly employed plot of response spectrum is based on the peak absolute
values of the function, as depicted in Figure 2.3.
A useful property of the absolute displacement response spectrum is that its asymptotic
value, for frequencies above f  approaches unity. The physical significance of the cut-off fre-
quency, f  , is that linear SDOF systems with natural frequency, fn > f  , respond “quasi-
statically” to the excitation environment, F .t /, i.e., the excitation is “slowly varying” for a system
 2.2. PART 2: GUIDELINES FOR SYSTEMATIC MODEL DEVELOPMENT 19
Impulsive Force Excitation and Normalized Displacement Response Spectrum
1

F(t) = sin(2̟t), for 0 ≤ t ≤ 0.5 sec;

F(t) = 0, for t > 0.5 sec.

F (lb)
0.5

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
t(sec)

2
Response Spectrum

-1

-2
0 1 2 3 4 5 6 7 8 9 10
SDOF Natural Frequency (Hz)

Figure 2.2: Half-sine pulse and normalized displacement response spectrum.

with relatively high natural frequency. This characteristic also applies to relative displacement
and absolute acceleration shock spectra.

2.2.8 ILLUSTRATIVE EXAMPLE: 1940 EL CENTRO CA GROUND

MOTION DATA
A measured ground motion time history recorded during May 18, 1940 El Centro CA earth-
quake is illustrated in Figure 2.4.
The normalized response spectrum (computed for n D 0:02), provided in Figure 2.5, sug-
gests a cut-off frequency, f  D 20 Hz for this particular event.

2.2.9 DEFINITION OF RELEVANT STRUCTURAL DYNAMIC MODELS

In order to develop a relevant dynamic model, general requirements should be addressed based
on:
(1) frequency band, 0 < f < f  , and intensity (F0 ) of anticipated dynamic environments;
and
(2) general characteristics of structural or mechanical components.
20 2. DEFINITION OF TEST ARTICLE FINITE ELEMENT MODELS
Impulsive Force: Normalized Displacement Response Spectrum

100

Response Spectrum
Response Spectrum
Low-Frequemcy Asymptote
High-Frequency Asymptote
Quasi-Static Initiation f *

10-1 -1
10 100 101 102

SDOF Natural Frequency (Hz)

Figure 2.3: Absolute normalized displacement response spectrum.

El Centro Earthquake, May 18, 1940 (NS Data)

0.4

0.3

0.2

0.1
Accel (g)

-0.1

-0.2

-0.3

-0.4
0 5 10 15 20 25 30 35 40 45 50

t(sec)

Figure 2.4: El centro NS acceleration time history.

 2.2. PART 2: GUIDELINES FOR SYSTEMATIC MODEL DEVELOPMENT 21
El Centro NS: Normalized Acceleration Shock Spectrum
3.5

Shock Spectrum
3
Quasi-Static Initiation, f*

2.5

Shock Spectrum
2

1.5

0.5

00 5 10 15 20 25
SDOF Natural Frequency (Hz)

Figure 2.5: El centro NS acceleration normalized response spectrum.

With the cut-off frequency (f  ) established based on shock and/or response spectra of
anticipated dynamic environments, the shortest relevant wavelength of forced vibration for com-
ponents in a structural assembly may be calculated. For finite element modeling, the quarter
wavelength .L=4/ is of particular interest, since it defines the grid spacing requirement needed
to accurately model dynamics [15]. Table 2.2 summarizes guidelines for typical structural com-
ponents.
In addition to the above grid spacing guidelines, the engineer must also consider limi-
tations associated with beam and plate technical theories. In particular, if the wavelength to
thickness ratio .L= h/ is less than about 10, a higher-order theory (than pure flexure) or 3-D
elasticity modeling should be considered [16]. Moreover, modeling requirements for capture of
stress concentration details may call for finer grid meshing than suggested by the cut-off fre-
quency. Finally, if the dynamic environment is of sufficiently high amplitude (F0 ), nonlinear
modeling may be required, e.g., if plate deflections are greater than thickness, h [17].

2.2.10 ILLUSTRATIVE EXAMPLE: ALUMINUM LAUNCH VEHICLE AND

SPACECRAFT COMPONENTS
Consider a cut-off frequency, f  D 50 Hz, which is typical for launch vehicle and spacecraft
primary structural loads. Assuming the structural components are aluminum (E  107 psi,
22 2. DEFINITION OF TEST ARTICLE FINITE ELEMENT MODELS
Table 2.2: Quarter wavelength relationships for typical structural components

Basis Component Wave Type L/4 Additional Data

Dilational ( E/ρ)/(4f *) E = stretch modulus
Continuum 3-D Elastic G = shear modulus
Shear ( G/ρ)/(4f*)
Mechanics B = bulk modulus
3-D Acoustic Dilational ( B/ρ)/(4f *) ρ = mass density
String Lateral ( T/ρA)/(4f *)
Axial ( E/ρ)/(4f*) T = tension
Rod A = cross-sectional area
Torsion ( G/ρ)/(4f *) EI = flexural stiffness
Technical
Theory Beam Bending (π/2)(EI/ρA)1/4 / 2π f *
Membrane Axial ( N/ρh)/(4f *) h = plate thickness
D = plate flexural stiffness
Plate Bending (π/2)(D/ρh)1/4 / 2π f * N = in-plane stress resultant

G  3:84  106 psi,   2:59  10 4 lb-sec2 /in4 ), the following quarter wavelengths .L=4/
are defined for continuum mechanics theory:
p
a. Dilational deformation: L=4 D E==.4f  /  982 in.
p
b. Shear deformation: L=4 D G==.4f  /  609 in.
Clearly, the cross-sectional dimension of rod, beam, and plate-shell components of the
largest feasible launch vehicle and spacecraft structures are substantially smaller than the above
L=4 estimates. Therefore, models based on technical theories are most feasible.

2.2.11 ILLUSTRATIVE EXAMPLE: ALUMINUM LAUNCH VEHICLE

FEEDLINE
Consider a large-scale launch vehicle’s propellant feedline with the following cross-sectional
dimensions (not representative of any particular system):

Outer diameter: OD D 15 in; wall thickness: h D 0:5 in:

The feedline’s cross-sectional parameters are A D 22:7 in2 , I D 599 in4 . Therefore, employing
the relevant technical theory formulae, the L=4 estimates are:
p
c. Axial deformation: L=4 D E==.4f  /  982 in.
p
d. Torsional deformation: L=4 D G==.4f  /  609 in.
In order to properly consider lateral bending dynamics of the propellant feedline, the
contained fluid must be considered. For the purposes of the present discussion, water is employed
 2.2. PART 2: GUIDELINES FOR SYSTEMATIC MODEL DEVELOPMENT 23
as an ersatz fluid with properties in the general range of liquid oxygen (LOX); the effective
bulk modulus of water at room temperature, accounting for radial flexibility of the feedline
structure is B  1:6  105 psi, and its mass density is   1:36  10 5 lb-sec2 /in4 . On the basis
of continuum theory, the quarter wavelength .L=4/ for the fluid within the feedline is:
p
e. Dilational fluid deformation: L=4 D B==.4f  /  207 in.
Clearly, the cross-sectional dimension of the feedline is an order of magnitude less than
L=4. Therefore, a fluid model corresponding to a structural “rod” enclosed within the feedline
structure, yet permitted to slide with respect to the rod, is most appropriate.
Since the contained fluid is constrained to move laterally with the feedline structure, the
“lateral” mass per unit length, A  :0213 lb-sec2 /in2 , is the sum of structural and fluid com-
ponents. Therefore, employing the relevant technical (Euler–Bernoulli) beam theory, the lateral
L=4 estimate for the feedline is
p
f. Feedline bending deformation: L=4 D .=2/
.EI=A/1=4 = 4f   80:8 in.
This result indicates that the ratio of quarter wavelength to feedline diameter is roughly 5
at f  D 50 Hz, suggesting that a feedline structural model accounting for cross-sectional shear
deformation (Timoshenko beam theory) may be most appropriate. Fortunately, most modern
FEM codes employ beam elements that are based on Timoshenko beam theory.

2.2.12 MODAL DENSITY AND THE EFFECTIVENESS OF FINITE

ELEMENT MODELS
Finite element modeling is an effective approach for study of structural and mechanical system
dynamics as long as individual vibration modes have “sufficient frequency spacing” or “low modal
density.” Modal density is typically described as the number of modes within a 1/3 octave fre-
quency band (f0 < f < 1:26f0 ). When modal density of a structural component or structural
assembly is greater than 10 modes per 1/3 octave band, details of individual vibration modes are
not of significance and statistical vibration response characteristics are of primary importance.
In such a situation, the Statistical Energy Analysis (SEA) method [18] applies.
Table 2.3 [15] gives formulae for modal density (as a mathematical derivative, d n=d!
(n D number of modes, ! D frequency in radians/sec), for typical structural components.
The above modal density relationships relate to structural components, rather than struc-
tural assemblies that are composed of a variety of components. More reliable estimation of a
particular system’s modal density is based on the assembled system’s modes.

2.2.13 ILLUSTRATIVE EXAMPLE: FLUID-FILLED CIRCULAR

CYLINDRICAL SHELL
The following example system, consisting of a fluid-filled, thin-walled circular cylindrical shell
(see Figure 2.6), was extensively studied by Abramson [19] during the mid-1960s. It contin-
ues to offer excellent, comprehensive guidance for modeling of launch vehicle propellant tank
24 2. DEFINITION OF TEST ARTICLE FINITE ELEMENT MODELS
Table 2.3: Modal densities of typical structural components

Component Motion Modal Density, dn/dω Additional Data

T = tension, A = area
String Lateral L/(π T/ρA) ρ = mass density
L = length
Rod Axial L/(π E/ρ) E = elastic modulus
Rod Torsion L/(π G/ρ) G = shear modulus
Beam Bending L/(2π)(ω EI/ρA)-1/2 EI = flexural stiffness
N = stress resultant
Membrane Lateral Asω/(2π)(N/ρh)
As = surface area
D = plate flexural stiffness
Plate Bending As/(4π) D/ρh)
h = plate thickness
B = bulk modulus
Acoustic Dilational Voω2/(2π 2)( B/ρ)3
Vo = enclosed volume

structures. The geometry, properties, and boundary conditions for the present example are as
follows.
• Geometry: diameter (D D 300 in), length (L D 600 in), wall thickness (h D 1 in).
• Material: aluminum (E D 107 psi), Poisson’s ratio ( D 0:3), density (s g D 0:1 lb/in3 ).
• Fluid: water (f g D 62:4 lb/ft3 D :036 lb/in3 ).
• Fluid B.Cs: fluid free surface at Z D 600 in, blocked tank bottom at Z D 0 in.
• Structure B.Cs: bottom (Z ) axial restraint, and free to radially expand without bending
restraint.
• Ullage Pressure: P0 D 0 psi, or 30 psi.
The structural shell mode shapes associated with empty and fluid-filled shell are of the
form:
ˆ.z;  I m; n/ D cos.mz=2L/
cos.n / or cos.mz=2L/
sin.n/;
m D 1; 3; 5; : : :; n D 0; 1; 2; 3; : : :: (2.51)
Closed-form expressions for empty structure natural frequencies (without and with ullage
pressure) are provided below. They are associated with bulge (n D 0), lateral (n 1) and shell
breathing (n > 1) for the empty shell are
   
1 k1 C k2 C k3 1=2
fmn D
; (2.52)
2 s h
 2.2. PART 2: GUIDELINES FOR SYSTEMATIC MODEL DEVELOPMENT 25
Z
Fluid-Free Surface

L
Y
The tank bottom is
axially fixed and free
to radially expand
X

Figure 2.6: Fluid-filled circular cylindrical shell.

where
 m 4
 
Eh L
k1 D
  (membrane stiffness parameter) (2.53)
R2 m 2  n 2 2
C
L R
   
Eh 3
m 2  n 2 2
k2 D  C (flexural stiffness parameter) (2.54)
12 .1  2 / L R
 
m 2  n 2
k3 D .Po R/
C (ullage “differential” stiffness parameter): (2.55)
L R
In addition, the empty structure’s axial and torsion modal frequencies are (noting that
G D E=.1 C 2//,

     
1 m  p 1 m  p
fZ;m D

E=s ; f;m D

G=s ; (2.56)
2 L 2 L
respectively.
Closed-form expressions for the fluid-filled structure natural frequencies (without and
with ullage pressure) are presented by Abramson in [19], for the “curious” reader. However, the
intent of this illustrative example is to provide insight into parameters affecting the system’s
modal frequencies.
26 2. DEFINITION OF TEST ARTICLE FINITE ELEMENT MODELS
The first informative result of closed-form modal analysis, provided in Figure 2.7, indicates
the roles of membrane, flexural, and ullage pressure strain energies on empty shell natural (m D
1) frequencies.

Sensitivity of Empty Shell “m = 1” Frequencies to Membrane, Flexure,

and Ullage Pressure Strain Energies
102

Frequency (Hz)

101 Membrane
Membrane+Flexure
Membrane_Flexure+Ullage

100
0 2 4 6 8 10 12 14 16
Circumferential Harmonic (n)

Figure 2.7: Sensitivity of empty shell natural frequencies to strain energy contributions.

The second informative result of closed-form modal analysis, provided in Figure 2.8, in-
dicates the roles of membrane, flexural, and ullage pressure strain energies on fluid-filled shell
natural .m D 1/ frequencies.
The profound effect of fluid mass on shell natural frequencies is summarized in Figure 2.9
for the “no-ullage pressure,” m D 1 modes.
The high modal density and ullage pressure sensitivity of empty shell natural frequencies
are illustrated in Figure 2.10.
Table 2.4 details the numerical natural frequencies corresponding to Figure 2.13. Note
that (1) modal frequencies exceeding 85 Hz are shaded in gray and (2) axial and torsion natural
frequencies are indicated.
The corresponding high modal density and ullage pressure sensitivity of fluid-filled shell
natural frequencies are illustrated in Figure 2.11.
Table 2.5 details the numerical natural frequencies corresponding to Figure 2.11. Note
that (1) modal frequencies exceeding 85 Hz are shaded in gray and (2) axial and torsion natural
frequencies are indicated.
The following insights are gained from the previous results:
 2.2. PART 2: GUIDELINES FOR SYSTEMATIC MODEL DEVELOPMENT 27

Sensitivity of Fluid-Filled Shell “m = 1” Frequencies to Membrane, Flexure,

and Ullage Pressure Strain Energies
102

101
Frequency (Hz)

Membrane
Membrane+Flexure
Membrane_Flexure+Ullage

100

10-1
0 2 4 6 8 10 12 14 16
Circumferential Harmonic (n)

Figure 2.8: Sensitivity of fluid-filled shell natural frequencies to strain energy contributions.

Sensitivity of Shell “m = 1” Frequencies to Fluid-Mass (Ullage Pressure = 0 psi)

102
Frequency (Hz)

101
Empty
Fluid-Filled

100
0 2 4 6 8 10 12 14 16
Circumferential Harmonic (n)

Figure 2.9: Sensitivity of natural frequencies to fluid mass loading.

28 2. DEFINITION OF TEST ARTICLE FINITE ELEMENT MODELS

Table 2.4: Modal density and sensitivity of empty shell natural frequencies to ullage pressure
Circumferential Harmonic (n), P0 = 0 psi
m

Circumferential Harmonic (n), P0 = 30 psi

m

Table 2.5: Modal density and sensitivity of fluid-filled shell natural frequencies to ullage pressure
Circumferential Harmonic (n), P0 = 0 psi
m

Circumferential Harmonic (n), P0 = 30 psi

m
 2.2. PART 2: GUIDELINES FOR SYSTEMATIC MODEL DEVELOPMENT 29
Sensitivity of Empty Shell Frequencies Ullage Pressure
102

Frequency (Hz)
P0= 0 psi
P0= 30 psi
101

100
0 2 4 6 8 10 12 14 16
Circumferential Harmonic (n)

Figure 2.10: Modal density and sensitivity of empty shell natural frequencies to ullage pressure.

a. Bulge (n D 0) bulge and lateral (n D 1) natural frequencies are associated with membrane
strain energy.
b. Bulge (n D 0) bulge and lateral (n D 1) natural frequencies are insensitive to ullage pressure.
c. Shell breathing (n > 1) natural frequencies are associated with membrane and flexural strain
energies.
d. Shell breathing (n > 1) natural frequencies are sensitive to ullage pressure.
e. Shell natural frequencies are profoundly sensitive to added fluid mass.
f. Axial and torsion natural frequencies are unaffected by fluid mass and ullage pressure.

2.2.14 FREE SURFACE SLOSH MODES

A discussion of the modal behavior of a fluid-filled structure is incomplete without inclusion of
free surface slosh modes. The pendulum mechanics of slosh are thoroughly discussed by Abram-
son [19], who described theoretical foundations, experimental results, and a variety of simplified,
lumped parameter dynamic models. Fluid-structure interaction FEMs that include free surface
effects, developed by Zienkiewicz [8], have been implemented in various finite element software
packages. Of primary importance to Abramson’s fluid-filled circular cylindrical shell is the fact
that the fundamental free surface slosh modes occur at frequencies (0.35 Hz) sufficiently be-
low the lowest flexible structure-fluid mode (1.20 Hz) to be modeled separately as interactions
between the sloshing fluid and “rigid” container structure. As a result of situations of this type,
30 2. DEFINITION OF TEST ARTICLE FINITE ELEMENT MODELS
Sensitivity of Fluid-Filled Shell Frequencies to Ullage Pressure
102

Frequency (Hz)
101
P0= 0 psi
P0= 30 psi

100
0 2 4 6 8 10 12 14 16
Circumferential Harmonic (n)

Figure 2.11: Modal density and sensitivity of fluid-filled shell natural frequencies to ullage pres-
sure.

a trivial (zero pressure) boundary condition is often imposed on fluid-filled FEMs. Moreover,
since it is quite difficult to instrument the fluid free surface, experimental modal tests most of-
ten do not engage in measurement of free surface slosh modes. Inclusion of free surface slosh
in predictive models is often accomplished by employing simplified, lumped parameter dynamic
models.

2.2.15 INTERFACE FLEXIBILITY IN STRUCTURAL ASSEMBLIES—A

RETROSPECTIVE
Mathematical solutions for structural components, subjected to a wide variety of boundary con-
ditions are a staple in the historical development of structural mechanics theory [5]. The math-
ematical solutions are complemented by a wealth of empirical data indicating variability of joint
stiffness as well as damping (especially when joints have slip-friction behavior). Significant de-
viations from assumed ideal joint behavior are also present in structures, which are composed of
components that are welded to one another.
A quite revealing illustration of non-ideal boundary conditions is noted in results of a
series of modal tests conducted on thin cylindrical shells (see Figure 2.12) at NASA/LARC in
the mid-1970s [20].
Initial COSMIC NASTRAN mathematical models of the test article were defined with
fixed end boundary conditions for all test conditions: (1) empty, unpressurized; (2) empty pres-
surized; and (3) half-filled with water unpressurized and pressurized. Natural frequencies of
 2.2. PART 2: GUIDELINES FOR SYSTEMATIC MODEL DEVELOPMENT 31
0.0813 CM Lip (Top & Bottom)

R = 25.4 CM 0.635 CM Resistance Weld

1.27 CM 1.905 CM

1.587 CM

P0 - Internal
Pressure
t = 0.0813 CM t = 50.8 CM

50.8 CM

Shell breathing modes mapped for: Bottom Plate Thickness Varies

From 1.905 CM at Outer Edge
1. Empty, un-pressurized to 1.27 CM at C/L
Resistance Weld
2. Empty pressurized 0.635 CM

3. Half-filled with water, un-pressurized 1.905 CM

1.905 CM

50.8 CM Diameter

Figure 2.12: NASA Langley Research Center cylindrical shell test article.

shell breathing modes for the initial models were significantly higher than all corresponding
test data. After changing the model end boundary conditions to pinned (shear diaphragm),
which was not intuitively obvious to the young engineer (this book’s author), all of the analytical
natural frequencies closely followed modal test data, as illustrated in Figure 2.13.
This lesson, experienced by many young engineers, is a clear example of non-ideal bound-
ary conditions that exist in real structures. It is most unfortunate that this point is so easily missed
by many practicing engineers due to the high degree of automation in day-to-day utilization of
today’s computer-aided engineering (CAE) tools.

2.2.16 DAMPING IN STRUCTURAL ASSEMBLIES

Most engineering organizations employ empirically based values for modal damping (n ), and
the modal approach for description of “viscous” damping, when practical, which circumvents
difficulties associated with the lack of a theoretical damping matrix.
A variety of artificially constructed mathematical forms for the damping matrix have been
defined over the past century. One form that has managed to find its way into most finite element
32 2. DEFINITION OF TEST ARTICLE FINITE ELEMENT MODELS
R = 25.4 CM
1.27 CM

500
Empty

P0 - Internal
P0 = 0

Pressure
400 t = 50.8 CM
t = 0.0813 CM
f|HZ|
Z
300

200
Mode Shape Functions: Legend
4 6 8 10 12
n wIZ 01 = wIZ cos nP △ Test
□ Anal. (w = |CL-CL|
2w
700 Anal. (w = —3Z
= 0)
Empty
300 NZ = 0 z=0l
P0 = 5.516 × 104 N/M2
600 ½ Full Water
P0 = 0
f|HZ| 200

500 f|HZ|

100

400

0
4 6 8 10 12 4 6 8 10 12
n n

Figure 2.13: Comparison of predicted and measured shell breathing mode frequencies.

codes, namely proportional damping, which is attributed to Rayleigh [21], is

ŒB D ˛ ŒM  C ˇ ŒK : (2.57)

Application of the modal transformation on this matrix form results in the following
distribution of modal damping, which does not resemble typical empirical data records:
˛ ˇ!n
n D C : (2.58)
2!n 2
While proportional damping constructs a mathematically elegant approach for computa-
tional structural dynamics, it is never consistent with experimental data, and merits abandon-
ment.
During the late 1920s, Kimball and Lovell [22] and Becker and Foppl [23], independently
determined by experiment, that damping in typical structures is simultaneously proportional to
displacement (strain) and in phase with velocity, Shortly thereafter, Kussner [24] and Kass-
ner [25] introduced the concept of complex structural damping, which appropriately describes
 2.2. PART 2: GUIDELINES FOR SYSTEMATIC MODEL DEVELOPMENT 33
the observations of Kimball, Lovell, Becker, and Foppl. In short, the mathematical description
of damping in typical structures shifted from a theoretical (viscous) formulation

ŒM  fug
R C ŒB fug
P C ŒK fug D fFe .t/g (2.59)

to a hysteretic formulation (known today as structural damping),

ŒM  fug
R C .1 C i/ ŒK fug D fFe .t/g : (2.60)

Note that structural damping and viscous modal damping coefficients are related to one
another as
 D 2n : (2.61)
The contributions of Kimball, Lovell, Becker, and Foppl represent a great contrast from
the situation which resulted from Rayleigh’s citation about proportional damping, which quoted
directly from Theory of Sound (Chapter V, Paragraph 97) states: “The first case occurs frequently,
in books at any rate, when motion of each part of the system is resisted by a retarding force,
proportional both to the mass and velocity of the part, the same exceptional reduction is possible
when F (the dissipation force) is a linear function of T (kinetic energy) and V (strain energy).”
The thorough treatment of structural damping, found in the text by Cremer, Heckl, and
Ungar [26], provides a wealth of empirical data along with a technical viewpoint that comple-
ments the prevailing, automated finite element mindset. Three crucial features inherent in many
structural systems are clearly noted in that text, namely:
1. “Solid” structures generally exhibit damping forces that are independent of frequency (dis-
placement dependent structural damping) for a wide range of building materials.
2. Structural damping is often extremely low (  104 ) for individual, unattached structural
members, such as bars, beams, plates and shells. This is typical for steel, aluminum and
other “hard” metals; damping may be two orders of magnitude greater for lead, concrete,
and brick.
3. Structural damping in assemblies is often on the order of   0:01, which is attributed to
losses in (welded, bolted, riveted, and bearing type) joints.

2.2.17 CLOSURE
Practical guidelines for pre-test structural dynamic modeling, presented in this chapter for the
integrated test analysis process, are as follows.
1. Determine the relevant frequency band (0 < f  f  ) by shock and/or response spectrum
analysis of anticipated operational environments. In the absence of such data, U.S. gov-
ernment standards (e.g., NASA STD-5002 and SMC-S-004) suggest accepted frequency
bands for typical launch vehicle and spacecraft structures.
34 2. DEFINITION OF TEST ARTICLE FINITE ELEMENT MODELS
2. Follow frequency-wavelength guidelines for definition of an appropriate structural dy-
namic FEM (most modern CAE/FEM software resources by default will inherently sat-
isfy this guideline). However, in many cases engineers take risky “shortcuts” in model
definition.
3. The role played by interface flexibility, as noted in the 1970s NASA/LARC shell modal
test project, points to the importance of detailed joint modeling as an essential part of FEM
modeling. Specifically, models should accurately follow engineering drawings (facilitated
by modern CAE/FEM software resources, but often not properly employed) in the vicinity
of structural joints.
4. Typical aerospace shell-type structures have relatively high modal density. In particular,
overall body bending (n D 1), torsional, and shell breathing (n > 1) modes occupy the
same frequency band. This phenomenon, which occurs for both empty and fluid-filled
structures, is descriptive of the “many modes” problem. The “many modes” problem, when it
is present, significantly affects all steps in the integrated test analysis process.
5. The profound effect of added fluid mass on the fundamental (m D 1, n D 0) bulge mode
of a propellant tank ( 208 Hz empty vs.  8 Hz fluid filled for the Abramson example)
points to a serious potential deficiency associated with modal testing that is limited empty
propellant tank structures. Specifically, a primary structural mode of the flight vehicle is
not subjected to verification and validation offered by the integrated test analysis process
(typically limited to the 0–50 Hz frequency band).
6. The dominant role of “body” modes in estimation of primary structure flight loads, and
(control system and Pogo) stability margins, suggests the potential cost-schedule benefits
associated with an integrated test analysis process that focuses primarily on “body” modes
(n D 0 and 1 for the Abramson example). This consequence offers an opportunity to po-
tentially streamline the modal test plan (Chapter 3).
7. “Body” modes appear to be primarily sensitive to membrane stiffness parameters, and in-
sensitive to shell flexural stiffness parameters and ullage pressure stiffness. This attribute
may greatly streamline the test-analysis correlation (Chapter 6) and test-analysis reconcil-
iation (Chapter 7) steps in the integrated test-analysis process.
8. Shell “breathing” modes (n > 1 for the Abramson example) are sensitive to membrane,
shell flexural, and ullage pressure stiffness parameters. Moreover, these modes are sensitive
to local details and manufacturing imperfections. This suggests a significant challenge to
the entire integrated test analysis process when “body” and shell breathing modes cannot
be separated into distinct classes.
9. In light of widespread misunderstandings related to damping in structures, the assumption
of proportional damping should never be employed in structural dynamic modeling. Modal
 2.3. REFERENCES 35
damping for predicted “real” modes appears to be a safer practice. However, the realities
associated with actual damping mechanisms (e.g., concentration at joints and nonlinearity)
lead to “unsettling” issues that ultimately point to measured “complex” modal behavior,
which cannot be effectively dismissed.

2.3 REFERENCES
[1] I. Newton, Naturalis Principia Mathematica, London, 1689. 7
[2] J. le Rond d’Alembert, Trait e de Dynamique, Paris, 1743. 7
[3] W. R. Hamilton, On a general method in dynamics, Philosophical Transactions of the Royal
Society, 1835. 7
[4] J. L. Lagrange, Mechanique Analitique, Paris, 1788. 7
[5] S. P. Timoshenko, History of Strength of Materials, McGraw-Hill, 1953. 7, 30
[6] A. Piersol and T. Paez, Eds., Harris’ Shock and Vibration Handbook, 6th ed., McGraw-
Hill, 2010. 7
[7] W. Ritz, Über eine neue methode zur Lösung gewisser variationsprobleme der math-
ematischen physik, Journal Für die Reine und Angewandte Mathematik, 1909. DOI:
10.1515/crll.1909.135.1. 7, 9
[8] O. C. Zienkiewicz, R. L. Taylor, and J. Z. Zhu, The Finite Element Method, its Basis and
Fundamentals, 6th ed., Elsevier, 2005. 7, 10, 29
[9] L. A. Pipes, Matrix Methods for Engineering, Prentice Hall, 1961. 7, 10
[10] B. G. Galerkin, …some questions of elastic equilibrium of rods and plates, Vestnik In-
zhenerov i Tekhnikov, 19, 1915. 7, 10
[11] A. H. Nayfeh and D. T. Mook, Nonlinear Oscillations, John Wiley & Sons, 1979. DOI:
10.1002/9783527617586. 7, 10
[12] E. Trefftz, Ein gegenstuck zum ritzschen verfahren, Proc. of the 2nd International Congress
of Applied Mechanics, 1926. 7, 10
[13] C. A. Brebbia and J. Dominguez, Boundary Elements an Introductory Course, 2nd ed.,
WIT Press, 1992. DOI: 10.1115/1.2897280. 7
[14] W. P. Rodden, Theoretical and Computational Aeroelasticity, Crest Publishing, 2011. 8
[15] A. G. Piersol and T. L. Paez, Harris’ Shock and Vibration Handbook, 6th ed., McGraw-
Hill, 2010. 14, 17, 18, 21, 23
36 2. DEFINITION OF TEST ARTICLE FINITE ELEMENT MODELS
[16] S. P. Timoshenko, History of Strength of Materials, McGraw-Hill, 1953. 21
[17] Y. C. Fung, Foundations of Solid Mechanics, Prentice Hall, 1965. 21
[18] R. H. Lyon and R. G. DeJong, Theory and Application of Statistical Energy Analysis, 2nd
ed., Butterworth–Heinemann, 1995. 23
[19] H. Norman Abramson, The Dynamic Behavior of Liquids in Moving Containers, NASA
SP-106, 1966. 23, 25, 29
[20] R. N. Coppolino, A Numerically Efficient Finite Element Hydroelastic Analysis, vol. 1 The-
ory and Results, NASA CR-2662, 1976. DOI: 10.2514/6.1976-1533. 30
[21] L. Rayleigh, The Theory of Sound, vol. 1 and vol. 2, Cambridge University Press, 2011.
DOI: 10.1063/1.3060230. 32
[22] A. Kimball and D. Lovell, Internal friction in solids, Physical Review, 30, 1927. DOI:
10.1103/physrev.30.948. 32
[23] E. Becker and O. Foppl, Dauerversuche zur bestimmung der festigkeitseigenschaften,
Beziehungen Zwischen Baustoffdampfung und Verformungeschwindigkeit, Forschungsh. Ver.
Deutsch. Ing., no. 304, 1928. 32
[24] H. Kussner, Augenblicklicher entwicklungsstand der frage des flugelflatterns, Luftfahrt-
forsch, 12(6):193–209, 1935. 32
[25] R. Kassner, Die berucksichtigung der inneren dampfung beim ebenen problem der
flugelschwingung, Luftfahrtforsch., 13(11):388–393, 1936. 32
[26] L. Cremer, M. Heckl, and E. Ungar, Structure Borne Sound, Springer-Verlag, 1973. DOI:
10.1007/978-3-662-10121-6. 33
 37

CHAPTER 3

Systematic Modal Test

Planning
3.1 PART 1: UNDERSTANDING MODAL DYNAMIC
CHARACTERISTICS
3.1.1 INTRODUCTION
Comprehensive understanding of a structure’s modal characteristics is realized by quantitative
analysis of the kinetic and strain energy distributions associated with its normal modes. Analysis
of energy distributions offers in-depth insights that are not necessarily realized by visualization
of frozen and animated displays of geometric modal deformations.
The modes of an undamped structural dynamic system are solutions of the real, symmetric
eigenvalue problem,


ŒK fˆn g ŒM  fˆn g n D f0g ; n D !n2 ; (3.1)

where fˆn g are distinct, orthogonal individual eigenvectors (or mode shapes), and n D !n2 , are
the corresponding eigenvalues. The collection of all (or a truncated set of ) eigenvectors defines
the system’s modal matrix, Œˆ, which (when normalized to unit modal mass) has the following
properties:
ŒˆT ŒM  Œˆ D ŒI  ; ŒˆT ŒK Œˆ D Œ : (3.2)
An interesting aspect of modal orthogonality for an undamped structural dynamic system
lies in the fact that the orthogonality relationships can be “unpacked” to describe modal kinetic
and strain energy distributions. Specifically, the respective kinetic and strain distributions for
each mode are the term-by term products:
DOF
X
fKEgn D fŒM  fˆgn g ˝ fˆgn ; KETOT;n D KEin D 1
i D1
(3.3)
DOF
X
fSEgn D fŒK fˆgn g ˝ fˆgn =n ; SETOT;n D SEin D 1:
i D1

The individual terms in each of these “energy” vectors are directly associated with the
dynamic system degrees of freedom. As such, they provide appropriately weighted metrics for
38 3. SYSTEMATIC MODAL TEST PLANNING
kinetic and strain energy distributions that are not indicated by the geometric modes shape
(e.g., “heavier” degrees of freedom have greater kinetic energy than “lighter” degrees of freedom
with equivalent modal displacements). A direct consequence of the algebraic eigenvalue prob-
lem, Equation (3.1), is that the modal kinetic and strain energy distributions for the assembled
system are identical. However, in a later discussion, segmentation of the system’s mass and stiff-
ness distributions into component partitions will be demonstrated as a means of discriminating
distributions modal kinetic and strain energies. That being said, modal kinetic energy grouped
sums, partitioned by global direction yield the direction of overall modal activity for a particular
mode, depicted as follows:
X X X
fKEgn ) fKETX gn ; fKETY gn ; fKETZ gn ;
X X X
fKERX gn ; fKERY gn ; fKERZ gn
X X X (3.4)
fKETX gn C fKETY gn C fKETZ gn
X X X
C fKERX gn C fKERY gn C fKERZ gn D 1:

In addition, modal kinetic energy grouped sums, partitioned by subsystem component,

provide additional valuable insights, i.e.,
X X X
fKEgn ) fKE1 gn ; fKE2 gn ; fKE3 gn ; : : :
X X X (3.5)
fKE1 gn C fKE2 gn C fKE3 gn ; : : : D 1:

3.1.2 ILLUSTRATIVE EXAMPLE: ISS P5 SHORT SPACER

The ISS (International Space Station) P5 Short Spacer was the subject an end-to-end exercise of
the Integrated Test Analysis Process in 2001. Modal testing was conducted by a team composed
of Boeing/Rocketdyne (the manufacturer), NASA/MSFC (the laboratory), and Measurement
Analysis Corporation (the consultant). The P5 test article and (space shuttle payload bay geo-
metric) laboratory test fixture are shown in Figure 3.1.
The 21,666 DOF (3611 grid points) FEM of the ISS P5 (weighing 3605 lb) and test
fixture is illustrated in Figure 3.2.
Descriptive geometric plots of the first two FEM modes, illustrated in Figure 3.3, reveal
the character and content of modal behavior in a superficial manner.
A summary of modal kinetic energies associated with the first 36 FEM modes of the P5
test article (note that “rotational DOF distributions are not indicated as they were negligible) is
provided in Table 3.1.
Note that the first 13 FEM modes have less than 1% fixture modal kinetic energy, and non-
negligible fixture modal kinetic energy occurs in most of the higher FEM modes. In addition,
modes with dominant fixture modal kinetic energy occur at 50 Hz and higher frequencies.
 3.1. PART 1: UNDERSTANDING MODAL DYNAMIC CHARACTERISTICS 39

Figure 3.1: ISS P5 short spacer.

Complete Test Article P5 Truss (red), P5 Grapple (blue)

Figure 3.2: ISS P5 and test fixture finite element model.

40 3. SYSTEMATIC MODAL TEST PLANNING

Z Z
X
Y
Y X
Mode 1: 18.67 Hz (KEX = 94%) Mode 2: 19.31 Hz (KEY = 89%)

Figure 3.3: Mode 1: 18.67 Hz (KEx D 94%), Mode 2: 19.31 Hz (KEy D 89%).

Table 3.1: P5 test article FEM modal kinetic energy distributions

Directional KE (%) Component KE (%) Directional KE (%) Component KE (%)
Mode Freq (Hz) Mode Freq (Hz)
X Y Z Truss Grapple Fixture X Y Z Truss Grapple Fixture

3.1.3 ILLUSTRATIVE EXAMPLE: AXISYMMETRIC SHELL FINITE

ELEMENT MODEL
The shell structure, shown in Figure 3.4, consists of five substructures, namely: (1) a lower cylin-
drical skirt (fully fixed at its base), (2) a lower hemispherical bulkhead, (3) lower cylindrical
section, (4) upper cylindrical section, and (5) upper hemispherical bulkhead.
In addition, allocation of unassembled component mass and stiffness matrices within the
5616 DOF model are illustrated in Figure 3.5.
The overall dimensions of the aluminum structure are length, L D 100 in, radius, R D 20
in, and wall thickness, h D 0:4 in. It should be noted that this illustrative example structure
does not represent a realistic design. The rather high thickness-to-radius ratio, h=R D 1=50, was
 3.1. PART 1: UNDERSTANDING MODAL DYNAMIC CHARACTERISTICS 41

Z
• System Point
Y X •• Substructure Boundary Point
Substructure Point

Figure 3.4: Illustrative example shell structure.

Skirt Lower Dome Lower Cylinder Upper Cylinder Upper Dome

0 0 0 0 0
1000 1000 1000 1000 1000
2000 2000 2000 2000 2000
3000 3000 3000 3000 3000
4000 4000 4000 4000 4000
5000 5000 5000 5000 5000
0 2000 4000 0 2000 4000 0 2000 4000 0 2000 4000 0 2000 4000

Figure 3.5: Allocation of shell structure component matrices.

selected to produce less shell breathing modes in the base frequency band (f < 2000 Hz) than
typical aerospace systems, while including modes of sufficient complexity to illustrate key aspects
of quantitative normal mode metrics. The subject structure, fully constrained at the bottom of
the lower cylindrical skirt, has 150 modes in the 0–1453 Hz frequency band. If the example
structure is viewed as a 1/20th scale model, the full-scale set of 150 modal frequencies becomes
0–73 Hz, which is representative of the range of interest for some spacecraft applications.
Modal kinetic and strain energy grouped sums, partitioned by subsystem component, yield
the distribution of component activity for a particular mode. It should be noted that the com-
ponent kinetic and strain energies are not necessarily distributed in the same manner as one
another when they are computed on the basis of separate component mass and stiffness matri-
42 3. SYSTEMATIC MODAL TEST PLANNING
ces in accordance with
X
fKEm gn D fŒMm  fˆgn g ˝ fˆgn ; KEmn D fKEm gn  1
DOF

(modal KE for component m, mode n)

(3.6)
X
fSEm gn D fŒKm  fˆgn g ˝ fˆgn =n ; SEmn D fSEm gn  1
DOF

(modal SE for component m, mode n):

Modal dynamics of axisymmetric shell structures are composed of “body” modes (i.e., ax-
ial, lateral two orthogonal directions, torsional, and bulge) and “breathing” modes associated
with n > 1 circumferential harmonics as well as axial harmonics). In order to systematically
classify modes of geometrically (but not necessarily physically) axisymmetric shell structures,
consider the seven “body” geometric patterns for successive shell circumferential stations illus-
trated in Figure 3.6.
The seven patterns are associated with cross-sectional lateral (“TX ” and “TY ”), and axial
(“TZ ”) translations, pitch, yaw, and torsional (“RX ,”“RY ,” and “RZ ”) rotations, and radial bulge
(“TR”) translation. In “circumferential harmonic terms,” the above seven patterns represent n=0
and 1 motions (or load patterns).
By organizing the above-described geometric patterns as a body displacement transfor-
mation matrix, Œ b , the discrete FEM shell displacements, Œˆ, are expressed as
Œˆ D Œˆb  C Œˆr  D Œ b  Œ'b  C Œˆr  ; (3.7)
where Œˆr  represents remaining (residual) displacements that are not represented by the body
patterns.
Employing weighted linear least-squares analysis, as described below, each system normal
mode may be partitioned into “body” and (remainder) “breathing” components, as follows.
Step 1: Weighted least squares
 T     
‰b Mˆ D ‰bT M ‰b Œ'b  C ‰bT Mˆr : (3.8)
Step 2: Orthogonal constraint
 T

b Mˆr D Œ0 : (3.9)
Step 3: Generalized “body” modes
  1  T 
Œ'b  D ‰bT M ‰b ‰b Mˆ : (3.10)
Step 4: “Body” and “breathing” distributions
ŒKEb  D ŒMˆb  ˝ Œˆb  ; ŒKEr  D ŒMˆr  ˝ Œˆr  : (3.11)
 3.1. PART 1: UNDERSTANDING MODAL DYNAMIC CHARACTERISTICS 43

Typical “FX” Load Patch Typical “FY” Load Patch Typical “FZ” Load Patch

Z
Z
0
20 20 20
10 20 10 20 10 20
10 10
0 0 0 10
0 0 0
-10 -10 -10 -10 -10
Y -20 -20 X Y -20 -20 X Y -10 X
-20 -20

Typical “MX” Load Patch Typical “MY” Load Patch Typical “MZ” Load Patch

5 5
Z

Z
0 0
-5 -5 20
20 20 10 20
20 20 0 10
10 10
10 10 0
0 0 -10 -10
0 0
-10 -10 -10 -20 -20
-10
Y -20 -20 X Y -20 -20 X Y X

Typical “FR” Load Patch

20
10 20
0 10
0
-10 -10
-20 -20

Y X

Figure 3.6: Shell axial station “body” geometric deformation patterns.

Step 5: “Body” and “breathing” total kinetic energies formed by column sums of Equa-
tion (3.11):
X
Body kinetic energy for mode “n” D KEb;n D fKEb gn
X (3.12)
Breathing kinetic energy for mode “n” D KEr;n D fKEr gn :

The sum of “body” and “breathing” modal kinetic energies for each mode is always unity
(or 100%). In the case of an ideal axisymmetric structure, the “body” and “breathing” modal
kinetic energy distributions are either 0 or 100%. When imperfections and/or localized features
are present (defining a perturbed shell structure) designations of “body” and “breathing” modal
kinetic energy distributions are less distinct. In such situations, the modal kinetic energy desig-
nation of mode “type” may be defined on the basis of majority components (i.e., a “body” mode
has more than 50% body kinetic energy).
44 3. SYSTEMATIC MODAL TEST PLANNING
A yet more subtle description of “body” and “breathing” modal kinetic energies is realized
by analysis of the “body” contribution in Equation (3.7), specifically since

Œˆb  D Œ‰b  Œ'b  ; (3.13)

      
ˆTb Mˆb D 'bT ‰bT M ‰b Œ'b  D 'bT Œmb  Œ'b  : (3.14)
Therefore, the unpacking of Equation (3.14) results in the “body” mode kinetic energy dis-
tribution expressed in terms of generalized coordinates associated with the seven shape patterns
per axial station, i.e.,
ŒKEb  D Œm'b  ˝ Œ'b  : (3.15)
This result offers the opportunity to describe the “body” modal kinetic energy distributions
in terms of generalized “directions.” Specifically,
a. “body X ” bending kinetic energy:
X X
KEbx;n D fKEb gn C fKEb gn : (3.16)
1;8;::: 5;12;:::

b. “body Y ” bending kinetic energy:

X X
KEby;n D fKEb gn C fKEb gn : (3.17)
2;9;::: 4;11;:::

c. “body Z ” axial kinetic energy:

X
KEbz;n D fKEb gn : (3.18)
3;10;:::

d. “body” torsion kinetic energy:

X
KEbt;n D fKEb gn : (3.19)
6;13;:::

e. “body .n D 0/” bulge kinetic energy:

X
KEbb;n D fKEb gn : (3.20)
7;14;:::

The sixth “breathing” kinetic energy is described by the previous result in Equation (3.12),
specifically
f. “breathing” kinetic energy: X
KEr;n D fKEr gn : (3.21)
1;2;3;:::
 3.1. PART 1: UNDERSTANDING MODAL DYNAMIC CHARACTERISTICS 45
A summary of ideal axisymmetric shell and perturbed shell structure modal frequencies
and kinetic energy metrics (sorted in terms of total “body” and “breathing” kinetic energy dis-
tributions (Equation set (3.13)), for ideal axisymmetric and perturbed shells, is presented in
Figure 3.7.
Ideal Axisymmetric Shell Perturbed Shell

Frequency (Hz)

Frequency (Hz)
1000 1000

500 500
System Mode System Mode
0 Body Mode 0 Body Mode

20 40 60 80 100 120 140 20 40 60 80 100 120 140

Body KE (%)

Body KE (%)
80 80
60 60
40 40
20 20
0 0
20 40 60 80 100 120 140 20 40 60 80 100 120 140
Breathing KE (%)

Breathing KE (%)
80 80
60 60
40 40
20 20
0 0
20 40 60 80 100 120 140 20 40 60 80 100 120 140
Mode Number Mode Number

Figure 3.7: Axisymmetric and perturbed shell structure modal frequencies and kinetic energy
metrics.

The above result is quite significant, as the vast majority of system modes are associated
with “breathing” deformation, which will be shown in the subsequent discussion on Target Mode
Selection to be relatively insignificant contributors to primary structure dynamic loads.
An in-depth summary of the present illustrative example’s modal metrics, for a represen-
tative subset of the first 150 FEM modes for an ideal axisymmetric shell is provided in Table 3.2.
The following observations result from analysis of modal energy characteristics for the
ideal axisymmetric structure.
1. Twenty-four “body” modes, 126 (n > 1) “breathing” modes.
2. Nine “body” bending modes in each of the “X ” and “Y ” directions.
3. Two “body” axial “Z ” modes.
4. Two “body” torsion modes.
5. Two “body” mixed axial-bulge modes.
46 3. SYSTEMATIC MODAL TEST PLANNING
Table 3.2: Selected modes of the example shell structure
Body & Breathing
Kinetic Energies (%) Body Mode Kinetic Energy Types (%) Component Kinetic Energies (%) Component Strain Energies (%)
Mode Freq (Hz) Body Breathing X-Bending Y-Bending Z-Axial Torsion n=0 Bulge Skirt Dome 1 Shell 1 Shell 2 Dome 2 Skirt Dome 1 Shell 1 Shell 2 Dome 2

6. Contrasting kinetic and strain energy distributions in fundamental bending, axial, torsion
modes.

Regarding the sixth observation, the contrasting kinetic and strain energy distributions
(i.e., dominant kinetic energy toward the top, dominant strain energy toward the foundation)
conform to fundamental mechanical expectations.
Graphical illustrations indicating the character of the fundamental (Y ) bending mode and
the lowest frequency shell breathing mode are provided in Figure 3.8. Kinetic (KE) and strain
or potential energy (PE) distributions are indicated in “pie” format.

3.1.4 CLOSURE
Comprehensive understanding of a structure’s modal characteristics employing modal kinetic
and strain energy metrics has been demonstrated with two key illustrative examples, namely:
(a) the International Space Station P5 test article, which has benefitted from an end-to-end
integrated test-analysis project and (b) the axisymmetric shell finite element model, which ad-
dresses aspects of the “many modes” problem.
Unpacking of fundamental orthogonality relationships mathematically yields modal ki-
netic and strain energy distributions. Physically meaningful distributions of modal kinetic and
strain energy distributions occur only if component mass and stiffness matrices are subdivided
into constituent, decoupled partitions; otherwise the assembled system kinetic and strain energy
distributions must be identical (a mathematical consequence of the algebraic eigenvalue prob-
 3.2. PART 2: TARGET MODE SELECTION 47
Model SHELL …Mode 1 122 Hz Model SHELL …Mode 3 176 Hz

Z
Z
Y X
KE X Y PE KE PE
<1% 1% Skirt <1% 8% <1% 1%
1% Skirt 3%
14%
11% CYLDome1 CYLDome1
CYLShell1 CYLShell1 7%
CYLShell2 CYLShell2
CYLDome2 27% 46% CYLDome2
51% 41%
51%

36% 65%
<1% 35%

Mode 1: Y-Bending Mode 3: Cylinder n = 3 Breathing

Figure 3.8: Graphical illustration of the character of two system modes.

lem solution). Classification of body mode properties (i.e., bending, axial, torsion, n D 0 bulge)
is made possible through the utilization of body geometric deformation patterns.
The “many modes” problem is partially managed by classification of a minority of the pop-
ulation as “body” modes. Additional guidance through the “many modes” problem is provided
by the process of target mode selection, which is the topic of the next discussion.

3.2 PART 2: TARGET MODE SELECTION

3.2.1 INTRODUCTION
Guidelines for target mode selection (a subset of “significant” system modes within the fre-
quency band, 0  f  f  ) discussed in Chapter 2, are documented in NASA STD-5002 [1].
While that document appropriately specifies criteria for dynamic systems excited by bound-
ary accelerations, it is not sufficiently clear in its reference to “augmented by modes which are
critical for specific load or deflection definition.” The present discussion offers further clarifica-
tion regarding target mode selection for cases in which the system is excited by (a) boundary
accelerations and/or (b) applied external loads.
In Chapter 2, matrix equations associated with response of a linear structural dynamic
system to dynamic excitations were described. That previous discussion is generalized herein
to include the potential presence of localized nonlinearities, primarily at joints, which occur in
some aerospace systems.
48 3. SYSTEMATIC MODAL TEST PLANNING
3.2.2 STRUCTURE EXCITED BY BOUNDARY ACCELERATIONS AND
APPLIED EXTERNAL LOADS
The form of matrix structural dynamic equations with localized nonlinearities is
˚ ˚
ŒM  UR C ŒB UP C ŒK fU g D Œ€e  fFe g C Œ€N  fFN g : (3.22)

Symmetric system mass, damping, and stiffness matrices, ŒM , ŒB, and ŒK, respectively,
are associated with discrete displacement degrees of freedom, fU g. The external loads, fFe g are
allocated to the discrete displacements in accordance with geometric distributions described by
the (linearly independent) columns of the matrix, Œ€e . Moreover, localized nonlinear internal
loads, fFN g are allocated to the discrete displacements in accordance with geometric distribu-
tions described by the (linearly independent) columns of the matrix, Œ€N .
The vital companion displacements associated with the nonlinear internal loads are
 T
fUN g D €N fU g : (3.23)

Since the most general case nonlinear internal loads may be described as nonlinear func-
tions of localized displacements, velocities, and parameters, the nonlinear loads follow a symbolic
algorithm: ˚

fFN ; pg t Ct D f UN ; UP N ; p t : (3.24)
If the parameters (p ) are updated as a function of time (or motion path), the nonlinearities
are “hysteretic.” Alternatively, if the parameters (p ) are fixed, the nonlinearities are “algebraic.”
By partitioning the displacements into “interior” and “boundary” subsets,
 
Ui
fU g D : (3.25)
Ub

The system equations (assuming for the purposes of this discussion that localized nonlinear
forces are associated with interior DOFs) are expressed as:
        
Mi i Mib UR i Bi i Bi b UP i Ki i Ki b Ui
C C
Mbi Mbb UR b Bbi Bbb UP b Kbi Kbb Ub
     (3.26)
€ie 0i b Fi e €iN
D C fFN g :
0bi Ibb Fb 0bN

Note that the boundary forces, fFb g, are not designated as “exterior loads” or “interface
loads” with another structure, but simply “boundary loads.”
The transformation of “interior” displacements to “relative” displacements with respect to
the “boundary” displacements is defined as:
    0    0 
Ui Ii i Ki i 1 Kib U i Ii i ‰ib U i
D D : (3.27)
Ub 0bi Ibb Ub 0bi Ibb Ub
 3.2. PART 2: TARGET MODE SELECTION 49
Resulting in the transformed system equations,
  0    0    0 
Mi i M 0 ib UR i Bi i B 0 ib UP i Ki i 0i b U i
0 0 R C 0 0 P C 0
M bi M bb Ub B bi B bb Ub 0bi K bb Ub
     (3.28)
€ie 0i b Fie €iN
D T C fFN g :
‰ib €ie Ibb Fb 0bN

Application of the modal transformation (real, undamped modes) of the upper partition
of Equation (3.27),
 0    
U i ˆin 0ib qn
D ; Œˆin T ŒMii  Œˆin  D ŒInn  ;
Ub 0bi Ibb Ub (3.29)
T
 2
Œˆin  ŒKii  Œˆin  D !n ;
results in the “modal” dynamic equations,
        
Inn Pnb qR n 2 n !n 0nb qP n !n2
0nb qn
C C
Pbn M 0 bb UR b 0bn B 0 bb UP b K 0 bb
0bn Ub
 T    T  (3.30)
ˆin €ie 0nb Fie ˆin €iN
D T C fFN g :
‰ib €ie Ibb Fb 0bN
Note, the damping coupling partitions are commonly assumed to have the following prop-
erties:    T 0   0 
ˆTin Bii ˆin D Œ2 n !n  ; ˆin B ib D Œ0nb  ; B bi ˆin D Œ0bn  : (3.31)

3.2.3 MODAL EFFECTIVE MASS AND TARGET MODE SELECTION

The mass coupling partitions, called modal partition factors are defined as,
 
ŒPnb  D ˆTin M 0 ib ; ŒPbn  D ŒPnb T : (3.32)
The modal effective mass associated with one individual mode “n” is defined as,
ŒMeff bb n D ŒPbn  ŒPnb  : (3.33)
The product of the above two modal participation matrices is called the modal effective
mass for mode “n”. The sum of all modal effective masses approaches the boundary mass parti-
tion, i.e., X  
ŒMeff bb n ! M 0 bb : (3.34)
n

A linear structure excited only by boundary acceleration has modal accelerations that are
stimulated by allocations proportional to the modal participation factors. A convenient form that
is commonly employed to evaluate the significance of individual modal responses (i.e., target
mode selection) is normalized modal effective mass, specifically defined:
50 3. SYSTEMATIC MODAL TEST PLANNING
a. total boundary mass is a “row” matrix composed of diagonal terms of ŒM 0 bb  (see Equa-
tion (3.30));

b. modal effective mass (for mode, “n”) composed of diagonal terms of ŒMeff bb n (see Equa-
tion (3.33)); and

c. normalized modal effective mass (for mode, “n”) is formed by the quotient
ŒMeff bb n = ŒM 0 bb .

In the most general case of a structural dynamic model that has a redundant boundary
(typically more than 6 DOF), application of the above scheme may be difficult to interpret.
However, in most applications, the boundary is constrained to behave as a rigid body referenced
to a convenient grid location (6 DOF). In those situations, the total boundary mass conforms
to the system’s rigid body mass referenced at the selected grid location.

3.2.4 ILLUSTRATIVE EXAMPLE: ISS P5 SHORT SPACER

An evaluation of FEM modal content, including modal effective mass, was conducted with
results for the lowest 18 FEM modes summarized in Table 3.3.

Table 3.3: ISS P5 short spacer test article FEM modal kinetic energy and modal effective mass
Directional KE (%) Directional KE (%) Modal Effective Mass (%) [restricted to Truss & Grapple]
Mode Freq (Hz) X Y X Truss Grapple Fixture X Y Z ΘX ΘY ΘZ Comments

Modes with Modal Effective

Mass >5% are Colored Yellow

Modes with Non-negligible

Test Fixture KE
Colored Gray

Relying on modal effective mass, the target mode selection criterion (modal effective mass
> 5%) indicates that only five modes should have been judged “significant.” In the conduct of
the ISS P5 test program, the first 10 modes were selected as target modes on the basis of NASA’s
position that the dynamic environment cutoff for the Space Shuttle was f  D 35 Hz.
 3.2. PART 2: TARGET MODE SELECTION 51
3.2.5 MODAL GAINS AND TARGET MODE SELECTION
By mathematically viewing boundary accelerations, external and nonlinear forces as excitation
sources (on a mode-by-mode basis), Equations (3.30) and (3.31) are recast as follows:
 
fqR n g C Œ2 n !n  fqP n g C !n2 fqn g
˚     (3.35)
D ŒPnb  UR b C ˆTin €i e fFi e g C ˆTin €iN fFiN g ;

 ˚  ˚    
fFb g D M 0 bb UR b C B 0 bb UP b C K 0 bb fUb g C ŒPbn  fqR n g T
‰ib €ie fFie g : (3.36)
The role of modal participation factors, ŒPnb  and ŒPbn , and modal effective mass, asso-
ciated with boundary acceleration excitation was described and demonstrated in the past two
sections of this
 chapter.
 Equation
 (3.35) highlights two additional modal metrics, namely the
modal gains, ˆTin €ie and ˆTin €iN , which indicate the degree of modal excitation of each par-
ticular mode associated with each applied and nonlinear force pattern). The relative magnitudes
of these modal gains therefore may serve as metrics for target mode selection in the presence of
applied external loads and localized nonlinear forces. When many separate applied force pat-
terns and local nonlinearities are present, modal gain metrics may offer an overly complicated
and perplexing target mode selection option.

3.2.6 THE MODE ACCELERATION METHOD AND TARGET MODE

SELECTION
Consider the following recasting of the upper partition of the matrix structural dynamic Equa-
tion (3.28):
˚ ˚ ˚  0 ˚  0 ˚
ŒKi i  U 0 i D ŒMi i  UR 0 i ŒBi i  UP 0 i M i b UR b B i b UP b
 0  (3.37)
K ib fUb g C Œ€ie  fFie g C Œ€iN  fFiN g :
In 1945, Williams [2] introduced the ingenious mode acceleration method, which revo-
lutionized computation of structural dynamic loads process (the benefit of his formulation will
be noted later in the present discussion). Since the contribution of damping forces is often small
compared to inertial forces, Equation (3.37) becomes
˚ ˚  0 ˚
ŒKi i  U 0 i  ŒMi i  UR 0 i M i b UR b
 0  (3.38)
K ib fUb g C Œ€i e  fFi e g C Œ€iN  fFiN g :
The final key step in William’s development of the mode acceleration method involves ˚
substitution of the modal accelerations (see Equation (3.30)) for the physical accelerations, UR i0 ,
resulting in
˚  0 ˚
ŒKii  U 0 i  ŒMii ˆin  fqR n g M ib UR b
 0  (3.39)
K ib fUb g C Œ€ie  fFie g C Œ€iN  fFiN g :
52 3. SYSTEMATIC MODAL TEST PLANNING
The right-hand side contributions are grouped into “modal acceleration (dynamic)” and
“quasi-static” structural forces, which are defined as

fFi gD  ŒMi i ˆin  fqR n g (modal acceleration forces); (3.40)

 0 ˚  
fFi gQS  M ib UR b K 0 ib fUb g
(3.41)
C Œ€ie  fFie g C Œ€iN  fFiN g (quasi-static forces):

The unique contributions associated with William’s mode acceleration method are as fol-
lows.

1. The quasi-static contribution of all higher modes that are not included in the lower fre-
quency subset defined by f  are included without need for their computation!

2. The coefficient terms associated with modal accelerations, fqR n g, are associated with dynamic
characteristics of system response, while the remaining coefficient terms are associated with
quasi-static characteristics of system response.

Further development of William’s mode acceleration method, especially in the aerospace

community, requires expansion of the physical solution, Equation (3.39) in terms of absolute
displacements employing Equation (3.27), resulting in
     
Ui K ii 1 Mii ˆin K i i1 M 0 ib ˚ R
 fqR n g C Ub
Ub 0bn 0bn
      (3.42)
K i i1 K 0 ib Ki i 1 €i e Kii 1 €iN
C fUb g C fFi e g C fFiN g :
Ibb Ibe IbN

The above result implies that the interior stiffness matrix, ŒKii , is non-singular. When the
stiffness matrix is singular, additional operations are required to achieve the above result.
Stress-strain recovery relationships for detailed structural loads are summarized by
 
Ui
f g D ŒK  fU g ; fU g D : (3.43)
Ub

By combining Equations (3.42) and (3.43), the mode acceleration-based dynamic loads
recovery relationship, with load transformation matrices, takes the form
  h i˚
fg D LTM qR fqR n g C LTM UR b UR b
    (3.44)
C LTM Ub fUb g C ŒLTM Fe  fFe g C LTM FN fFN g ;
 3.2. PART 2: TARGET MODE SELECTION 53
where the load transformation matrices are
  h i  
  K ii 1 Mii ˆin K ii 1 M 0 ib
LTM qR D ŒK 
; LTM UR b D ŒK 
;
0bn 0bn
 
  K ii 1 K 0 i b
LTM Ub D ŒK 
; (3.45)
Ibb
   
Kii 1 €ie   Kii 1 €iN
ŒLTM Fe  D ŒK 
; and LTM FN D ŒK 
:
Ibe IbN

It should be noted that the above derived form of William’s mode acceleration method is
one among many other specific forms employed throughout the aerospace community. In partic-
ular, forms including damping effects, and more convenient forms associated with substructure
(component mode synthesis) and fully assembled vehicle systems are prevalent in the aerospace
community. That being said, delineation of accumulated (1) modal acceleration (dynamic) and
(2) quasi-static components is most relevant to the topic of target mode selection, specifically,
the modal acceleration and quasi-static components, respectively, are
 
fgD D LTM qR fqR n g ; (3.46)

h i˚    
f gQS D LTM UR b UR b C LTM Ub fUb g C ŒLTM Fe  fFe g C LTM FN fFN g : (3.47)

Comprehensive mode acceleration based for target mode selection exploits evaluation
of modal and quasi-static contributions to (a) structural forces delineated in Equations (3.40)
and (3.41), and/or (b) a representative set of system dynamic loads delineated in Equations (3.46)
and (3.47) (termed by some organizations “watch list” loads).

3.2.7 ILLUSTRATIVE EXAMPLE: AXISYMMETRIC SHELL FINITE

ELEMENT MODEL
The following demonstration analysis is applied to the axisymmetric shell FEM, interpreting
the results described in Figure 3.7 and Table 3.2, as associated with a 1/20th sub-scale model.
Therefore, the full-scale frequencies of the finite element model are 1/20th of the sub-scale
model.
Consider four applied dynamic load distributions, Œ€ie , applied in the two lateral direc-
tions, the axial direction, and locally normal to the shell wall, respectively (applied boundary
accelerations are not included in this exercise), as depicted in Figure 3.9.
In addition, the time history force, Fe , along with its normalized shock spectrum (Fig-
ure 3.10) is imposed separately on the applied load distributions,
The normalized shock spectrum associated with the applied force time history suggests
a full-scale value for f  , the dynamic response cut-off frequency, on the order of 70–80 Hz.
54 3. SYSTEMATIC MODAL TEST PLANNING

Localized
FEX FEY FEZ
FEY

Z Z Z Z
Y X Y X Y X Y X

Figure 3.9: Applied load distributions.

Applied Force Time History and Normalized SRS

1

0.5
F(t)

-0.5

-1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t (sec)

4
Normalized SRS

0
0 10 20 30 40 50 60 70 80 90 100
Frequency (Hz)

Figure 3.10: Applied force time history and normalized shock spectrum.
 3.2. PART 2: TARGET MODE SELECTION 55
The 150th mode of the shell finite element model is about 73 Hz (corresponding to the 1/20th
sub-scale value of 1452 Hz).
For the present example, the modal response equation (see Equation (3.35)) reduces to,
   
fqR n g C Œ2 n !n  fqP n g C !n2 fqn g D ˆTn €e fFe g : (3.48)
 T 
Therefore the modal gains for each of the four applied load distributions are ˆn €e .
Regarding the first three applied load distributions described in Figure 3.5, the dynamic
loads, f g, of interest are the cumulative “body” loads at a series of 26 axial stations, defined by
the following particular mode acceleration equations:
 
fg D fgD C fgQS D LTM qR fqR n g C ŒLTM Fe  fFe g ;
      (3.49)
LTM qR D ‰bT ŒM  Œˆ ; ŒLTM Fe  D ‰bT Œ€e  :

For the fourth applied load distribution case involving a single concentrated load and
an associated coinciding local response load of interest is described by the load transformation
matrices,  
LTM qR D mDOF ŒˆDOF  ; ŒLTM Fe  D 1; (3.50)
where the “DOF” subscript denotes the local matrix components of interest.
An evaluation of mode-by-mode contributions to dynamic response for all four applied
load distribution cases is summarized in Figure 3.11. (It should be noted that the results are
associated with “unit” values for all modal accelerations.) For each of the four applied load dis-
tributions, there are groupings of three plots; the top plot indicates the distribution of modal
frequencies, the middle plot summarizes relative modal gain norms, and the lower plot indi-
cates an absolute summation “norm” of all dynamic loads relevant for the particular applied load
distribution case.
The following conclusions are drawn from the above results.
1. Inertial “body” load responses, representative of “primary” structure loads, associated with
the first three applied load distributions, are the result of contributions associated with
“body” modes (24 out of a total of 150 modes).
2. Local inertial load response, representative of “secondary” structure loads, associated with
the fourth applied load case is the result of many of the 150 “body” and “breathing” modes.
3. In all cases, modal gains offer an indication of the relative prominence of individual mode
contributions, but the dynamic response load norms for fgD offer a more definitive as-
sessment of the significant modal contributions to a responding loads-modes “watch list.”
4. It appears that predicted system dynamic response “decompositions” are the most relevant
discriminators for target mode selection. Modal gain and modal effective mass may also
serve as less definitive target mode selection indicators.
56 3. SYSTEMATIC MODAL TEST PLANNING
Applied Force-X Drivers & Peak Body Inertial Loads Applied Force-Y Drivers & Peak Body Inertial Loads
100 100

Frequency

Frequency
(Hz)

(Hz)
50 50

00 50 100 150 00 50 100 150

100 100

Modal Gain

Modal Gain
(%)

(%)
50 50

0 0
20 40 60 80 100 120 140 20 40 60 80 100 120 140
100 100
Inertial Load

Inertial Load
(%)

(%)
50 50

0 0
20 40 60 80 100 120 140 20 40 60 80 100 120 140
Mode Mode
Localized Applied Force-Y Drivers &
Applied Force-Z Drivers & Peak Body Inertial Loads
Peak Grid Point 685 Inertial Loads
100 100
Frequency

Frequency
(Hz)

(Hz)
50 50

0 0
0 50 100 150 0 50 100 150
100 100
Modal Gain

Modal Gain
(%)

(%)
50 50

0 0
20 40 60 80 100 120 140 20 40 60 80 100 120 140
100 100
Inertial Load

Inertial Load
(%)

(%)
50 50

0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140

Mode Mode

Figure 3.11: Mode-by-mode contributions to dynamic response.

The relative significance of quasi-static and modal dynamic contributions to system tran-
sient responses, for the first three applied load distribution cases, is summarized in Table 3.4.
In all three distributed applied load distribution cases, the quasi-static contributions to
system response are generally significant. However, the fourth concentrated applied load re-
sponse case, depicted in Figure 3.12, is most interesting in that the local responding load is
almost completely dominated by quasi-static response (in spite of the fact that many modes
contribute to the minor, dynamic component of response.
While the results for the fourth applied load case are anticlimactic, (i.e., (a) dynamic
response is associated with the contributions of many modes and (b) total response is nearly en-
tirely due to quasi-static response), the vital role played by Williams’ mode acceleration method
is most clearly evident. Specifically:

1. system loads due to applied dynamic environments are the sum of “quasi-static” and “modal
dynamic” contributions;
 3.2. PART 2: TARGET MODE SELECTION 57
Table 3.4: Significance of quasi-static and modal dynamic contributions to system response
Load Case 1: Distributed “X” Loading Load Case 2: Distributed “Y” Loading Load Case 3: Distributed “Z” Loading
Geometry |FX| |MY| Geometry |FY| |MX| Geometry |FZ| |FR|
R Z Static Total Static Total R Z Static Total Static Total R Z Static Total Static Total

Normalized Applied Load History

100
Applied Load

50
0
-50
-100
0 0.1 0.2 0.3 0.4 0.5 0.6
t (sec)

Normalized Response Load History

100 Quasi-Static Component
Modal Dynamic Component
50
Response

Total Response

0
-50
-100
0 0.1 0.2 0.3 0.4 0.5 0.6
t (sec)

Figure 3.12: Quasi-static and modal dynamic contributions to system response (load case 4).

2. there are situations in which response loads may be almost entirely the result of “quasi-
static” behavior; hence precise knowledge of modal dynamics is unnecessary; and

3. in the majority of situations for which modal dynamics play a significant role, response
loads (especially primary structural loads) are the product of some system mode contri-
58 3. SYSTEMATIC MODAL TEST PLANNING
butions; in these situations, plans for modal testing are best focused on a selected “target
mode” subset.

3.2.8 CLOSURE
A variety of target mode selection metrics for modal test planning have been explored. When
a subject system is expected to experience dynamic environments imposed as interface dynamic
motions (no other applied load sources), modal effective mass appears to provide an adequate
metric for target mode selection, as employed in the ISS P5 modal test plan and specified in
NASA STD-5002. For the more general case of a subject system expected to experience dis-
tributed applied forces (and in some cases interface dynamic motions), a detailed structural stress
and member response load (also called a loads “watch list”) modal decomposition provides the
most reliable metric for target mode selection. The “watch list” modal decomposition process
relies on Williams’ mode acceleration method and load decomposition matrices that catego-
rize stress and member response load contributions in (a) quasi-static and (b) modal dynamic
categories. As demonstrated with the axisymmetric shell finite element model, the “watch list”
decomposition process offers an effective strategy for dealing with the challenging “many modes”
problem.

3.3 PART 3: RESPONSE DOF SELECTION FOR MAPPING

EXPERIMENTAL MODES
3.3.1 INTRODUCTION
The Residual Kinetic Energy (RKE) method for response “measurement” DOF selection to
map experimental normal modes is described and demonstrated on a simple rod finite element
modal and a pre-test FEM associated with a modal test article. The method utilizes a detailed
pre-test finite element model of the subject structure. An initial response “measurement” DOF
set is first selected for mapping modes in the frequency band of interest. The remaining DOFs
of the structural model are assumed to respond in a quasi-static manner following a Guyan re-
duction transformation. Modes formed from the “measurement” partition of the FEM modal
vectors and expanded using the Guyan reduction transformation comprise the “approximate”
normal mode set. While orthogonality of the complete FEM is mathematically perfect, orthog-
onality associated with the approximate normal mode set is necessarily imperfect. The generally
accepted standard for U.S. government sponsored aerospace systems is specified by NASA and
the USAF Space Command. Residual error vectors are formed based on differences between
complete FEM normal modes in the frequency band of interest and the set of corresponding
“approximate” modes. A RKE matrix directly identifies additional DOFs (or DOF groupings)
which must be instrumented in order to experimentally map the normal modes of interest.
 3.3. PART 3: RESPONSE DOF SELECTION FOR MAPPING EXPERIMENTAL MODES 59
3.3.2 SELECTION OF DOFS FOR MODAL MAPPING (THE RKE
METHOD)
The most commonly used approach for development of a TAM is Guyan eduction [3] in which
the “measurement” DOFs are retained in the “analysis” set. The matrix dynamic equations for
free vibration (with all physical constraints applied) are
  
KAA1 KAO ˆA
ŒK fˆg ŒM  fˆg  D f0g !
KOA KOO ˆO
     (3.51)
MAA1 MAO ˆA 0
D ;
MOA MOO ˆO 0
where the FEM modal matrix is expressed in terms of “analysis” and “omitted” partitions. De-
noting the modal matrix as, Œˆ, perfect orthogonality of (unit mass normalized) modes is ex-
pressed as
ŒOR D ŒˆT ŒM  Œˆ D ŒI  : (3.52)
The Guyan reduction transformation which is used to form an approximate test-analysis
mass (TAM) mass matrix is  
I
ŒT RED  D : (3.53)
KOO1 KOA
If the “analysis” set partition of the modal matrix is assumed to represent “measured”
modal DOFs, then the approximate TAM modal matrix is
   
ˆA I
D ŒˆA  ; (3.54)
‰O KOO1 KOA
where the partition, ‰O , represents the approximate “omitted” DOF partition. Moreover, the
TAM mass matrix, ŒMAA , is defined as
 T   
I MAA1 MAO I
ŒMAA  D : (3.55)
KOO1 KOA MOA MOO KOO1 KOA
Imperfect orthogonality of the modal partition, ŒˆA , corresponding to the “measured”
DOFs with respect to the TAM mass matrix (for unit mass normalized modes, ŒˆA ) is expressed
as,
ŒORA  D ŒˆA T ŒMAA  ŒˆA  ¤ ŒIAA  : (3.56)
The U.S. government test mode orthogonality [1, 4] criterion requires jORA;ij j  10% for
all i ¤ j terms in the TAM orthogonality matrix, ŒORA .
The residual displacement error matrix based on the difference between the exact FEM
and TAM (Equation (3.54)) modal matrices is defined as
     
ˆA ˆA 0
ŒR D D : (3.57)
ˆO ‰O ˆO ‰O
60 3. SYSTEMATIC MODAL TEST PLANNING
Note that the residual error associated with the “measured” or “analysis” DOF partition
is null. The modal kinetic energy distribution for the complete system is

ŒEˆ  D ŒMˆ ˝ Œˆ ; (3.58)

where the column sum for each individual mode is unity (if the modes are normalized to unit
modal mass).
The RKE matrix is defined in a similar manner as

ŒER  D ŒMR ˝ ŒR : (3.59)

Like the residual displacement error, ŒR, the RKE matrix is exactly Œ0 at the rows corre-
sponding to measured DOFs. The expected characteristic that residual energy is pronounced at
“omitted” yet dynamically significant DOFs in any particular mode will be demonstrated with
the uniform, free-free rod described by a 19-grid point FEM, illustrated in Figure 3.13.

1 2 3 4 5 6 7 8 9 10 11
1 12 13 14 15 16 17 18 19

Figure 3.13: Simple rod structure.

The first six (6) axial vibration modes are graphically depicted in Figure 3.14.
DOFs 1, 3, 5, 7, 9, and 19 were first selected as “measured” DOFs. The TAM orthogo-
nality matrix and RKE associated with this selected DOF set are illustrated in Figure 3.15.
The orthogonality matrix and RKE plot indicate that the selected “measured” DOFs are
inadequate for mapping of modes 3–6. Pronounced RKE in DOFs 11–17 suggests addition of
“measured” DOFs in modes that vicinity.
Addition of accelerometers at DOFs 11, 13, 15, and 17 was found to improve both the or-
thogonality matrix (conforming to the NASA-STD-5002 and SMC-S-004 criteria) and RKE,
as indicated in Figure 3.16.
Since its introduction in 1998 [5], the RKE method has been employed by many organi-
zations in the U.S. aerospace industry for modal test planning. A variety of enhancements by a
number of authors, involving the application of iterative and genetic schemes and modifications
to the basic Guyan reduction method, have been introduced over the years. However all of the
enhancements are based on the original RKE formulation.

3.3.3 ILLUSTRATIVE EXAMPLE: ISS P5 SHORT SPACER

The RKE method was employed for determination of accelerometer allocation for the ISS P5
Short Spacer modal test program. An allocation of 86 triaxial accelerometers (258 DOF) was se-
lected, resulting in a TAM mass matrix that produced impressive orthogonality for the majority
of 36 FEM modes, as indicated in Figure 3.17.
 3.3. PART 3: RESPONSE DOF SELECTION FOR MAPPING EXPERIMENTAL MODES 61
Mode 1 Mode 2 Mode 3
0.2 0.3 0.3
0.15
0.2 0.2
0.1
0.05 0.1 0.1

Phi

Phi

Phi
0 0 0
-0.05
-0.1 -0.1
-0.1
-0.15 -0.2 -0.2
-0.2 -0.3 -0.3
5 10 15 5 10 15 5 10 15
DOF DOF DOF

Mode 4 Mode 5 Mode 6

0.3 0.3 0.3

0.2 0.2 0.2

0.1 0.1 0.1

Phi

Phi

Phi
0 0 0

-0.1 -0.1 -0.1

-0.2 -0.2 -0.2

-0.3 -0.3 -0.3

5 10 15 5 10 15 5 10 15
DOF DOF DOF

Figure 3.14: Simple bar structure modes.

Cumulative Sum of RKE

CumSum (RKE) (%)

Figure 3.15: TAM orthogonality and RKE for the 6 DOF accelerometer set.
Cumulative Sum of RKE
CumSum (RKE) (%)

Figure 3.16: TAM orthogonality and RKE for the 10 DOF accelerometer set.
62 3. SYSTEMATIC MODAL TEST PLANNING

100

80

Orthogonality
60

40

20

0
30
20 30
20
10 10
0 0
Mode Mode

Figure 3.17: ISS P5 pre-test modal orthogonality prediction.

Table 3.5: Summary of pre-test modal characteristics and TAM mass matrix modal orthogo-
nality
Kinetic Energy (%)
Mode Freq (Hz) |Orthogonality| (%)
Truss Grapple Fixture

A more detailed summary justifying the RKE selected TAM mass matrix and accelerom-
eter allocation plan (for predicted modes 1–17) is provided in Table 3.5.
The above summary suggests that the RKE-derived modal test plan includes sufficient
accelerometer channels to map modes (associated with the P5 test article and test fixture) in
the 0–50 Hz frequency band, while satisfying NASA-STD-5002 TAM orthogonality criteria.
Note modes 14–17 include non-negligible modal kinetic energy content in the test fixture.
 3.3. PART 3: RESPONSE DOF SELECTION FOR MAPPING EXPERIMENTAL MODES 63
3.3.4 ALLOCATION OF MODAL EXCITATION RESOURCES
Planning of excitation resources for a modal test involves placement and orientation of shakers
based on
1. accessibility and structural integrity of candidate test article locations; and
2. ability to adequately excite all target modes (using modal gains, see Equation (3.48)).
Employing the above criteria, diagonally oriented shakers (and in-line accelerometer
channels intended for measurement of drive point frequency response functions, to be discussed
in Chapter 4) were allocated as depicted in Figure 3.18.

38

25
1

Z
Y X

Location (Coordinates) Direction Cosines (Radians)

TAM Node Description X (in) Y (in) Z (in) Fx/Ftot Fy/Ftot Fz/Ftot
1 Lower-Center-Front 776.38 0 380.31 0.58 0.58 0.58
25 Lower-Left-Rear 890.33 -54.5 380.31 0.58 0.58 0.58
38 Grapple Fixture 820.24 0 466.16 0.58 0.58 -0.58

Figure 3.18: ISS P5 excitation allocation.

The anticipated adequacy of excitation resources was evaluated on the basis of predicted
modal gains associated with the test article finite element model, as summarized in Table 3.6.
Note that the most prominent shaker location for each individual mode is designated in bold
print and modes with non-negligible test fixture kinetic energy are shaded.
64 3. SYSTEMATIC MODAL TEST PLANNING
Table 3.6: Predicted adequacy of ISS P5 modal excitation resources

Kinetic Energy (%) Normalized Modal Gain, [ΦTΓ]

Mode Freq (Hz)
Truss Grapple Fixture 1 (Truss) 25 (Truss) 38 (Grapple)

3.3.5 CLOSURE
The theoretical basis of the RKE method for accelerometer allocation and TAM mass matrix
definition as part of the modal test planning process has been described. An early application
of the RKE method for the ISS P5 modal test, conducted in 2001 is summarized in the above
discussion. One of many subsequent modal tests conducted in the U.S. aerospace community,
specifically NASA MSFC’s ISPE modal test, conducted in 2016, is employed as an illustrative
example throughout this book. In addition to accelerometer allocation, the allocation of shaker
resources based on pre-test FEM predicted modal gains (a common practice) completes the
process of accelerometer and excitation allocation specification.

3.4 REFERENCES
[1] Load analysis of spacecraft and payloads, NASA-STD-5002, 1996. 47, 59
[2] D. Williams, Dynamic loads on aeroplanes under given impulsive load with particular
reference to landing and gust loads on a large flying boat, Great Britain Royal Aircraft
Establishment Reports SME 3309 and 3316, 1945. 51
[3] R. Guyan, Reduction of stiffness and mass matrices, AIAA Journal, 3, 1965. DOI:
10.2514/3.2874. 59
[4] Independent structural loads analysis, U.S. Air Force Space Command, SMC-S-004, 2008.
59
[5] R. N. Coppolino, Automated response DOF selection for mapping of experimental nor-
mal modes, IMAC XVI, 1998. 60
 65

CHAPTER 4

Measured Data Analysis

4.1 PART 1: PRELIMINARY MEASURED DATA ANALYSIS
4.1.1 INTRODUCTION
Physical phenomena are either deterministic or random (non-deterministic) [1]. A determin-
istic time history record follows an explicit mathematical relationship (e.g., a sinusoid) while a
random time history can only be known in terms of statistical characteristics (e.g., mean value,
standard deviation). Deterministic time history data classifications are summarized in Table 4.1.

Table 4.1: Deterministic time history data classifications

Deterministic Time History Records

Class Sub-Class Example

Periodic Sinusoidal A · sin(ωt)

N
Complex Periodic ∑ An sin(nω 0t + λn)
n=1
N
∑ An sin(ωnt + λn)
Non-Periodic Almost Periodic n=1
ωn+1/ωn is not an integer
N
Transient ∑ Ane-σnt sin(ωnt + λn)
n=1

Several descriptive terms for random data records are now introduced. A collection of
time history records from separate “tests” is called an ensemble. A single test record is called
stationary if the statistical properties (e.g., mean, standard deviation, etc.) are the same for all
sub-records of “reasonable” duration (selected segments from the entire test record). Otherwise,
the record is classified as non-stationary. If each data record within an ensemble is stationary
and the statistical properties of all data records are the same, the ensemble is classified as ergodic.
Note that all ergodic processes are stationary, but all stationary processes are not ergodic. Random
data classifications [1] are summarized in Table 4.2.
Preliminary measured data analysis (which employs basic data analysis functions) is re-
quired for two primary reasons, specifically:
66 4. MEASURED DATA ANALYSIS
Table 4.2: Random time history data classifications

Random Time History Records

Class Sub-Class Example

Stationary Ergodic All tests have the same standard deviation

All tests are stationary, but individual record
Non-Ergodic
"intensities" differ

Non-Stationary See Bendat and Piersol [1] See Bendat and Piersol [1]

1. overall measured data quality and content evaluation and

2. detection of nonlinear dynamics for a subject test article.

4.1.2 KEY PRELIMINARY MEASURED DATA ANALYSIS FUNCTIONS,

MEAN, VARIANCE, AND STANDARD DEVIATION
The mean value, , of a time history data record, Œx, with “N ” samples, is calculated within
as [1]
N
X
 D .1=N / x.n/; (4.1)
nD1

and the (sample) variance,  2 , of a time history data record is calculated as [1]
N
X
 2 D .1=N / Œx.n/ 2 : (4.2)
nD1

The (sample) standard deviation,  , is simply the square root of the (sample) variance.

4.1.3 NORMALIZED PROBABILITY DENSITY AND IDEAL GAUSSIAN

DISTRIBUTION
The normalized probability density function for a measured time history record is calculated
based on the normalized data record, z , defined as [1]

z D .x /=: (4.3)
Dividing the normalized data record (composed of “N ” samples) into “n” equally spaced
containers or “bins” between the minimum and maximum values of z , the histogram function,
h.z/, is formed as the number of samples within each “bin” divided by “N ”. For a data record,
the normalized probability density function is defined as [1];
 4.1. PART 1: PRELIMINARY MEASURED DATA ANALYSIS 67
   
n h.z/
p.z/ D
: (4.4)
zmax zmin N
The ideal normalized Gaussian probability density function, pG .z/, is defined as [1];
 p  2
pG .z/ D 1= 2
e .z =2/ : (4.5)
Both the actual normalized probability density and normalized Gaussian probability den-
sity functions have the integral property [1]
Z1 Z1
p.z/dz D pG .z/dz D 1: (4.6)
1 1

4.1.4 TOTAL NORMALIZED PROBABILITY FUNCTION

The total probability that the normalized time history will fall within a specified range is defined
as [1],
Zz2
P .z1  z  z2 / D p.z/dz; (4.7)
z1

and the probability that the value of z will not exceed an upper bound, z2 , is defined as
Zz2
P . 1  z  z2 / D p.z/dz: (4.8)
1

4.1.5 AUTOSPECTRUM OR POWER SPECTRAL DENSITY FUNCTION

The autospectrum or power spectral density function is defined in the literature by a variety of
relationships. For a thorough exposition of this function [1], a physically descriptive definition
of the autospectrum is based on the way this quantity was estimated prior to the popular use of
digital signal processing techniques. The analog process for estimating autospectrum involved
four steps.
1. Filter a time history, x.t/, with a narrow bandpass filter of bandwidth, f , and center
frequency, f (Hz), resulting in x.f; f; t /.
2. Form the square of the filtered signal, x 2 .f; f; t /.
3. Calculate the average value of the squared, filtered signal,

ZT
2 1
xN .f; f; t/ D x 2 .f; f; t/dt : (4.9)
T
0
68 4. MEASURED DATA ANALYSIS
4. Divide the result by the filter bandwidth, f , resulting in the autospectrum (or power
spectrum),

ZT
xN 2 .f; f; t / 1
Gxx .f / D D x 2 .f; f; t/dt : (4.10)
f .f
T /
0

The (real-valued) autospectrum provides information on the frequency distribution of the

standard deviation of a time history record. It therefore has the following property:

Z1
2
 D Gxx .f /df : (4.11)
0

A modern digital, state-of-the-art method for calculating the autospectrum employs the
finite Fourier transform of the time history record. By subdividing the record into “nd ” distinct
sub-records or “windows” (of duration, T ) the autospectrum is defined as,

nd
2 X
Gxx .f / D
Xi .f /  Xi .f /: (4.12)
.nd T /
i D1

X.f / is the finite Fourier transform of the time domain series, x.t/, where the temporal sam-
pling rate is t . The autospectrum, GXX .f /, is calculated at discrete frequencies, fk D k
f ,
where 0  k  nw =2. Important guidelines for computation of an appropriate autospectrum are:

1 1
fmax D (Nyquist frequency); f D (bandwidth resolution)
2
t T
T (4.13)
nw D (discrete Fourier transforms “window” length)
t
Tmax D nd
T (total record length for “nd ” distinct averages):

In practice, the originally measured “analog” signal, x.t/, should be low-pass filtered with
a “cut-off ” frequency (typically 0:8  fmax ) in order to avoid aliasing before conversion to a time
series with digital bandwidth, fmax . In addition, a sufficiently long data record, Tmax , should be
recorded to estimate an autospectrum associated with a desired bandwidth resolution, f , and
number of distinct averages, nd (typically in excess of 10 to minimize contributions associated
with extraneous noise sources). It should finally be noted that specialized “windowing” functions
and correction factors (e.g., the Hanning window and “bow-tie” correction factors) and overlap
processing (non-distinct successive records) are commonly applied practices [1].
 4.1. PART 1: PRELIMINARY MEASURED DATA ANALYSIS 69
4.1.6 CROSS-SPECTRUM OR CROSS-POWER SPECTRAL DENSITY
FUNCTION
Closely related to the autospectrum is the cross-spectrum, which will be featured in Part 2 of
this chapter. In a manner similar to Equation (4.13), the digital cross-spectrum is defined as [1]

nd
2 X
Gyx .f / D
Yi .f /  Xi .f /: (4.14)
.nd T /
i D1

Guidelines and practices for estimation of a digital cross-spectrum and autospectrum are
identical.

4.1.7 THE SPECTROGRAM FUNCTION

An extremely informative application of the autospectrum, namely the spectrogram, is defined
as a “running” autospectrum, which is a function of individual or “grouped” autospectra that are
computed at successive time segments of a data record. The value of spectrograms and all of the
above described preliminary measured data analysis functions is demonstrated in the following
discussion.
A series of illustrative examples are provided to demonstrate application of preliminary
data analysis fundamentals in various situations. Overall classification of individual time history
records is effected through calculation and display of the time history, spectrogram, autospec-
trum and probability density function. Data quality and content is evaluated by closer examina-
tion of the autospectrum (including the spectrogram), probability density and total probability,
and response spectrum.

4.1.8 ILLUSTRATIVE EXAMPLE: SINUSOIDAL TIME HISTORY WITH

BACKGROUND RANDOM NOISE
Consider a measured time history record consisting of a 20 Hz sinusoidal signal, contaminated
by broadband random noise, and sampled at dt D :005 s over a duration of 50 s. The composite
preliminary data analysis display shown in Figure 4.1 includes a color spectrogram (upper left),
time history trace (lower left), autospectrum (upper right), and probability density (lower right).
The dashed curve in the probability density plot indicates the ideal Gaussian probability density
function. In addition, the title indicates the window length .NW /, frequency bandwidth (f ),
and number of distinct averages (Nav ) of autospectrum sub-records, as well as the mean and
standard deviation (Std) values for the data record.
Based on general uniformity of the spectrogram and time history plots with respect to
time, the data record is judged stationary. Both the spectrogram and autospectrum indicate that
the record is dominated by a 20 Hz sinusoidal signal. The record, in spite of the presence of
broad band noise, may be classified as predominantly deterministic due to general consistency
70 4. MEASURED DATA ANALYSIS
of the sinusoidal component. The probability density function differs substantially from the ideal
Gaussian distribution in a manner suggesting a sinusoid (see [2]).

File: Sinex1.mat Chan:Channel1 |dt=0.005, Nw=512, df=0.39, Nav=19, Mean=0.0052, Std=0.82|

100

80 -10

20 Log10 (PSD)
Frequency
-20
60
-30

40 -40
-50
20 -60
0 20 40 60 80 100
0 Frequency
5 10 15 20 25 30 35 40 45
Time

2 0.4
1 0.3
Signal

PDF
0 0.2
-1
0.1
-2
5 10 15 20 25 30 35 40 45 0 -2 -1 0 1 2
Time Standardized Variable

Figure 4.1: Preliminary data analysis of a sinusoidal time history with random noise record.

4.1.9 ILLUSTRATIVE EXAMPLE: SDOF LINEAR SYSTEM RESPONSE TO

RANDOM EXCITATION
An SDOF linear system, described by the second-order dynamic equation,

u.t/ P C !n2 u.t/ D F .t/;

R C 2 n !n u.t/ (4.15)
is subjected to broad band Gaussian random excitation, F .t/, and has the displacement response,
u.t /. Both excitation and response are sampled at dt D :005 s over a duration of 50 s. The (pre-
liminary data analysis displays for excitation, F .t /, and response, u.t /, are shown in Figure 4.2.
General uniformity of the spectrogram and time history plots with respect to time for both
excitation and response records indicate that the process is stationary. Both the spectrogram and
autospectrum for the excitation indicate broad band frequency content. The output, u.t/, spec-
trogram and autospectrum reveal narrow band random response character. Gaussian probability
density distribution for the response history suggests that the system is a linear SDOF system
with 20 Hz natural frequency [1].
 4.1. PART 1: PRELIMINARY MEASURED DATA ANALYSIS 71

Figure 4.2: Linear SDOF system broadband random excitation (left) and response (right).

4.1.10 ILLUSTRATIVE EXAMPLE: ISPE MODAL TEST

In 2017, NASA/MSFC conducted a modal test on the Integrated Spacecraft and Payload Ele-
ment (ISPE), general overview (courtesy of NASA/MSFC) depicted in Figure 4.3.

Figure 4.3: ISPE test article overview.

The modal test sensor suite included four electrodynamic shakers (excitation forces) and
273 accelerometer channels (dynamic responses). Preliminary measured data analysis displays
for excitation “1” and response “1” are shown in Figure 4.4.
Both the spectrogram and autospectrum for the excitation indicate broadband frequency
content. The drop-off in the autospectrum above 90 Hz is the result of low-pass filtering of
measured data histories. The response spectrogram and autospectrum reveal the presence of
many peaks in the 15–90 Hz frequency band, suggesting the presence of many test article modes.
Gaussian probability density distribution for the response history suggests that the ISPE test
article exhibited linear MDOF system behavior in the 15–90 Hz frequency band.
72 4. MEASURED DATA ANALYSIS

Figure 4.4: ISPE test article broad band random excitation (left) and response (right).

4.1.11 ILLUSTRATIVE EXAMPLE: WIRE ROPE TEST ARTICLE

In 2006, during the conducting of U.S. Army sponsored research at Measurement Analysis
Corporation, time history data was measured on a test article consisting of a “rigid mass” sup-
ported by four wire-rope isolators, which was subjected to random excitation delivered by a
vertically oriented electrodynamic shaker. Time histories of acceleration of the rigid mass and
the shaker applied load were recorded. Employing a series of mathematical operations on the
measured data records, time history records of (a) isolator deflection and (b) isolator internal
force were estimated. In one particular test, broadband random excitation was delivered by the
shaker. Preliminary data analysis of the estimated isolator deflection and isolator internal force
probability density functions is summarized, along with a photograph of the test configuration,
in Figure 4.5.
The distortion in isolator internal force probability density with respect to ideal Gaussian
probability density indicates the presence of nonlinear behavior [2].

4.1.12 ILLUSTRATIVE EXAMPLE: ISS P5 MODAL TEST

During conduct of the ISS P5 modal test in 2001, nonlinear behavior of the test article was
suspected. In particular, modal frequencies and mode shapes associated with successive data
records were inconsistent with one another. As a result, a series of data records were acquired
and subjected to in-depth preliminary data analysis. The source of anomalous dynamic behavior
was quickly localized to a left-side trunnion that connected the ISS P5 test article to the test
fixture; left- and right-side trunnion locations are illustrated in Figure 4.6.
The preliminary data analysis strategy employed to evaluate and locate suspected nonlinear
behavior was based on the theoretical premise that:

a. response of a linear system to sinusoidal excitation at one frequency will also occur at the
excitation frequency;
 4.1. PART 1: PRELIMINARY MEASURED DATA ANALYSIS 73

Enidine WR2-200-10B
X Z

Wire-Rope Isolators

Isolator Deflection Isolator Internal Force

Figure 4.5: Wire rope isolators subjected to broadband random excitation.

Left Trunnion Right Trunnion

Figure 4.6: ISS P5 test article trunnion locations.

74 4. MEASURED DATA ANALYSIS
b. response of a system with localized nonlinearity to sinusoidal excitation at one frequency
will include frequency components at other frequencies (harmonic distortion); and

c. nonlinear harmonic distortion tends to be most prominent at locations near the source of
nonlinear behavior (advice provided by the late Allan G. Piersol).

Employing swept-sine applied load excitation (shaker 2 only), response activity of all 258
measured accelerometer channels was reviewed. Swept-sine responses in the form of spectro-
grams and focused response spectrum (SRS) calculations were used to identify the source of
nonlinear behavior. Results associated with left- and right-side trunnion locations are illustrated
in Figure 4.7.

SRS SRS

Left-Side Trunnion Right-Side Trunnion

Figure 4.7: Trunnion responses to swept-sine excitation.

Clearly, the left-side trunnion indicates strong harmonic distortion (second harmonic) in
the computed SRS function, while the right-side trunnion indicates minor harmonic distortion.

4.1.13 CLOSURE
Preliminary measured data analysis is vital to integrity of all experimental steps of the integrated
test analysis process. Failure to conduct a thorough preliminary screening of measured time
history records is known to result in unreliable and ambiguous outcomes for modal and (more
general) dynamic characterization tests.
An effective strategy for screening of measured data records includes review of probability
densities, autospectra, and spectrograms for all measured time history channels. When the sub-
ject test article is subjected to continuous, stationary, broadband random excitation(s) following
Gaussian probability distributions, two outcomes are possible: (1) all responses follow Gaussian
probability distributions indicating linear system behavior and (2) some responses exhibit prob-
ability distributions that deviate from ideal Gaussian distribution indicating nonlinear behavior.
Measured ISPE test data indicates clearly linear system behavior, while measured wire rope test
data indicates strongly nonlinear behavior.
4.2. PART 2: FREQUENCY RESPONSE FUNCTION ESTIMATES FROM MEASURED DATA 75
A second strategy for screening of measured data records includes review of swept sine
excitation induced spectrograms for all excitation and response records. The occurrence of har-
monic distortion (frequency content other than the excitation frequency) provides an indication
of nonlinear dynamic behavior. Moreover, in the case of localized nonlinear behavior, proximity
of response channels to a nonlinear mechanism is indicated by severity of harmonic distortion.
Application of this strategy was quite effective in localization of nonlinear behavior during ISS
P5 modal testing. In addition, focused response spectrum analyses during time segments indict-
ing strongest nonlinear activity provided confirming data related to nonlinear behavior.

4.2 PART 2: FREQUENCY RESPONSE FUNCTION

ESTIMATES FROM MEASURED DATA
4.2.1 INTRODUCTION
The response characteristics of linear dynamic systems, namely frequency response functions
(FRFs) are theoretically expressed in the frequency domain based on response to simple har-
monic excitation. FRFs are estimated employing spectral analysis techniques [1] of measured
excitation and response time history records. Quality of the measured FRF estimates is judged
on the basis of coherence functions that are indicative of “signal to noise” ratios as functions of
frequency.
Well-developed spectral analysis techniques were adapted in the mid-1990s [2] to ad-
dress estimation and identification of dynamic systems exhibiting “algebraic” nonlinear behav-
ior. Treatment of systems exhibiting “hyseteretic” nonlinear behavior, however generally require
employment of specialized time-domain strategies.
After completion of preliminary data evaluations, detailed analyses are often performed
to estimate input-output characteristics of a subject dynamic system. The following sections
provide an overview of single input/single output (SI/SO) and multiple input/single output
(MI/SO) data analysis procedures. Spectral and correlation analysis of SI/SO and MI/SO data
records is the primary tool used for estimation of system FRF characteristics.

4.2.2 MI/MO FREQUENCY RESPONSE EQUATIONS FOR SYSTEMS

WITH ALGEBRAIC NONLINEARITIES
Consider the time domain matrix equation set (introduced in Chapter 3), which includes non-
linear force terms, FN .t /,

˚ ˚
ŒM  UR .t / C ŒB UP .t / C ŒK fU.t/g D Œ€e  fFe .t/g C Œ€N  fFN .t /g : (4.16)

The companion Fourier transform of the above equation set is expressed as

76 4. MEASURED DATA ANALYSIS

˚ h 1
i h
1
i
UR .f / D Z.f / €e fFe .f /g C Z.f / €N fFN .f /g
D ŒHe .f / fFe .f /g C ŒHN .f / fFN .f /g ; (4.17)

where the frequency domain impedance matrix is defined as

h i
ŒZ.f / D .2
f / 2 K C i
.2
f /
B .2
f /2
M : (4.18)

The FRF matrices and forces may be “folded” into the all-encompassing matrices and
arrays
 
    Fe .f /
R .f / D
HUF He .f / HN .f / ; fF .f /g D : (4.19)
FN .f /
Therefore, the general FRF relationship that implicitly includes the nonlinear force terms
adopts the common linear input-output relationship,

˚  
UR .f / D HUF
R .f / fF .f /g or more generally fY.f /g D ŒHYX .f / fX.f /g : (4.20)

Benefits associated with “folding” in of the nonlinear partitions into the more convenient
linear form will be exploited in development of nonlinear system estimation strategy (for systems
with “algebraic” nonlinearities).

4.2.3 SI/SO, MI/SO AND MI/MO FREQUENCY RESPONSE FUNCTION

ESTIMATION
For simplicity, the frequency domain arrays and FRF matrix described in Equation (4.20) are
now presented in a form dropping the brackets and frequency designations. Therefore, the
MI/MO relationship describing measured data that includes noise is

Y D H
X C N: (4.21)
Post-multiplying the above equation set by the complex conjugate transpose (Hermitian
conjugate) of the force array results in




Y
X D H
Y
X C N
X : (4.22)
Post-multiplication of Equation (4.21) by its Hermitian conjugate results in

Y
Y  D H
X
X
H  C N
X
H  C H
N
X C N
N  : (4.23)
4.2. PART 2: FREQUENCY RESPONSE FUNCTION ESTIMATES FROM MEASURED DATA 77
Application of the averaging process used to define the autospectrum and cross-spectrum
for individual time history series (see Equations (4.11)–(4.14)), the following spectral matrix
relationships are defined (now reintroducing the matrix, array, and frequency labels):

ŒGNX .f / D Œ0 ; ŒGXN .f / D Œ0 (uncorrelated excitations and noise); (4.24)

ŒGYX .f / D ŒHYX .f /
ŒGXX .f / (output-input relationship); (4.25)

ŒGYY .f /C D ŒHXX .f /
ŒGXX .f /
ŒHYX .f / (coherent output autospectrum); (4.26)
ŒGYY .f / D ŒGYY .f /C C ŒGNN .f / (total output autospectrum): (4.27)

In the case of a single applied force excitation and a single response variable, SI/SO alge-
braic manipulation yields the following estimates:

HYX .f / D GYX .f /=G XX .f / (SI/SO FRF); (4.28)

2 2
2 jHYX .f /j GXX .f / jGYX .f /j
YX .f / D D
GYY .f / GXX .f /
GYY .f /
1 (SI/SO ordinary coherence): (4.29)

Multiple Input/Single Output or MI/SO analysis represents a simple extension to SI/SO

analysis. The most direct relationship for MI/SO frequency functions involves direct frequency-
by-frequency inversion of the cross-spectral matrix of generally partially correlated inputs
(forces), i.e.,

1
ŒHYX .f / D ŒGYX .f /
ŒGXX .f / ; (MI/SO FRF array): (4.30)
In addition, since the MI/SO coherent output and total output autospectra are non-matrix
functions, the total ordinary coherence function is simply,

2 .GYY .f //C
YX .f / D  1; (MI/SO ordinary coherence): (4.31)
GYY .f /
The above MI/SO formulation does not offer details related to contributions associated
with multiple excitations to the output response. An alternative MI/SO analysis formulation
using triangular decomposition is thoroughly developed by Bendat and Piersol [1].
Cholesky factorization of the input cross-spectral matrix, ŒGXX .f / yields,

ŒGXX .f / D Œ€XZ .f /
Œ€ZX .f / ; (4.32)

where Œ€XZ .f / is a complex, lower triangular matrix and Œ€ZX .f / is its upper triangular, Her-
mitian transpose. A triple product decomposition version of Cholesky factors is formed by unit
diagonal normalization of Œ€XZ .f / and Œ€ZX .f / resulting in
78 4. MEASURED DATA ANALYSIS

ŒGXX .f / D ŒLXZ .f /
ŒGZZ .f /
ŒLZX .f / ; (4.33)

where ŒLXZ .f / and ŒLZX .f / are lower and upper triangular, unit diagonal matrices, respectively,
and ŒGZZ .f / is a positive-definite, diagonal matrix.
Physical significance of the normalized triangular factors is recognized by noting the input
variable transformation

fX.f /g D ŒLXZ .f /
fZ.f /g : (4.34)

The successive terms in fZ.f /g are

Z1 .f / D X1 .f /;
Z2 .f / D X2 .f / “swept” of contributions from X1 .f /; (4.35)
Z3 .f / D X3 .f / “swept” of contributions from X1 .f / and X2 .f /: : ::

Therefore, ŒGZZ .f / is the diagonal autospectrum matrix of uncorrelated, generalized inputs,

fZ.f /g.
Substitution of the triple product ŒGXX .f / decomposition relationship (Equation (4.33))
into Equation (4.30) results in the frequency response functions associated with uncorrelated
generalized inputs, fZ.f /g,

1
ŒHYZ  D ŒGYZ 
ŒGZZ  ; (4.36)
where

1
ŒHYZ  D ŒHYX 
ŒLXZ  ; ŒGYZ  D ŒGYX 
ŒLZX  : (4.37)
Therefore, the frequency responses associated with physical inputs, fX.f /g, are now expressed
as

1
ŒHYX  D ŒHYZ 
ŒLXZ  : (4.38)
The generalized MI/SO equation in terms of generalized inputs is defined as

Y.f / D ŒHYZ .f /
fZ.f /g C N.f /: (4.39)

Since the generalized inputs, are uncorrelated, the output autospectrum (dropping the
“f ” designation for simplicity) is expressed as

ˇ ˇ2 ˇ ˇ2 ˇ ˇ2
GYY D ˇHYZ 1 ˇ
GZ1 Z1 C ˇHYZ 2 ˇ
GZ2 Z2 C


C ˇHYZ N ˇ
GZK ZK C GNN : (4.40)
4.2. PART 2: FREQUENCY RESPONSE FUNCTION ESTIMATES FROM MEASURED DATA 79
Coherent output autospectra and cumulative coherence function pairs are therefore defined
as follows:

ˇ ˇ2 ˇ ˇ2 
.GYY /1 D ˇHYZ 1 ˇ
GZ1 Z1 ; Y2 1 D ˇHYZ 1 ˇ
GZ1 Z1 =GYY for input x1 .t/; (4.41)
ˇ ˇ2 ˇ ˇ2
.GYY /2 D ˇHYZ 1 ˇ
GZ1 Z1 C ˇHYZ 2 ˇ
GZ2 Z2 ;
ˇ ˇ2 ˇ ˇ2 
Y2 2 D ˇHYZ 1 ˇ
GZ1 Z1 C ˇHYZ 2 ˇ
GZ2 Z2 =GYY for inputs x1 .t/ C x2 .t/; (4.42)
ˇ ˇ2 ˇ ˇ2 ˇ ˇ2
.GYY /K D ˇHYZ 1 ˇ
GZ1 Z1 C ˇHYZ 2 ˇ
GZ2 Z2 C


C ˇHYZ N ˇ
GZK ZK ;
ˇ ˇ2 ˇ ˇ2 ˇ ˇ2 
2
YK D ˇHYZ 1 ˇ
GZ1 Z1 C ˇHYZ 2 ˇ
GZ2 Z2 C


C ˇHYZ N ˇ
GZK ZK =G YY ;
for inputs; x1 .t/ C


C xk .t /: (4.43)

The cumulative coherence function family has the property

0  Y2 1  Y2 2  : : :  YK
2
 1: (4.44)

When graphically displayed as a function of frequency, the coherence function family ap-
pears as a waterfall plot series that clearly indicates the relative contributions of the accumulated
excitation sources. The benefits of such a display will be demonstrated in a series of illustrative
examples.
Finally, it is noted that MI/MO analysis represents a simple extension to MI/SO analysis.

4.2.4 ILLUSTRATIVE EXAMPLE: ISPE MODAL TEST

The ISPE test article was excited with broad band random excitation forces at four separate
locations, and MI/MO correlation and spectral analysis was preformed employing 8192 length
windows with 50% overlap processing. Cumulative coherences for all 265 TAM response chan-
nels plus 4 drive point response channels were computed. MI/MO plots associated with all 4
excitation forces and 4 drive point responses are detailed in Figure 4.8.
The plots in Figure 4.8 are arranged in a 4  4 matrix “plot map.” For example, the plots
in the (2,3) position correspond to excitation 2, drive point response 3. Each plot “i; j ” entry
includes the cumulative coherence (top), FRF phase angle (middle), and FRF magnitude (bot-
tom). An important feature of the plot matrix is FRF “reciprocity” (e.g., the “i; j ” component
is consistent with the “j; i ” component, with regard to both magnitude and phase of the FRF).
Greater detail can be discerned from the plot group for the “1, 1” component shown in Fig-
ure 4.9. Of particular interest is the (top) cumulative coherence plot, which (a) indicates the
successive contributions of the first three excitations (to drive point 1 response autospectum)
that are substantially below unity and (b) indicates the significant role of the fourth excitation
producing near unit cumulative coherence.
80 4. MEASURED DATA ANALYSIS

Figure 4.8: Representative ISPE MI/MO FRF and cumulative coherence plots.

File: ISPETESTDFRFs…MI/MO: Navg=16, df=0.025, Input(1)=2302104X-, Output(1)=2302104X+ 21-Sep-2019 3:54 PM

1
Coherence

0.5

0
15 20 25 30 35 40 45 50
Hz

200
Phase

-200
15 20 25 30 35 40 45 50
Hz

10-1

10-2
Magnitude

10-3

10-4
15 20 25 30 35 40 45 50
Hz

Figure 4.9: ISPE MI/MO FRF and cumulative coherence plots for excitation 1, drive point
response 1.
4.2. PART 2: FREQUENCY RESPONSE FUNCTION ESTIMATES FROM MEASURED DATA 81
The cumulative coherences associated with nearly all response channels were close to unity
confirming extremely high-quality FRF estimates.

4.2.5 ILLUSTRATIVE EXAMPLE: WIRE ROPE MI/SO TEST DATA

ANALYSIS
Recalling the preliminary measured data analysis, summarized in Figure 4.5, the hypothe-
sized “algebraic” nonlinear system composed of measured time histories (applied force and
acceleration response) and synthesized “measured” time histories (cubed displacement and
velocity
jvelocityj) depicted in Figure 4.10 was subjected to MI/SO analysis.

F(t) H1

F (U(t))3 H2 Ü(t)

U
U̇(t) ∙ |U̇(t)| H3

Figure 4.10: Hypothesized MI/SO nonlinear system.

The cumulative coherence plot, shown in Figure 4.11 indicates that incorporation of the
two nonlinear terms produces a nearly unit value cumulative coherence (red curve), while the
ordinary coherence associated with a linear model (blue curve) indicates reduced coherence.
The results from Figure 4.11 provide clear evidence that the behavior of the wire rope
isolators is nonlinear. However, further definitive data analysis, described in Chapter 7, will
indicate that the nonlinear behavior is “hysteretic” rather than “algebraic” as currently hypoth-
esized. That being said, the present “algebraic” nonlinear model serves an important role as an
“intermediate” data analysis.

4.2.6 ILLUSTRATIVE EXAMPLE: ISS P5 MODAL TEST

The ISS P5 test article was excited with broadband random excitation forces at three separate
locations, MI/MO correlation and spectral analysis was preformed employing 1024 length win-
dows with 50% overlap processing. Cumulative coherences for all 261 TAM response channels
plus 3 drive point response channels were computed. MI/MO plots associated with the three
drive point FRFs (complete drive point FRF “matrix” as depicted for the ISPE in Figure 4.8 is
not provided here) are detailed in Figure 4.12.
82 4. MEASURED DATA ANALYSIS

File: WireNonR…MI/MO: Navg=29, df=0.488, MI/SO Cumulative Coherence

1

0.9
Due to Force
0.8 Due to Force + U3Z3
Due to Force + U3Z3 + V3Z*|V3Z|

Cumulative Coherence
0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
10 15 20 25 30 35 40 45 50
Hz

Figure 4.11: Hypothesized nonlinear system cumulative coherence.

Drive Point FRF “1” Drive Point FRF “25”

1 1
Coherence

Coherence
0.5 0.5

0 0
10 15 20 25 30 35 40 45 50 10 15 20 25 30 35 40 45 50
200 Hz 200 Hz
Phase

Phase
0 0

-200 -200
10 15 20 25 30 35 40 45 50 10 15 20 25 30 35 40 45 50
Hz Hz
-1 -1
10 10

-2 -2
Magnitude

10 Magnitude 10

-3 -3
10 10

-4 -4
10 10
10 15 20 25 30 35 40 45 50 10 15 20 25 30 35 40 45 50
Hz Hz

Drive Point FRF “38” ISS P5 Test Article Drive Points

Coherence

38
10 15 20 25 30 35 40 45 50
Hz
Phase

25
10 15 20 25 30 35 40 45 50 1
Hz
Magnitude

Z
10 15 20 25 30 35 40 45 50 Y X
Hz

Figure 4.12: ISS P5 MI/MO FRF and cumulative coherence plots.

 4.3. REFERENCES 83
The reduced cumulative coherences in the 30–35 Hz frequency band were judged pro-
vide additional evidence of test article nonlinear behavior, which was noted in preliminary data
analysis results summarized in Figure 4.7.

4.2.7 CLOSURE
Systematic techniques for estimation of FRFs associated with single and multiple excitation
resources, due to Bendat and Piersol, provide a sound foundation for detailed measured analy-
sis of measured time history records. Automation and reinterpretation of triangular (Cholesky)
decomposition of the excitation cross-spectral (frequency dependent) matrix leads to a straight-
forward, unambiguous cumulative coherence function family. The cumulative coherence family
provides insight into the role played by each individual excitation (which is not necessarily com-
pletely uncorrelated with the other excitation sources). Benefits offered by cumulative coherence,
as demonstrated with modal test data records, include: (1) further confirmation and localization
of ISS P5 test article behavior (noted in preliminary data analyses) and (2) confirmation of ISPE
test article linear behavior, along with determination of the relative prominence of modal test ex-
citation resources. In addition, employment of cumulative coherence facilitates characterization
of “algebraic” nonlinear systems as demonstrated in testing of a wire rope shock and vibration
isolator.

4.3 REFERENCES
[1] J. S. Bendat and A. G. Piersol, Random Data Analysis and Measurement Procedures, 4th
ed., Wiley, 2010. DOI: 10.1002/9781118032428. 65, 66, 67, 68, 69, 70, 75, 77
[2] J. S. Bendat, Nonlinear Systems Techniques and Applications, 2nd ed., Wiley, 1998. 70, 72,
75
 85

CHAPTER 5

Experimental Modal Analysis

5.1 PART 1: PRELIMINARY EXPERIMENTAL MODAL
ANALYSIS
5.1.1 INTRODUCTION
The discipline of experimental modal analysis developed over two distinct time periods, namely:
(1) the analog era prior to 1971 and (2) the digital era from 1971 to the present time. Key
publications offering an historical perspective on the discipline have been authored by Bishop
and Gladwell [1], covering the analog era up to 1961, and Brown and Allemang [2], spanning
the analog and digital eras through 2007. In spite of availability of many excellent automated
experimental modal analysis procedures, developed during the digital era, review of measured
FRF data (preliminary experimental modal analysis) is always a prudent step in the overall ex-
perimental modal analysis process. Preliminary experimental modal analysis is highly dependent
on analysis of graphical FRF displays based on (1) strategies dating back to the analog era and
(2) modal indicator displays introduced during the digital era [3].

5.1.2 FREQUENCY RESPONSE CHARACTERISTICS OF LINEAR SDOF

SYSTEMS
An elementary linear structural dynamic or mechanical dynamic system is described in terms
of a SDOF linear mechanical system. The dynamic response, u.t/, of a SDOF system which is
excited by an applied force, F .t / and/or base (foundation) motion input, u0 .t/, is governed by
the ordinary differential equation (schematic depicted in Figure 5.1),

mu.t
R / C b u.t
P / C ku.t/ D F .t / C b uP 0 .t/ C ku0 .t/; (5.1)
where m, b , and k are the constant mass, viscous damping, and elastic stiffness coefficients,
respectively. By defining the relative displacement variable, uR .t/ D u.t/ u0 .t/, and dividing
by the mass, m, Equation (2.1) simplifies to

uR R .t/ C 2 n !n uP R .t/ C !n2 uR .t/ D F .t/=m uR 0 .t/; (5.2)

where the undamped natural frequency (rad/sec) and critical damping ratio, respectively, are,
r
k b
!n D ; n D : (5.3)
m .2m!n /
86 5. EXPERIMENTAL MODAL ANALYSIS
u(t), F(t)
m

k b uo(t)

Figure 5.1: Linear SDOF dynamic system schematic.

In addition, the damped natural frequency (rad/sec) is defined as,

q
!d D !n 1 n2 : (5.4)

Steady-state response of a SDOF system subjected to simple harmonic force excita-

tion, F .!/e i!t , is of the form, U.!/e i !t , where the following FRF quantities are derived after
substitutions into the system differential equation (and conversion of the frequency variable,
! D 2f ):

ku.f / 1
H.f / D D (displacement FRF); (5.5)
F .f / .1 .f =fn /2 C 2in .f =fn //
mu.f
R / .f =fn /2
HA .f / D D (acceleration FRF); (5.6)
F .f / .1 .f =fn /2 C 2in .f =fn //
b u.f
P / C ku.f / .1 C 2in .f =fn //
TR.f / D D (transmissibility): (5.7)
F .f / .1 .f =fn /2 C 2in .f =fn //

Steady-state response of a SDOF system subjected to harmonic base motion excitation,

u0 .!/e i!t , is of the form uR .!/e i!t , where the following FRF quantities are derived after sub-
stitutions into the system differential equation:

uR R .f / .f =fn /2
HA .f / D D (relative acceleration FRF); (5.8)
uR 0 .f / .1 .f =fn /2 C 2in .f =fn //
u.f
R / .1 C 2in .f =fn //
TR.f / D D (transmissibility): (5.9)
uR 0 .f / .1 .f =fn /2 C 2in .f =fn //

Each of the above frequency response functions exhibits peak response amplification when
the excitation occurs near the natural frequency. Moreover, the phase of each FRF shifts 180ı
as the excitation frequency increases from below to above the natural frequency. Shown in Fig-
ures 5.2–5.4 are plots of the three types of FRFs in terms of magnitude and phase, real and
imaginary parts vs. frequency .f =fn /, and real vs. imaginary parts (where fn D 1; n D :02).
 5.1. PART 1: PRELIMINARY EXPERIMENTAL MODAL ANALYSIS 87
FRF Display: Displacement FRF 04-Aug-1998 9:12 PM

Magnitude
100
0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5
Frequency (Hz)

20

Magnitude 0
10
101 0

Real
-5
-10

Imaginary
-10 -20

0.9 1 1.1 0.9 1 1.1

-15

-20 20
100
Phase (Deg)

10

Imaginary
-25
0 0
-10 0 10
-10
-100
Real
-20

0.9 1 1.1 0.9 1 1.1

Frequency (Hz) Frequency (Hz)

Figure 5.2: SDOF displacement FRF, H.f / vs. f .

FRF Display: Acceleration FRF 04-Aug-1998 9:23 PM
Magnitude

100

0.5 1 1.5
Frequency (Hz)

20
Magnitude

101
Real

0
20
Imaginary

-20

0.9 1 1.1 0.9 1 1.1

10

20
100
Phase (Deg)

Imaginary

0
0 0
-10 0 10

-100
Real
-20

0.9 1 1.1 0.9 1 1.1

Frequency (Hz) Frequency (Hz)

Figure 5.3: SDOF acceleration FRF, Ha .f / vs. f .

88 5. EXPERIMENTAL MODAL ANALYSIS
FRF Display: Transmissibility 04-Aug-1998 9:27 PM

Magnitude
100
0.5 1 1.5
Frequency (Hz)

20

Magnitude
101 0

Real
0

Imaginary
-10 -20

0.9 1 1.1 0.9 1 1.1

-20 20
100
Phase (Deg)

Imaginary
0 0
-10 0 10

-100
Real
-20

0.9 1 1.1 0.9 1 1.1

Frequency (Hz) Frequency (Hz)

Figure 5.4: SDOF transmissibility, TR.f / vs. f .

A closer examination of the SDOF acceleration FRF reveals some well-known properties
of this (as well as SDOF displacement and transmissibility FRF) response functions, as displayed
in Figure 5.5.
In particular, the red dot near the peak value of the FRF magnitude, imaginary com-
ponent, and polar plot is indicative of the natural frequency, fn . In addition, the frequency
spacing of peaks in the imaginary component, indicated by the green dots, facilitates estimation
of damping.

5.1.3 FREQUENCY RESPONSE CHARACTERISTICS OF LINEAR MDOF

SYSTEMS
Consider a structural dynamic system subjected to applied excitation forces,

ŒM  fug
R C ŒB fug
P C ŒK fug D Œ€e  fFe g : (5.10)

Employing the assumption of simple harmonic excitation and responses, in the same man-
ner as employed for SDOF systems, MDOF frequency response to each separate “unit” excita-
tion force is described as
 5.1. PART 1: PRELIMINARY EXPERIMENTAL MODAL ANALYSIS 89
FRF Display: Acceleration FRF 04-Aug-1998 9:23 PM

Magnitude
100

0.5 1 1.5
Frequency (Hz)

2ζnfn
20
Magnitude
101

Real
0
20

Imaginary
-20
1.1
0.9 1 0.9 1 1.1
10

20
100
Phase (Deg)

Imaginary
0
0 0
-10 0 10
Real
-100
-20

0.9 1 1.1 0.9 1 1.1

Frequency (Hz) Frequency (Hz)

Figure 5.5: SDOF acceleration FRF, Ha .f / vs. f with annotated points.

˚   1
fH .!/g D UR .!/ D ! 2 K C i! ŒB !2M f€e g (noting that ! D 2f ): (5.11)

Estimation of MDOF system frequency responses from measured time history data was
described in Chapter 4 along with illustrative example processed test data. It is instructive at
this point to review some of the ISPE modal test estimated FRF functions within the context
of preliminary experimental modal analysis.

5.1.4 ILLUSTRATIVE EXAMPLE: ISS P5 MODAL TEST

While preliminary experimental modal analysis (employing review of measured FRF data) was
conducted during P5 modal testing at NASA MSFC, the primary focus of that effort shifted
toward preliminary investigation of nonlinear behavior (as described in Chapter 4) and non-
repeatability (sensitivity) of the test article dynamic characteristics. Therefore, further discussion
of the ISS P5 modal test will resume in Part 2 of this chapter.

5.1.5 ILLUSTRATIVE EXAMPLE: ISPE MODAL TEST

Experimental FRF displays associated with drive point 1, and two selected response DOFs are
provided in Figures 5.6 and 5.7, respectively.
90 5. EXPERIMENTAL MODAL ANALYSIS
FRF Display: File: ISPETESTDFRFs.mat, Input(1)=2302104X-, Output(1)=2302104X+ 24-Sep-2019 11:35 AM

Magnitude
10-2

10 15 20 25 30 35 40 45 50 55 60
Frequency (Hz)

0.05

Magnitude

Real
0
0
10-2
-0.02

Imaginary
-0.05

20 30 40 50 60 -0.04 20 30 40 50 60
Phase (Deg)

0.05

Imaginary
100 -0.06

0 0
-0.02 0 0.02 0.04 0.06
-100 Real -0.05

20 30 40 50 60 20 30 40 50 60
Frequency (Hz) Frequency (Hz)

Figure 5.6: ISPE drive point 1 FRF detail.

FRF Display: File: ISPETESTDFRFs.mat, Input(1)=2302104X-, Output(134)=2302208X- 24-Sep-2019 11:39 AM

Magnitude

10 20 30 40 50 60
Frequency (Hz)

0.05
Magnitude

0.06

Real
0
0.04
Imaginary

0.02 -0.05
20 30 40 50 60 20 30 40 50 60
0
Phase (Deg)

0.05
100 -0.02 Imaginary
0 0
-0.02 0 0.02 0.04 0.06
-100 Real -0.05
20 30 40 50 60 20 30 40 50 60
Frequency (Hz) Frequency (Hz)

Figure 5.7: ISPE response point FRF detail.

The above two FRFs are certainly more complicated than the idealized SDOF FRFs de-
picted in Figures 5.2–5.5.
A more comprehensive overview of the “many modes” evident in the ISPE test article is
evidenced by display of two composite signature functions, namely: (a) a normalized square-
 5.1. PART 1: PRELIMINARY EXPERIMENTAL MODAL ANALYSIS 91
root of the summation of the autospectra for all modal test response accelerations and (b) a
global FRF skyline function. The global FRF skyline function is defined as a complex frequency
domain function whose real and imaginary parts are the sum of the absolute values of the real
component of response FRFs and the sum of the absolute values of the imaginary component
of response FRFs, that is,
NH
X NH
X
SKY.f / D jreal .Hk .f //j C i jimag .Hk .f //j : (5.12)
kD1 kD1

A display of the two global skyline functions for the ISPE test article is provided in Fig-
ure 5.8.
ISPESTSTD Skyline Functions
100
PSD1/2 Skyline

10-1

10-2

10-3
15 20 25 30 35 40 45 50 55 60 65
Frequency (Hz)
100
|FRF| Skyline

10-1

10-2
15 20 25 30 35 40 45 50 55 60 65
Frequency (Hz)
phase(FRF) Skyline

1
0.8
0.6
0.4
0.2
0
15 20 25 30 35 40 45 50 55 60 65
Frequency (Hz)

Figure 5.8: ISPE test article global skyline functions.

The phase FRF skyline offers the clearest indication of the presence of many modes in the
15–65 Hz frequency band. While the skylines indicate the presence of “many modes,” further
investigation of local response FRFs will provide insight into the presence of additional localized
modes (a difficult, slow task for a structural test article that is mapped by 265 response channels
and 4 excitations, that is 1060 individual FRFs!). Other strategies for discerning the presence
of “many modes” as well as closely spaced modes have been developed and successfully applied
by leaders in the experimental structural dynamics community [3].
92 5. EXPERIMENTAL MODAL ANALYSIS
5.1.6 CLOSURE
Several intuitive experimental modal analysis tools have been described. SDOF frequency re-
sponse functions displayed in three ways, namely: (1) magnitude and phase components vs. fre-
quency; (2) real and imaginary components vs. frequency (also called co- and quad- functions);
and (3) polar real vs. imaginary components vs. frequency. All three display types provide intu-
itive, well-known means for estimation of modal frequency and damping parameters. Review
of FRFs associated with individual FRFs for actual MDOF systems offer some insights for es-
timation of modal frequency and damping parameters when modes are well separated (not the
case for the ISPE test article and other shell-type structures).
Global skyline functions offer some insight into the presence of multiple modes in a des-
ignated frequency band. However, such functions do not necessarily highlight the presence of
localized modes. A more comprehensive approach involving review of many or all measured
FRF response functions is a tedious, time-consuming endeavor.

5.2 PART 2: THE SIMULTANEOUS FREQUENCY DOMAIN

METHOD

5.2.1 INTRODUCTION
Difficulties encountered by NASA/MSFC on the Integrated Spacecraft Payload Element
(ISPE) modal survey in the fall of 2016 bring an important challenge to the forefront. Specifi-
cally, which estimated test modes are “authentic,” and which modes are due to “noise” associated
with measured FRFs? The present discussion on experimental modal analysis (EMA) focuses
on mathematical isolation of individual estimated test mode FRFs in a manner that is similar to
the concept developed by Mayes and Klenke [4]. While the presently discussed EMA approach
ought to be quite independent of the investigator’s choice of experimental modal analysis algo-
rithm, the results herein apply to methods that explicitly estimate the tested system’s state-space
plant matrix such as the Simultaneous Frequency Domain (SFD) method [5–7].
The latest version of SFD (SFD-2018) employs mathematical operations aimed at isolat-
ing individual candidate experimental modes without direct reliance on information associated
with the subject system’s TAM) mass matrix. The key to mathematical and visual isolation of
individual modes from measured data is the left-hand eigenvector. Virtually all modern experi-
mental modal analysis techniques produce estimates of (right-hand) eigenvectors and eigenval-
ues (modal frequency and damping). While techniques for estimation of left-hand eigenvec-
tors are well-known (e.g., the “real mode transpose times TAM mass matrix product” and the
Moore–Penrose pseudo-inverse [8]), they have been judged inadequate. The purest approach to
estimation of left-hand eigenvectors is a consequence specifically those techniques that estimate
the measured system’s plant or effective dynamic system matrix, such as SFD. Since a complete
set of raw experimental modes are identified consistent with the order of the estimated plant, the
 5.2. PART 2: THE SIMULTANEOUS FREQUENCY DOMAIN METHOD 93
left-hand eigenvectors are calculated exactly from the inverse of the complete, raw right-hand
eigenvector set.

5.2.2 EFFECTIVE DYNAMIC SYSTEMS

The SFD method [5], introduced in 1981, has undergone substantial revision and refinement
since that time [6, 7], primarily by this writer and principals at The Aerospace Corporation.
SFD implicitly assumes that FRFs associated with a series of “N ” excitations may be described
in terms of a transformation described by

  h i
UR 1 .f / UR 2 .f / : : : UR N .f / D ŒV  R1 .f / R2 .f / : : : RN .f / : (5.13)

By performing singular value decomposition (SVD) analysis [9] of the FRF collec-
tion,
h a dominant set of real generalized
i trial vectors, ŒV , and complex generalized FRFs,
R1 .f / R2 .f / : : : RN .f / , is obtained. Unit normalization of the SVD trial vectors is
expressed as

ŒV T ŒV  D ŒI  : (5.14)

Theoretically, the generalized FRF array describes the following dynamic system equations
associated with the individual applied forces

h i  h i    
R / C BQ .f
.f P / C KQ Œ.f / D €Q ŒF .f / ;
       
where BQ D M 1 ŒB ; KQ D M 1
ŒK ;
   
and €Q D M 1 Œ€ : (5.15)
     
The real, effective dynamic system matrices, BQ , KQ , and €Q , are estimated by linear
least-squares analysis [10].
Estimation of experimental modal parameters is performed by complex eigenvalue anal-
ysis of the state variable form of the effective dynamic system,
( )     
R BQ KQ P €Q
D C fF g ; (5.16)
P I 0  0

which is of the general type

   
fg
P D A fg C € fF g : (5.17)
94 5. EXPERIMENTAL MODAL ANALYSIS
Complex eigenvalue analysis of the effective dynamic system produces the following re-
sults:
      1
.a/ fg D ˆ fqg ; where the “left-handed” eigenvectors are ˆL D ˆ ;
        
.b/ ˆL
ˆ D ŒI  ; ˆL
A ˆ D Œ (complex eigenvalues);
    (5.18)
.c/ ˆL
€ D Œ  (modal gains); and
.d / qPj j qj D . /j ŒF .f / (frequency response of individual modes):

Recovery of experimental modes in terms of the physical DOFs involves back transfor-
mation employing the trial vector matrix, ŒV , specifically,

    
Œˆ D ŒV  ˆ ; ŒˆL  D ˆL V T ; ŒOR D ŒˆL  Œˆ  ŒI  : (5.19)

5.2.3 THE SFD METHOD PRIOR TO 2018

Estimation of the effective dynamic system with the SFD method (and more generally any
method that performs similar system “plant” estimation operations) will pick up spurious “noise”
degrees of freedom and associated spurious modes. Over the years since 1981, the writer has
employed a heuristic practice in versions of SFD algorithms that select “authentic” modes from
the complete set, which is estimated in selected frequency bands. The heuristic criteria include:
(1) elimination of modes having negative damping; (2) modes with very low modal gain; and
(3) other modes that appear spurious from any number of physical/experience based consider-
ations. Prior to 2018, the SFD method (this writer’s version) did not make use of the complex
modes associated with the effective dynamic system (Equations (5.15)–(5.19)).
The theoretical relationship between FRFs and modal parameters (assuming that modal
vectors are real) is

ŒH.f / D Œˆ
Œh.f / ; (5.20)

where Œˆ is the unknown real modal matrix and Œh.f / is the SDOF acceleration FRF matrix.
The terms of Œh.f / are defined as

.f =fn /2
hn .f / D ; (5.21)
.1 C 2in .f =fn / .f =fn /2 /
where the fn and n are the modal frequency and damping associated with the particular exper-
imental mode. Since the modal SDOF acceleration matrix is completely known, the real modal
matrix is obtained by linear least squares analysis. At the experimental modal analyst’s discretion
(highly recommended), low- and/or high-frequency residual modal frequencies may be added
to the set of identified eigenvalues (the low-frequency residual FRF has a frequency close to “0”
 5.2. PART 2: THE SIMULTANEOUS FREQUENCY DOMAIN METHOD 95
and a user-selected damping value, e.g., n D :01, and the high-frequency residual has a fre-
quency substantially higher than the highest experimental modal frequency and a user-selected
damping value, e.g., n D :01) to enhance accuracy of modal vector estimates.
The theoretical relationship between FRFs and modal parameters (assuming that modal
vectors are complex) is

N 
X 
ˆn ˆn
ŒH.f / D .2
f /
C  ; (5.22)
nD1
n i.2
f / n i.2
f /

where ˆn is the unknown nth complex modal residue vector and n is the nth (positive imaginary
part) complex eigenvalue. The complex modal residue vectors are proportional to the complex
system modes. Complex eigenvalues (when critical damping ratio, J , is less than 1.0) are

n D n !n C i! n : (5.23)

Since the complex eigenvalues are completely known, the complex modal residue vectors
(proportional to the complex modal vectors) are obtained by linear least squares analysis. As for
the case of real modes, low and/or high frequency residual modal frequencies may be added to
the set of identified eigenvalues to enhance accuracy of modal vector estimates.
Regardless of whether the user chooses to estimate real or complex modal vectors, recon-
structed FRFs calculated from identified modal parameters serve as a means for a quality check
on the overall (pre-2018) SFD process.

5.2.4 ILLUSTRATIVE EXAMPLE: ISS P5 MODAL TEST

Due to problematic nonlinear behavior and variability of modal characteristics of the ISS P5
test article, 17 separate measured time tistory data sets were recorded and evaluated employing
the preliminary and detailed data analysis techniques described in Chapter 4. An overview of
the measured data sets is provided in Table 5.1.
The three highlighted sets were selected by NASA and the test team for detailed exper-
imental data analysis, while all measured data sets were employed for the nonlinear-sensitivity
investigation.
Detailed experimental modal analysis of measured data set TSS2, employing the (pre-
2018) SFD method, resulted in estimation of 23 test article modes in the 0–63 Hz frequency
band as summarized in Table 5.2. Note the first 10 modes (shaded in yellow) were selected for
subsequent test-analysis correlation and model reconciliation analyses.
Integrity of the 23 estimated ISS P5 modes for measured data set TSS2 was established
by comparison of MI/MO and reconstructed FRFs calculated from identified modal parameters
as illustrated in Figure 5.9.
96 5. EXPERIMENTAL MODAL ANALYSIS
Table 5.1: ISS P5 measured data sets

Data Set Duration, Averages Excitation Type Excitation Level, Shakers

TSS1 13.3 minutes, 100 averages Burst random 10 lb RMS, 3 shakers
TSS2 10 minutes, 75 averages Continuous Random 10 lb RMS, 3 shakers
TSS3 13.3 minutes, 100 averages Burst random 10 lb RMS, shakers 1&2
TSS4 10 minutes, 75 averages Continuous Random 10 lb RMS, shakers 1&2
TSS5 13.3 minutes, 100 averages Burst random 10 lb RMS, shakers 1&2
TSS6 13.3 minutes, 100 averages Burst random 10 lb RMS, 3 shakers
TSS7 13.3 minutes, 100 averages Burst random 25 lb RMS, 3 shakers
TSS8 10 minutes, 75 averages Continuous Random 25 lb RMS, 3 shakers
TSS9 “long,” max overlap Sine Sweep Up-Down 3 lb up-10 lb down, shakers 1
TSS10 “long,” max overlap Sine Sweep Up-Down 3 lb up-10 lb down, shakers 2
TSS11 “long,” max overlap Sine Sweep Up-Down 3 lb up-10 lb down, shakers 3
TSS12 6.7 minutes, max overlap Sine Sweep Up 10 lb, shaker 3
TSS13 6.7 minutes, max overlap Sine Sweep Up 10 lb, shaker 2
TSS14 6.7 minutes, max overlap Sine Sweep Up 10 lb, shaker 1
TSS15 6.7 minutes, 50 averages Burst random 10 lb RMS, 3 shakers
TSS16 13.3 minutes, 100 averages Burst random 10 lb RMS, 3 shakers
TSS17 10 minutes, 75 averages Continuous Random 10 lb RMS, 3 shakers

Note that the MI/MO estimated FRF is indicated by the blue dots, the reconstructed FRF
is indicated by the green curves, and the frequencies associated with the 23 estimated modes are
indicated by the red circles.
Following established NASA and USAF Space Command practices [11, 12], orthogo-
nality of the 23 ISS P5 (data set TSS2) experimental modes, expressed in percentage units) is
summarized in Figure 5.10 and Table 5.3, respectively.
Review of the contents of Table 5.3 indicates that the NASA orthogonality criterion
(goal) [11] was achieved for most of the 23 test modes, including modes 11–23 which had
non-negligible kinetic energy content associated with the test fixture.
 5.2. PART 2: THE SIMULTANEOUS FREQUENCY DOMAIN METHOD 97

Table 5.2: ISS P5 estimated modes for data set TSS2

Directional KE (%) Component KE (%)

Mode Freq (Hz) Zeta (%) X Y Z Truss Grapple Fixture
1 16.94 2.13 73 27 1 85 15 0
2 17.58 1.08 24 73 2 91 9 0
3 25.19 1.08 68 4 28 77 23 0
4 28.44 0.80 48 47 5 94 6 0
5 31.12 1.25 72 8 20 50 50 0
6 32.6 0.84 18 17 65 61 39 0
7 33.66 1.55 15 11 74 54 46 0
8 35.19 0.72 33 60 7 95 5 0
9 36.39 0.40 81 11 8 98 2 0
10 38.38 0.32 4 29 67 99 1 0
11 42.86 1.44 13 34 53 68 31 1
12 45.16 1.49 25 36 39 78 22 0
13 46.95 0.79 9 18 74 82 10 8
14 48.4 1.04 40 17 43 69 24 7
15 49.1 2.30 34 45 22 53 5 42
16 51.69 1.15 19 14 67 60 34 6
17 54.06 1.54 22 43 35 70 21 9
18 54.45 0.83 65 25 10 95 3 2
19 55.3 0.67 45 43 12 98 2 1
20 58.1 0.84 71 16 13 97 2 0
21 60.15 1.55 9 17 74 81 12 7
22 61.78 0.04 1 96 4 3 0 97
23 62.84 0.65 16 8 76 25 72 3
98 5. EXPERIMENTAL MODAL ANALYSIS

Global Mode Fit…Input=F-1, Output=A005-2

10

Magnitude
10-3

10-4
20 30 40 50 60
Hz

100
Phase (Deg)

-100

20 30 40 50 60
Hz

Figure 5.9: Typical ISS P5 MI/MO FRF reconstruction based on pre-2018 SFD methodology.

100

80
Orthogonality

60

40

20

0
20
20
10 15
10
5
0 0
Mode Mode

Figure 5.10: ISS P5 (data set TSS2) experimental mode orthogonality.

 5.2. PART 2: THE SIMULTANEOUS FREQUENCY DOMAIN METHOD 99
Table 5.3: ISS P5 (data set TSS2) experimental mode orthogonality
Mode Freq (Hz) Orthogonality (%)

5.2.5 SFD 2018: A FRESH LOOK AT EXPERIMENTAL MODAL ANALYSIS

The initial point of departure from pre-2018 SFD practice is estimation of an effective dynamic
system over the entire frequency band of interest (rather than selected sub-frequency bands).
In order to achieve a satisfactory estimation for the effective dynamic system, the “tolerance”
factor (") employed in the SVD process is set to a sufficiently low value .10 5 /; in previous
sub-frequency band SFD calculations, the SVD “tolerance” factor was set to a value of 10 2 .
Computation of effective dynamic system modal parameters, from the first-order system de-
scribed in Equation (5.17), yields complex modes with eigenvalues having negative and positive
imaginary parts. The first level of mode down-selection is to eliminate all modal eigenvalues
and eigenvectors that are outside the positive frequency band of interest; for the ISPE modal
test, there are 106 complex eigenvalues in the 15–65 Hz frequency band. A vital component of
the mode down-selection process is selection of left-hand eigenvectors. ŒˆL  that correspond
to their Œˆ  counterparts; this circumvents issues associated with more involved procedures for
computation of a truncated left-hand eigenvector set.
Two computational procedures estimate uncoupled experimental modal FRFs. The first
method computes the exact modal solution from the estimated modal parameters of Equa-
tion (5.18)d. Specifically,
 


i 2f

hj A
D qPj A
D j : (5.24)
i2f j
The second method estimates uncoupled experimental modal FRFs from linear combina-
tions of the generalized FRFs (see Equations (5.18)a and (5.18)b) as follows:


 
hj E
D qPj E
D ˆL j
./
P : (5.25)
100 5. EXPERIMENTAL MODAL ANALYSIS
Verification and validation of any candidate estimated experimental mode is now to be
judged on the basis of: (a) graphical displays of the modal FRFs and (b) a new modal coherence
metric, which is defined as
ˇ    ˇˇ2
ˇ
ˇ hj .f / A
hj .f / E ˇ
COHj D ˇ    ˇˇ ˇˇ    ˇˇ : (5.26)
ˇ
h
ˇ j .f / A

hj .f /

Aˇ ˇ j
h .f / E

hj .f / Eˇ

5.2.6 ILLUSTRATIVE EXAMPLE: ISPE MODAL TEST

The Integrated Spacecraft Payload Element (ISPE), introduced in Chapter 4, was the subject
of modal testing at NASA/MSFC in the fall of 2016. Measured FRF data was quite extensive,
and the NASA team had a great deal of difficulty in estimation of modal parameters in the 0–
65 Hz frequency band due in part to close modal spacing and significant modal density. These
challenges led to development of SFD-2018. For completeness, results of pre-2018 experimental
modal analysis methodology and measured FRF reconstruction (similar to the graphic provided
in Figure 5.9 for the ISS P5 test article) are illustrated in Figure 5.11.

File: ISPETESTDFRF Global Real Mode Fit…Input=2302113X-, Output=2302113X+

10-1
Magnitude

10-2

10-3 20 25 30 35 40 45 50 55 60 65
Hz
150
Measured FRF
100 Least Square Fit FRF
Phase (Deg)

50 Estimated Modal Frequency

0
-50
-100
-150
20 25 30 35 40 45 50 55 60 65
Hz

Figure 5.11: Typical ISPE test article MI/MO FRF reconstruction based on pre-2018 SFD
methodology.

Establishment of a selected set of experimental modes employing uncoupled SFD-2018

modal FRF estimates, computed using Equations (5.24) (red for reconstructed FRFs) and
Equation (5.25) (blue for experimental FRFs) for several candidate modes is illustrated in Fig-
 5.2. PART 2: THE SIMULTANEOUS FREQUENCY DOMAIN METHOD 101
ures 5.12–5.14 include FRF magnitude, magnitude and phase, polar, and real and imaginary
parts.

FRF Display: Mode 1, Input 1, Coh = 99% 20-Jul-2018 12:43 PM

Magnitude
h(f)
10-2 hmodal(f)

20 25 30 35 40 45 50 55 60
Frequency (Hz)

0.03
0.02
Magnitude

0.02
0

Real
0
0.01 -0.005

Imaginary
-0.01 -0.02
-0.015
15.5 16 16.5 17 17.5 18 18.5 15.5 16 16.5 17 17.5 18 18.5
-0.02
Phase (Deg)

-0.025 0.02

Imaginary
100
-0.03
0 0
-0.01 0 0.01
-100 Real -0.02
15.5 16 16.5 17 17.5 18 18.5 15.5 16 16.5 17 17.5 18 18.5
Frequency (Hz) Frequency (Hz)

Figure 5.12: EMA for SFD candidate mode 1, excitation 1 (99% coherence).

It is clear in the above three figures that candidate modes 1 and 3 appear valid based on
close agreement of the two types of uncoupled modal FRF estimates. In contrast, candidate
mode 6 is clearly “spurious.”
Table 5.4 summarizes additional information related to the process of selection of 63
acceptable experimental modes from the 106 candidate modes (the first 30 candidate modes
and associated modal coherences are illustrated).
Table 5.4 provides a clear demonstration of the utility of the newly introduced EMA
metric. In particular, the modal coherence metric, defined in Equation (5.26), for which a value
of 85% or greater is assumed to indicate a validly estimated mode.
Returning to the issue of compliance with the NASA-STD-5002 test mode orthogo-
nality criterion, it is quite informative to compare orthogonality matrices computed by: (1) the
conventional method based on weighted orthogonality of real modes (in this case the real part
of the ISPE test modes) with respect to the TAM mass matrix and (2) exact complex test mode
orthogonality assured by left-hand eigenvectors (modes) based on Equation (5.18)b. Results for
the 63 ISPE test modes are depicted in Figure 5.15.
While results of orthogonality based on the NASA-STD-5002 criterion indicate good to
excellent orthogonality for the first 34 modes, the alternative based on the exact mathematical
property of complex SFD-2018 estimated modes opens up the opportunity for automatic sat-
isfaction of the orthogonality criterion without reliance on the (potentially flawed) FEM-based
102 5. EXPERIMENTAL MODAL ANALYSIS

FRF Display: Mode 3, Input 4, Coh = 97% 20-Jul-2018 12:46 PM

Magnitude
h(f)
10-2 hmodal(f)

10-4
20 25 30 35 40 45 50 55 60
Frequency (Hz)

10-1 0.1
Magnitude

Real
0
0
10-2

Imaginary
-0.05 -0.1

17.5 18 18.5 19 19.5 17.5 18 18.5 19 19.5

-0.1
Phase (Deg)

0.1

Imaginary
100 -0.15
0 0
-0.05 0 0.05 0.1
-100 Real -0.1

17.5 18 18.5 19 19.5 17.5 18 18.5 19 19.5

Frequency (Hz) Frequency (Hz)

Figure 5.13: EMA for SFD candidate mode 3, excitation 4 (97% coherence).

FRF Display: Mode 6, Input 4, Coh = 4% 20-Jul-2018 12:52 PM

Magnitude

h(f)
hmodal(f)

10-5
20 25 30 35 40 45 50 55 60
Frequency (Hz)

0.05
Magnitude

10-2

Real
0.04
0
0.03
10-3
Imaginary

0.02
-0.05
0.01
18.5 19 19.5 20 20.5 18.5 19 19.5 20 20.5
0
Phase (Deg)

-0.01 0.05
Imaginary

100
-0.02
0 0
0 0.02 0.04 0.06
-100 Real -0.05
18.5 19 19.5 20 20.5 18.5 19 19.5 20 20.5
Frequency (Hz) Frequency (Hz)

Figure 5.14: EMA for SFD candidate mode 6, excitation 4 (4% coherence).
 5.2. PART 2: THE SIMULTANEOUS FREQUENCY DOMAIN METHOD 103
Table 5.4: EMV modal selection criterion (COHj  85%) for candidate modes 1–30

Candidate Eigenvalue Modal Coherences

Mode Freq (Hz) Zeta (%) Excitation 1 Excitation 2 Excitation 3 Excitation 4
1 16.74 2.32 99.15 98.92 98.31 88.97
2 17.03 2.25 72.99 97.67 98.62 98.75
3 18.12 0.12 70.97 69.87 73.51 97.16
4 18.16 0.12 97.08 78.26 85.15 42.25
5 18.79 0.12 94.47 87.10 97.41 95.45
6 19.40 0.05 2.80 3.24 1.91 3.54
7 20.08 3.22 93.57 76.25 94.66 92.82
8 20.31 3.07 95.26 94.92 90.67 88.32
9 20.96 0.10 86.89 96.19 93.10 91.52
10 21.01 0.09 96.43 91.40 89.60 88.14
11 23.23 2.14 39.31 23.98 4.12 34.13
12 24.05 0.02 27.91 29.69 18.24 40.46
13 24.07 0.48 83.55 99.34 98.94 99.46
14 24.26 0.45 99.71 99.46 99.33 34.16
15 24.95 1.81 36.78 44.96 54.73 33.49
16 25.82 0.15 99.69 96.55 95.00 95.85
17 25.89 0.12 78.20 96.91 89.74 98.99
18 27.70 1.40 98.84 95.49 95.59 98.57
19 27.79 1.69 96.85 98.79 98.58 96.55
20 28.91 -1.44 0.76 0.36 1.09 2.55
21 30.15 0.07 87.03 88.28 69.70 86.86
22 32.52 0.95 8.29 81.46 91.37 97.65
23 32.52 0.11 28.00 27.15 3.47 25.12
24 33.42 0.16 87.29 95.81 95.64 91.84
25 33.65 1.56 66.09 96.78 93.51 15.07
26 34.61 2.28 90.14 79.40 65.22 93.04
27 35.53 1.17 83.46 80.42 64.20 88.29
28 35.76 0.86 72.07 72.16 62.79 22.99
29 36.24 0.15 95.91 96.93 80.51 93.78
30 36.46 0.26 99.76 99.38 99.68 99.75
104 5. EXPERIMENTAL MODAL ANALYSIS
ISPE Test Modal Orthogonality ISPE Test Modal Orthogonality
(NASA STD-5002 Method) (SFD-2018 Complex Modes)
100 100
90 90
10 10
80 80
20 70 20 70
60 60

Mode

Mode
30 30
50 50
40 40 40 40
30 30
50 20 50 20
10 10
60 60
10 20 30 40 50 60 10 20 30 40 50 60
Mode Mode

Figure 5.15: ISPE test mode orthogonality estimates.

TAM mass matrix. Further discussion on the merits associated with SFD-2018 is included in
Chapter 6.

5.2.7 CLOSURE
Experimental modal analysis is a mature discipline in the structural dynamics community, which
is as much an “art” as it is a “science.” Modern procedures for estimation of modal parameters
from measured data are highly automated; however, applications involving complicated struc-
tural systems and/or systems with closely spaced, parametrically sensitive modes require the test
engineer’s experience and judgment (“art”) to discern the difference between authentic and spu-
rious (“junk” or “noise”) system modes. A prevailing metric for experimental modal data valida-
tion is the orthogonality check, which relies on a model-based (TAM) mass matrix. In addition,
reconstructive synthesis of measured FRF data is another widely used strategy for experimental
mode validation. The present EMA study employs mathematical operations aimed at isolating
individual candidate experimental modes without reliance on a TAM mass matrix.
The key to mathematical and visual isolation of individual modes from measured data is
the left-hand eigenvector. The most effective approach to determination of left-hand eigenvec-
tors stems from employment of techniques that estimate the measured system’s plant or effective
dynamic system matrix. Since a complete set of (authentic and “noise”) system modes are es-
timated for the plant, left-hand eigenvectors are determined from the inverse of the complete
right-hand eigenvector set.
The following metrics provide a systematic basis for EMA.
1. The estimated SDOF modal FRF, formed by the product of a single estimated left-hand
eigenvector and FRF matrix, is plotted in terms of real and imaginary components vs.
 5.3. REFERENCES 105
frequency, magnitude, and phase components vs. frequency, and polar real vs. imaginary
components. Authenticity of an estimated mode is then judged on the basis of quality of
the plots.

2. The SDOF modal FRF is also formed from exact mathematical solution of the estimated
effective dynamic system. Graphical comparison of this result with the above left-handed
product information offers further means of authentic vs. “junk” mode discrimination.

3. Finally, a coherence metric based on comparison of the results of “1” and “2” provides a
0–100% figure of merit for estimated experimental modes.

5.3 REFERENCES
[1] R. E. D., Bishop and G. M. L. Gladwell, An investigation into the theory of resonance
testing, Philosophical Transactions, Royal Society of London, Series A, 225(A-1055):241–
280, 1963. DOI: 10.1098/rsta.1963.0004. 85

[2] D. L. Brown and R. J. Allemang, The modern era of experimental modal analysis, Sound
and Vibration Magazine, January 2007. 85

[3] A. G. Piersol and T. L. Paez, Eds., Harris’ Shock and Vibration Handbook, 6th ed.,
McGraw-Hill, 2010. 85, 91

[4] R. Mayes and S. Klenke, The SMAC modal parameter extraction package, IMAC XVII,
1999. 92

[5] R. N. Coppolino, A simultaneous frequency domain technique for estimation of modal

parameters from measured data, SAE Paper 811046, 1981. DOI: 10.4271/811046. 92,
93

[6] R. N. Coppolino and R. C. Stroud, A global technique for estimation of modal param-
eters from measured data, SAE Paper 851926, 1985. DOI: 10.4271/851926. 93

[7] R. N. Coppolino, Efficient and enhanced options for experimental mode identification,
IMAC XXI, 2003. 92, 93

[8] R. Penrose, A generalized inverse for matrices, Proc. of the Cambridge Philosophical Society,
51:406–13, 1955. DOI: 10.1017/s0305004100030401. 92

[9] G. H. Golub and C. Reinsch, Singular value decomposition and least squares solutions,
Numerische Mathematik, 14:403–420, 1970. DOI: 10.1007/bf02163027. 93

[10] K. Miller, Complex linear least squares, SIAM Review, 15(4):706–726, 1973. DOI:
10.1137/1015094. 93
106 5. EXPERIMENTAL MODAL ANALYSIS
[11] Load analysis of spacecraft and payloads, NASA-STD-5002, 1996. 96
[12] Independent structural loads analysis, U.S. Air Force Space Command, SMC-S-004, 2008.
96
 107

CHAPTER 6

Systematic Test Analysis

Correlation
6.1 PART 1: CONVENTIONAL MASS WEIGHTED
CORRELATION OF TEST AND FEM MODES
6.1.1 INTRODUCTION
Developments in structural dynamic modeling and modal testing were strongly motivated by air-
craft structural failures as early as the first decade of the 20th century. The space race, dating back
to the 1950s, and its early failures resulted development of strict government modal test-analysis
correlation standards [1, 2] based on mass-weighted orthogonality and cross-orthogonality met-
rics. The present discussion focuses on (1) review of mass weighted test mode orthogonality and
test-analysis cross-orthogonality metrics and (2) introduction of a modal coherence metric. All
of the noted metrics are the consequence of a rigorous weighted least-squares formulation [3].
Employment of the conventional mass weighted correlation metrics is demonstrated on modal
data associated with two modal test projects.

6.1.2 DERIVATION OF MASS WEIGHTED TEST-ANALYSIS

CORRELATION METRICS
The relationship between target test modes, Œˆ t , and their analytical counterparts, Œˆa , is de-
scribed by the transformation

Œˆt  D Œˆa  ŒCOR C ŒR ; (6.1)

where ŒCOR is the cross-orthogonality matrix and ŒR is the residual error matrix. Employ-
ing the TAM mass matrix, ŒMaa , as a weighting matrix, the least squares solution for cross-
orthogonality is

1
 T 
ŒCOR D ŒORa  ˆa Maa ˆt ; (6.2)
where the analysis mode orthogonality matrix is
 
ŒORa  D ˆTa Maa ˆa  ŒI  ; (6.3)
and the residual error matrix is orthogonal to the test modes, i.e.,
108 6. SYSTEMATIC TEST ANALYSIS CORRELATION

Œˆa T ŒMaa  ŒR  Œ0 : (6.4)

The modal coherence matrix, ŒCOH, is defined based on manipulation of Equation (6.1)
(taking into account the results stated in Equations (6.2)–(6.4)) resulting in

   
ŒCOH D ŒI  ŒORt  1=2 RT MR ŒORt  1=2
 
D ŒORt  1=2 CORt ORa COR ŒORt  1=2 ; (6.5)

where the test mode orthogonality matrix is

 
ŒORt  D ˆTt Maa ˆt : (6.6)
The modal coherence matrix, ŒCOH, provides a metric of the ability of mathematical
model modes, Œˆa , to reconstruct the measured modes, Œˆ t ; in particular, if the diagonal term,
jCOH nn j, associated with measured mode “n” is unity, the measured mode is a perfect linear
combination of the mathematical model modes, Œˆa .
Having already discussed U.S. government criteria for test mode orthogonality, it is in-
formative to note corresponding criteria for test-analysis correlation. NASA-STD-5002 [1]
states “Agreement between test and analysis natural frequencies shall, as a goal, be within 5%
for the significant modes (aka target modes)…Mode shape comparisons shall be required via
cross-orthogonality checks using the test modes Œˆ t , the analytical modes Œˆa , and the ana-
lytical mass matrix ŒMaa . The cross-orthogonality matrix is computed as Œˆ t T ŒMaa Œˆa . As
a goal, the absolute value of the cross-orthogonality between corresponding test and analyti-
cal mode shapes should be greater than 0.9; and all other terms of the matrix should be less
than 0.1 for all significant modes.” It should be noted that the NASA-STD-5002 [1] defini-
tion of cross-orthogonality differs from Equation (6.2), which allows for imperfect TAM mode
orthogonality.
It is also informative to note the somewhat stricter U.S. Air Force Space Command cri-
terion [2] for test-analysis correlation, which states, “As a goal, the analytical model frequencies
should be within three percent of the measured values, and the cross-orthogonality between
the analytical and measured modes, each set normalized to yield a unit generalized mass ma-
trix, should yield values equal to or greater than 0.95 on the diagonal, and equal to or less than
0.10 on the off-diagonal of the cross-orthogonality matrix. Any modeling adjustments/changes
made to achieve the above-stated criteria must be consistent with the actual hardware and its
drawings.”

6.1.3 ILLUSTRATIVE EXAMPLE: ISS P5 MODAL TEST

At the time of the ISS P5 modal test, the modal coherence matrix was not employed as part of the
modal test-analysis correlation process. It is quite informative nevertheless to review results of
 6.1. PART 1: CONVENTIONAL MASS WEIGHTED CORRELATION 109
cross-orthogonality analysis, which was employed for correlation of (pre-2018 SFD) test modes
and FEM analytical modes. As noted in Chapters 3–5, the ISS P5 modal test suffered from
challenges associated with localized nonlinearity and non-repeatability of experimental modal
data. It is informative to review cross-orthogonality results for the TSS2 data set associated
with the ISS P5 pre-test and revised post-test models. A more extensive discussion of the ISS
P5 test-analysis reconciliation process (model updating) is provided in Chapter 7. A summary
of pre-test ISS P5 modal frequencies, TSS2 data set modal frequencies and the pre-test FEM
cross-orthogonality matrix (for the first 10 modes) is provided in Table 6.1.

Table 6.1: ISS P5 data set TSS2 test to pre-test FEM modal frequencies and cross-orthogonality
matrix

TSS2 Modes
Mode
Freq (Hz) |Cross-Orthogonality| (%)
Pre-Test FEM

The TSS2 data set pre-test modal frequency correspondence and cross-orthogonality ma-
trix clearly do not satisfy NASA-STD-5002 [1] test-analysis correlation criteria (goals). Model
updating operations (to be discussed in Chapter 7) substantially improved the test-analysis cor-
relation situation, as summarized in Table 6.2.

Table 6.2: ISS P5 data set TSS2 test to updated FEM modal frequencies and cross-orthogonality
matrix

TSS2 Modes
Mode
Freq (Hz)
|Cross-Orthogonality| (%)
Updated FEM
110 6. SYSTEMATIC TEST ANALYSIS CORRELATION
The level of compliance of TSS2 modal test data and the updated P5 FEM model with
NASA-STD-5002 test-analysis correlation goals is summarized in Table 6.3.

Table 6.3: ISS P5 data set TSS2 test to updated FEM NASA-STD-5002 compliance status

Pre-Test
TSS2 Post-Test FEM
Mode FEM Comments
Freq (Hz) Freq (Hz) Freq (Hz) ∆F(%) COR (%)
1 18.67 16.94 16.48 -2.72 98
2 19.31 17.58 18.08 2.84 97 Satisfies NASA-STD-5002 Modal
3 24.88 25.19 25.66 1.87 96 Frequency and On-Diagonal Cross-
4 28.68 28.44 28.65 0.74 92 Orthogonality Goals
5 29.37 31.12 30.74 -1.22 94
6 30.07 32.6 32.28 -0.98 83
Frequency Band Associated with Reduced
7 34.01 33.66 33.32 -1.01 82
MI/MO Coherence (strong nonlinearity)
8 34.65 35.19 34.64 -1.56 85
9 35.22 36.39 36.58 0.52 90 Satisfies NASA-STD-5002 Modal
Frequency and On-Diagonal Cross-
10 35.61 38.38 36.98 -3.65 97
Orthogonality Goals

6.1.4 ILLUSTRATIVE EXAMPLE: ISPE MODAL TEST

ISPE modal test data, described in Chapter 5, and ISPE FEM modes are now compared em-
ploying the conventional mass weighted correlation procedure. Employing the real part of the
ISPE SFD-2018 complex modes and ISPE FEM modes, NASA-STD-5002 test mode orthog-
onality (Equation (6.3)), cross-orthogonality (Equation (6.2)), and modal coherence (Equa-
tion (6.5)) matrices are provided in Figure 6.1.
While the cross-orthogonality matrix indicates poor test-analysis modal correlation for
many of the 63 SFD-2018 estimated modes, the modal coherence matrix indicates that approx-
imately 50 of the SFD-2018 modes are linear combinations of the 75 lowest-frequency FEM
modes. In Chapter 7, test-analysis reconciliation methodology will demonstrate the potential
for drastic improvement of ISPE test-analysis correlation results.

6.1.5 CLOSURE
Systematic correlation of FEM based theoretical predictions and experimental modal analysis
data, from the viewpoint of a widely accepted U.S. aerospace community practice, relies on
weighted least squares based correlation metrics and standardized NASA and USAF criteria.
The fundamental assumption inherent in the practice is that test article dynamics may be re-
 6.1. PART 1: CONVENTIONAL MASS WEIGHTED CORRELATION 111
ISPE Test Modal Orthogonality ISPE TAM to Test Modal Cross-Orthogonality
(NASA STD-5002 Method) (NASA STD-5002 Method)
100
90
90 10
10 80
80
20 70
20 70
30 60

TAM Mode
60
30 50
Mode
50 40
40 40
40
50 30
30
50 20 60 20
10 70 10
60
10 20 30 40 50 60 10 20 30 40 50 60
Mode Test Mode
ISPE TAM to Test Modal Coherence (Real Test Modes)
90
10
80
20 70
60
Test Mode

30 50

40 40
30
50 20
10
60
10 20 30 40 50 60
Test Mode

Figure 6.1: ISPE conventional mass weighted test-analysis correlation matrices.

liably approximated in terms of real normal modes, in spite of the facts that (a) a test article
may exhibit nonlinear behavior and (b) experimental modes are generally complex. In addition,
(c) the correlation metrics are dependent on assumed validity of the TAM mass matrix derived
from the theoretical FEM. Due to these inherent assumptions, which at times are questionable,
standardized NASA and USAF criteria are specified as goals rather than requirements.
Two test projects serve as illustrative examples for application of the widely accepted test-
analysis correlation practice. The ISS P5 modal test, conducted in 2001, demonstrates perfor-
mance of the weighted least squares correlation metrics in the presence of nonlinear test article
behavior. The more recent ISPE modal test, conducted in 2016, demonstrates performance of
the weighted least squares correlation metrics in the presence of complex test modes associated
with close modal frequency spacing.
112 6. SYSTEMATIC TEST ANALYSIS CORRELATION
6.2 PART 2: CORRELATION OF TEST AND FEM MODES
USING LEFT-HAND EIGENVECTORS
6.2.1 INTRODUCTION
Over many years, members of the experimental modal analysis community have been chal-
lenged over the use of NASA and USAF Space Command [1, 2] modal orthogonality and
cross-orthogonality criteria for validation of experimental modal vectors and for assessment of
test-analysis correlation, respectively. At the heart of the challenge is the role played by the po-
tentially inaccurate TAM mass matrix, which is derived from a mathematical model. Recent
work that exploits left-hand eigenvectors, estimated by the SFD technique, provides a promis-
ing way out of the TAM mass matrix impasse. Modal orthogonality, defined as the product of
left- and right-handed experimental eigenvectors (real or complex) is mathematically an iden-
tity matrix. This guarantees that SFD-2018 estimated modes are always perfectly orthogonal.
An alternative cross-orthogonality definition, based on weighted complex linear least-squares
analysis [3], is evaluated and found to possess the desired property. Employment of (1) the left-
and right-handed experimental eigenvector based orthogonality matrix and (2) the weighted
complex linear least-squares based cross-orthogonality matrix represents a “game changer” that
potentially frees the experimental modal analysis community from the potentially inaccurate
TAM mass matrix.

6.2.2 MODAL TEST DATA RESULTING FROM ESTIMATED

STATE-SPACE MODELS
The SFD-2018 method (discussed in Chapter
h i5), employs SVD to estimate (real) trial vectors,
R
ŒV , and generalized acceleration FRFs, .f / , associated with measured physical DOF FRFs,
 
UR .f / , as follows:
  h i
UR .f / D ŒV  R1 .f / ; ŒV T ŒV  D ŒI  : (6.7)

Assuming the generalized DOF follow the effective dynamic system form,
h i  h i    
R / C BQ .f
.f P / C KQ Œ.f / D €Q ŒF .f / ; (6.8)

the effective state-space system modes are computed by solution of the eigenvalue problem,

( )   
R BQ KQ P  
D ; which is of the general type; fg
P D A fg : (6.9)
P I 0 

Complex eigenvalue analysis of the effective dynamic system produces the generalized
complex eigenvectors,
 6.2. PART 2: CORRELATION OF TEST AND FEM MODES 113

 
fg D ˆ fqg : (6.10)
The “left-hand” generalized eigenvectors are defined as
    1
ˆL D ˆ : (6.11)
Expansion of the generalized eigenvectors using the real trial vectors, ŒV , defines the test
eigenvectors (complex modes) in terms of the physical DOF, specifically,
    
ŒˆTEST D ŒV  ˆ ; ŒˆL TEST D ˆL V T : (6.12)

The state-space eigenvectors resulting from experimental modal analysis using SFD are
exactly orthogonal to one another, in particular,

ŒORTEST D ŒˆL TEST ŒˆTEST  ŒI  ; (6.13)

eliminating the need for satisfaction of NASA and/or USAF test mode weighted orthogonality
criteria! Moreover, orthogonality of state-space eigenvectors is not dependent on the TAM mass
matrix, ŒMTAM .

6.2.3 THEORETICAL SYSTEM MODES IN STATE-SPACE FORM

Modes associated with an undamped or damped theoretical model may be formulated in a man-
ner similar to the considerations employed in the previous discussion. If one substitutes the
Guyan [4] (or other suitable) reduction transformation for SVD trial vectors,

fUFEM g D Œ‰TAM  fUTAM g ; (6.14)

the reduced-order TAM modal dynamic equations become
˚ ˚
ŒMTAM  UR TAM C ŒBTAM  UP TAM C ŒKTAM  fUTAM g D f0g : (6.15)
Converting the above system to a state-space description, the following state-space eigen-
value problem is posed:

ŒATAM  ŒˆTAM  D ŒˆTAM  ŒTAM  ; (6.16)

 1 1

MTAM BTAM MTAM KTAM
ŒATAM  D : (6.17)
I 0
The (generally complex) right- and left-eigenvectors for the state-space TAM automati-
cally satisfy perfect orthogonality in the same manner as the test eigenvectors, specifically,

1
ŒˆL TAM D ŒˆTAM ; ŒˆL TAM ŒˆTAM D ŒI  : (6.18)
114 6. SYSTEMATIC TEST ANALYSIS CORRELATION
Results of both experimental and theoretical state-space modal analyses automatically
satisfy NASA-STD-5002 and SMC-S-004 orthogonality criteria [1, 2]. Most relevant is the
fact that experimental mode orthogonality does not require a TAM mass matrix.

6.2.4 CORRELATION OF EXPERIMENTAL AND THEORETICAL

MODES (COMPLEX LEAST SQUARES)
Systematic correlation of two sets of state-space, generally complex vectors (with identical DOF
designations) is defined in a manner analogous to the case of real vector sets, wherein the trans-
pose operator, is replaced by the complex conjugate transpose [3]. The correlation procedure is
formally defined by the following steps:

Step 1: Normalization of complex mode sets

1=2
ŒQTEST  D diagonal ˆTEST ˆTEST ;

1=2
ŒQTAM  D diagonal ˆTAM ˆTAM (6.19)

ŒˆTEST n D ŒˆTEST 
ŒQTEST  ; ŒˆTAM n D ŒˆTAM 
ŒQTAM  : (6.20)

Step 2: Pseudo-orthogonalities and pseudo cross-orthogonality

ŒORTEST  D ŒˆTEST n ŒˆTEST n ; ŒORTAM  D ŒˆTAM n ŒˆTAM n (6.21)

ŒCOR D ŒˆTAM n ŒˆTEST n : (6.22)

Step 3: Modal coherence

ŒC  D ŒORTAM n 1 ŒCOR ; ŒR D ŒˆTEST n ŒˆTAM n ŒC  (6.23)

ŒCOH D ŒORTEST  ŒR R D ŒC  ŒORTAM  ŒC  (6.24)

Extensive evaluation of the complex least squares process indicates that both the com-
plex cross-orthogonality (Equation (6.22)) and coherence (Equation (6.24)) matrices behave
in a manner that parallels the behavior of their real eigenvector counterparts. Therefore, com-
plex modal orthogonality and test-analysis mode correlation may be separated into a “recipe”
composed of two distinct sub-tasks.
 6.2. PART 2: CORRELATION OF TEST AND FEM MODES 115
1. Complex mode orthogonality is perfectly satisfied by left-hand eigenvector counterparts
to state-space complex eigenvectors for test and TAM eigenvector sets (Equations (5.19)
and (6.13), respectively).
2. Complex mode cross-orthogonality (correlation) and complex modal coherence are de-
fined based on the complex least-squares process (Equations (6.22) and (6.24), respec-
tively).
Due to (a) the presence of many closely spaced shell breathing modes of the test article
and (b) non-proportional damping (which is a basic reality of all built-up structural dynamic
systems), the test article’s modes are generally complex (and are not readily approximated as
“real” modes).

6.2.5 ILLUSTRATIVE EXAMPLE: ISPE MODAL TEST

The ISPE TAM model is described 75 real, undamped modes in the 0–65 Hz frequency band
computed from the reduced TAM model (expressed in state-space form according to Equa-
tions (6.15)–(6.18)). ISPE modal test data consists of 63 complex modes in the 0–65 Hz fre-
quency band, estimated by the SFD-2018 algorithm. Following the “recipe” based on Equa-
tions (6.13), (6.22), and (6.24), complex mode orthogonality, cross-orthogonality, and modal
coherence matrices are summarized in Figure 6.2.
The TAM and test mode orthogonality matrices are perfect identity matrices. The cross-
orthogonality matrix appears similar in form to its state-space counterpart (see Figure 6.1), and
the modal coherence matrix indicates that the majority of lower frequency (50%) of test modes
are strong linear combinations of the TAM modes.
Recalling the fact, introduced in Chapter 3, that the orthogonality matrix can be “un-
packed” to describe a subject system’s modal kinetic energy distribution (Equation (3.3)), a corre-
sponding “unpacking” of the complex state-space mode orthogonality matrix (Equation (6.13))
similarly describes the kinetic energy distribution of the complex modes.
In particular, the modal orthogonality and kinetic energy distribution relationship pair for
a complex state-space modal set is defined, in terms of left- and right-hand eigenvectors as,

ŒOR D ŒˆL  Œˆ D ŒˆV;L  ŒˆV  C ŒˆU;L  ŒˆU  ;

      (6.25)
ŒKEˆ  D conj ˆL ˝ Œˆ D conj ˆV;L ˝ ŒˆV  C conj ˆU;L ˝ ŒˆU  :
Note that Œˆ in the present context corresponds to the complex conjugate transpose
of the matrix, Œˆ; the same operator designation applies to the left-hand eigenvector matrix.
Partitioning of the complex state-space left- and right-hand eigenvectors into “velocity” and
“displacement” partitions (designated by the subscripts “V ” and “U ”, respectively), in order to
describe the modal kinetic energy distributions in terms of the measured DOFs, the above defi-
nitions of complex state-space eigenvector orthogonality and modal kinetic energy distributions
are independent of the TAM mass matrix.
116 6. SYSTEMATIC TEST ANALYSIS CORRELATION
ISPE Test Modal Orthogonality ISPE TAM Mode Orthogonality
(SFD-2018 Complex Modes) (State-Space Format)
100 100

10 90 10 90

80 80
20
20 70 70

Test Mode

TAM Mode
60 30 60
30
50 40 50

40 40 40
50
30 30
50 60
20 20

10 10
60 70
0
10 20 30 40 50 60 10 20 30 40 50 60 70
Test Mode TAM Mode

ISPE TAM to Test Modal Cross-Orthogonality ISPE TAM to Test Modal Coherence
(Complex Least Squares) (Complex Least Squares)
90 90
10 10
80 80

20 70 70
20
60 60
TAM Mode

Test Mode
30
50 30 50
40
40 40
40
50
30 30

60 20 50 20

10 10
70
60
10 20 30 40 50 60 10 20 30 40 50 60
Test Mode Test Mode

Figure 6.2: Orthogonality, cross-orthogonality coherence of TAM and test complex eigenvec-
tors.

6.2.6 ROADMAP FOR A HIGHLY IMPROVED INTEGRATED TEST

ANALYSIS PROCESS
Introduction of the complex state-space viewpoint for experimental modal analysis (SFD-2018)
and the TAM mass matrix independent metrics described in the present chapter, suggests a po-
tential paradigm shift for the integrated test-analysis process. A roadmap for this envisioned
highly improved test analysis process, encompassing experimental modal analysis and system-
atic test analysis correlation, employing ISPE modal test data and mathematical model data, is
outlined in Figures 6.3–6.6.
 6.2. PART 2: CORRELATION OF TEST AND FEM MODES 117

Current Practice Emerging Option

• Mode V&V via “Curve Fitting” • V&V via Isolated Mode FRF
FRF Display: File: ISPETESTDFRFs, Mode 1, Input 1, Coh=99%, 06-Mar-2019 11:35 AM
Magnitude

Magnitude
h(f)
hmodal(f)

Frequency (Hz)

Hz

Magnitude

Real
Imaginary
Phase (Deg)

Phase (Deg)

Imaginary
Real

Frequency (Hz) Frequency (Hz)

Hz

• Mode V&V: Self-orthogonality • Perfect Self-orthogonality

[ORt] = [ΦTt MaaΦ] [OR]TEST = [ΦL]TEST [Φ]TEST ≡ [I]
Orthogonality of Normalized SFD Modes Orthogonality of Normalized SFD Modes
100 100
90 90
10 10
80 80
20 70 20 70
60 60
30
Test

30

Test
50 50

40 40 40 40
30 30
50
50 20 20

10 60 10
60
10 20 30 40 50 60
10 20 30 40 50 60 Test
Test

Note: Emerging option is independent of the math model’s TAM mass matrix

Figure 6.3: ISPE experimental modal analysis and modal orthogonality.

118 6. SYSTEMATIC TEST ANALYSIS CORRELATION

Current Practice Emerging Option

• Test-Analysis Cross-Orthogonality • Test-Analysis Cross-Orthogonality
[COR] = [ORa]-1[ΦaTMaaΦt] [COR] = [ΦTAM]n* [ΦTEST]n
Cross-Orthogonality, |COH| of normalized (Real) TAM vs. Test Modes Cross-Orthogonality, |COH| of normalized Shade-Space TAM vs. Test Modes

90 90
10 10
80 80
20 70 20 70

30 60 30 60
Test

Test
50 50
40 40
40 40
50 50
30 30
60 60
20 20
70 10 70 10

10 20 30 40 50 60 10 20 30 40 50 60
Tam Tam

• Test-Analysis Modal Coherence

[COH ] = ![I ]–[ORt ]-1/2"RTMR#[ORt ]-1/2$= • Test-Analysis Modal Coherence
[ORt ]-1/2"CORtORaCOR#[ORt ]-1/2 [COH] = [ORTEST]–[R*R] = [C]* [ORTAM][C]
Coherence, |COH| of normalized (Real) TAM vs. Test Modes Coherence, |COH| of normalized Shade-Space TAM vs. Test Modes

90 90
10
10 80 80

70 20 70
20
60 60

Test
30
Test

30 50 50

40 40
40 40
30 30
50 50 20
20
10 10
60 60
10 20 30 40 50 60 10 20 30 40 50 60
Test Test

Note: Emerging option is independent of the math model’s TAM mass matrix

Figure 6.4: ISPE test-analysis cross-orthogonality and modal coherence.

 6.2. PART 2: CORRELATION OF TEST AND FEM MODES 119

Current Practice Emerging Option

• Modal Kinetic Energy • Modal Kinetic Energy
[KEΦ] = [MΦ]⊗[Φ] [KEΦ] = conj[Φ*L ]⊗[Φ]
• Mode Shape 1 (real) • Mode Shape 1 (complex)

• Kinetic Energy 1 (real) • Kinetic Energy 1 (complex)

Note: Emerging option is independent of the math model’s TAM mass matrix

Figure 6.5: ISPE test mode 1 mode shape and modal kinetic energy.
120 6. SYSTEMATIC TEST ANALYSIS CORRELATION

Current Practice Emerging Option

• Modal Kinetic Energy • Modal Kinetic Energy
[KEΦ] = [MΦ]⊗[Φ] [KEΦ] = conj[Φ*L ]⊗[Φ]
• Mode Shape 3 (real) • Mode Shape 3 (complex)

• Kinetic Energy 3 (real) • Kinetic Energy 3 (complex)

Note: Emerging option is independent of the math model’s TAM mass matrix

Figure 6.6: ISPE test mode 3 mode shape and modal kinetic energy.
 6.2. PART 2: CORRELATION OF TEST AND FEM MODES 121
The roadmap for a highly improved integrated test analysis process offers the following
innovations.

1. Experimental modal analysis moves from the current, prevailing practice of verification
and validation through curve fitting back to past (pre-digital era) practice of single mode
isolation made possible by the SFD-2018 technique.

2. Established goals for experimental mode orthogonality (NASA-STD-5002 and USAF

Space Command SMC-S-004) are automatically satisfied, without dependence on a pos-
sibly flawed and/or inaccurate TAM mass matrix.

3. Test-analysis cross-orthogonality goals (NASA-STD-5002 and USAF Space Command

SMC-S-004) are addressed without dependence on a possibly flawed and/or inaccurate
TAM mass matrix. Moreover, the modal coherence analysis is defined independent of the
TAM mass matrix.

4. While the conventionally used “real” test mode approximation often is similar in content
to the corresponding SFD-2018 estimated test mode, modal kinetic energy distributions
associated with the “real” and “complex” test modes may differ from one another. This is
attributed to the fact that the conventional “real” test mode based modal kinetic energy
depends on a possibly flawed and/or inaccurate TAM mass matrix. This is clearly demon-
strated by the lack of agreement between conventional and complex modal kinetic energy
distributions for ISPE mode 1, and close agreement between conventional and complex
modal kinetic energy distributions for ISPE mode 3.

5. The emerging option for a highly improved integrated test analysis process must be re-
viewed, applied in parallel with established with established practice, and aggressively
“poked and prodded” by the technical community before this “paradigm shift” is accepted.

6.2.7 CLOSURE
The ISPE modal test experience led to introduction of experimental modal analysis and test-
analysis correlation tool enhancements designed to deal with particularly difficult modal testing
challenges. Discrimination of valid and spurious experimental modes was addressed in Chap-
ter 5. Test-analysis correlation challenges associated with complex, closely-spaced experimental
modes were addressed in the above discussion. Conclusions associated with the newly intro-
duced test-analysis correlation enhancements are as follows.

1. The left-hand, right-hand eigenvector based state-space orthogonality matrix for complex
experimental modes automatically satisfies NASA-STD-5002 and USAF SMC-004 re-
quirements. The orthogonality matrix for experimental modes is mathematically perfect
by definition, and it is independent of the approximate TAM mass matrix.
122 6. SYSTEMATIC TEST ANALYSIS CORRELATION
2. Employment of the complex least-squares formulation for test-TAM correlation and
modal coherence appears to be an appropriate enhancement for incorporation in NASA-
STD-5002 and USAF SMC-004 test-analysis correlation standards. It is recognized that
the task of modal correlation does not specifically require perfectly orthogonal, real modes.
3. Introduction of the complex least-squares formulation for test-TAM correlation and
modal coherence opens the opportunity for inclusion of damping in correlation and up-
date endeavors, e.g., structures with non-negligible modal complexity due to localized,
non-proportional damping mechanisms (joints).
The developments influenced by the ISPE test experience offer potential enhancements
to U.S. Government modal testing standards that must undergo wider study and critique before
establishment of general acceptance in the structural dynamics technical community.

6.3 REFERENCES
[1] Load analysis of spacecraft and payloads, NASA-STD-5002, 1996. 107, 108, 109, 112,
114
[2] Independent Structural Loads Analysis, U.S. Air Force Space Command, SMC-S-004,
2008. 107, 108, 112, 114
[3] K. Miller, Complex linear least squares, SIAM Review, 15(4):706–726, 1973. DOI:
10.1137/1015094. 107, 112, 114
[4] R. Guyan, Reduction of stiffness and mass matrices, AIAA Journal, 3, 1965. DOI:
10.2514/3.2874. 113
 123

CHAPTER 7

Reconciliation of Finite
Element Models and Modal
Test Data
7.1 PART 1: FINITE ELEMENT MODEL MODAL
SENSITIVITY
7.1.1 INTRODUCTION
Efficient computation of structural dynamic modal frequency and mode shape sensitivities as-
sociated with variation of physical stiffness and mass parameters is essential for (1) practical
design sensitivity and uncertainty studies and (2) reconciliation of finite element models with
modal test data. Sensitivity analysis procedures fall in two distinct categories, namely (a) modal
derivatives for small parametric variation and (b) altered system modes associated with “large”
parametric variation. The latter category is generally applicable to modal testing, which often
requires significant local parameter changes at joints to effect FEM-test reconciliation. How-
ever, many investigators and commercial software packages employ estimated modal derivatives
in optimization strategies, which address FEM-test reconciliation objectives.
Since the 1960s, methods for computation of modal frequency and mode shape deriva-
tives have evolved. Fox and Kapoor [1] introduced an exact derivative formulation that required
knowledge of all modes of the original system; application of the procedure when a truncated
set of modes was employed produced compromised derivatives. In response to this difficulty,
Nelson [2] derived an exact formulation for computation of mode shape derivatives for trun-
cated mode sets. Efforts to refine and extend application of mode shape derivatives for finite
parameter change sensitivity computations have been pursued by many investigators (including
the present author). However, the need for modal frequency and mode shape sensitivities that
map over very large ranges for multiple parameters suggests application of alternative Ritz [3]
strategies.
The Ritz method is one of the most significant developments in analytical mechanics of
the past century. This method provides a logical energy formulation for consistent reduction of
mass and stiffness matrices employing a set of trial vectors as a reduction transformation. Effec-
tiveness and accuracy of the reduction process depends on selection of an appropriate trial vector
set. When a truncated set of baseline system mode shapes is used as the trial vector set (popularly
124 7. RECONCILIATION OF FINITE ELEMENT MODELS AND MODAL TEST DATA
known as Structural Dynamic Modification (SDM)) [4], the Ritz method often produces poor
estimates for the altered system. Augmentation of the truncated baseline system mode shapes
with appropriately defined additional vectors, however, has been found to produce extremely
accurate altered system modal frequencies and mode shapes. Quasi-static residual vectors [5],
appended to a truncated set of mode shapes, were found to produce extremely accurate modes
for offshore oil platform models subjected to localized alterations [6]. Residual Mode Augmen-
tation (RMA), introduced in 2002 [6] and subsequently refined [7], is a procedure that defines
augmented trial vectors, which are appropriate for structures subjected to highly distributed, as
well as localized, alterations.

7.1.2 SENSITIVITY ANALYSIS STRATEGIES

The present discussion focuses on Ritz procedures that address structural sensitivities due to
stiffness and mass alterations described by large (as opposed to small) parametric variations.
Therefore, formulations that address computation of eigenvalue and mode shape derivatives are
not considered.
The matrix equations describing exact free vibration of baseline and altered structures,
respectively, are

ŒKO  ŒˆO  ŒMO  ŒˆO  ŒO  D Œ0 ; (7.1)

ŒKO C p
K Œˆ ŒMO C p
M  Œˆ Œ D Œ0 : (7.2)

It is implicitly assumed that the stiffness and mass changes scale linearly with respect to
the parameter, p . Therefore, changes in “beam” depth may not be directly applied, since the
axial stiffness (AE) scales linearly with depth and the flexural stiffness (EI) scales as the cube
of depth. The appropriate formulation for Equation (7.2) permits linear sensitivity of “AE” and
“EI” separately.
The relationship between mode shapes of the baseline and altered structures is expressed
as the cross-orthogonality of orthonormal mode shape sets
 
ŒCOR D ˆTO ŒMO  Œˆ ; (7.3)
where the modal self-orthogonality properties are

   
ŒORO  D ˆTO ŒMO  ŒˆO  D ŒIO  ; ŒOR D ˆT ŒMO C pM  Œˆ D ŒI  : (7.4)

The most fundamental Ritz approximation, commonly used in SDM [4] employs a trun-
cated set of low frequency eigenvalues as the reduction transformation described by

Œˆ D ŒˆOL  Œ' : (7.5)

 7.1. PART 1: FINITE ELEMENT MODEL MODAL SENSITIVITY 125
The reduced baseline structure stiffness and mass matrices are, respectively,
   
ŒkO  D ˆTOL KO ˆOL D ŒOL  ; ŒmO  D ˆTOL MO ˆOL D ŒIOL  ; (7.6)
and the reduced stiffness and mass sensitivity matrices are, respectively,
   
Œk D ˆTOL KˆOL ; Œm D ˆTOL MˆOL : (7.7)
Therefore, the reduced altered structure free vibration equation is

ŒOL C p
k Œ' ŒIOL C p
m Œ' Œ D Œ0 : (7.8)

A well-known result of this type of trial vector reduction strategy is that the approximate
altered structure eigenvalues are generally higher than results for the exact solution, and the
approximate mode shapes do not closely follow the exact shapes when parametric alterations are
large.

7.1.3 RESIDUAL VECTOR AUGMENTATION FOR LOCAL

ALTERATIONS
The static displacements for a baseline structure subjected to unit loads (at physical degrees of
freedom where the structure is to be altered) described by the columns of a load array, Œ€, are
the solutions of

ŒKO  ŒUS  D Œ€ : (7.9)

A low-frequency modal approximation of static displacements for the above system em-
ploys the transformation

ŒUSL  D ŒˆOL  ŒqL  ; (7.10)

resulting in the approximate static displacements
    
ŒqL  D L1 ˆTOL Œ€ ; ŒUSL  D ˆOL L1 ˆTOL Œ€ : (7.11)
The difference between the exact and approximate static solutions defines MacNeal’s [5]
quasi-static residual vectors,

     
‰ D ŒUS  ŒUSL  D KO1 ˆOL L1 ˆTOL Œ€  ˆOH H1 ˆTOH Œ€ ; (7.12)

which have been mathematically proven to be the quasi-static displacements associated with all
of the high-frequency mode shapes. An orthonormalized set of residual vectors is defined by
solution of the residual eigenvalue problem,
126 7. RECONCILIATION OF FINITE ELEMENT MODELS AND MODAL TEST DATA

      
k ' m '  D Œ0 ; (7.13)
where the generalized residual mass and stiffness matrices are
   T     T 
k D ‰ KO ‰ ; m D ‰ MO ‰ : (7.14)
The augmented trial vector set (replacing the reduction transformation of Equation (7.5))
is

   
N̂ OL D ˆOL ˆ ; (7.15)
    
ˆ D ‰ ' : (7.16)

When structural alterations are localized, relatively few residual vectors adequately de-
scribe the content of changed system mode shapes. The previously described innovation loses
its appeal when structural alterations are well-dispersed requiring utilization of many residual
vectors.

7.1.4 RESIDUAL MODE AUGMENTATION (RMA) FOR DISPERSED

ALTERATIONS
Definition of residual vectors associated with dispersed, independent alterations of a baseline
structure, described by Equation (7.1), is accomplished by first computing the lowest-frequency
mode shapes of the baseline structure (Equation (7.5)) as well as the lowest mode shapes asso-
ciated with each independent alteration of the structure

ŒKO C pNi Ki  ŒˆiL  ŒMO C pNi Mi  ŒˆiL  ŒiL  D Œ0 .for i D 1; : : :; N /: (7.17)

The selected value of each independent scaling parameter is sufficiently large to produce
a substantial change in mode shapes (with respect to the baseline structure). An initial set of
trial vectors that adequately (and perhaps redundantly) encompass all potential (low frequency)
altered system mode shapes is
 
Œ‰ D ˆ1L ˆ2L ::: ˆNL : (7.18)
This set of trial vectors is expressible as the sum of (a) a linear combination of baseline
system mode shapes and (b) trial vectors (that are linearly independent of the baseline system
mode shapes)
 
Œ‰ D ŒˆOL  ŒCOR C ‰ 0 : (7.19)
 7.1. PART 1: FINITE ELEMENT MODEL MODAL SENSITIVITY 127
The cross-orthogonality coefficient matrix is determined based on the following least-
squares solution:

     
ˆTOL MO ‰ D ˆTOL MO ˆOL ŒCOR C ˆTOL MO ‰ 0 D ŒIOL  ŒCOR C Œ0 ; (7.20)

where
 
ŒCOR D ˆTOL ŒMO  Œ‰ ; (7.21)
 0  
‰ D IOL ˆOL ˆTOL MO Œ‰ : (7.22)

The “purified” trial vector set is linearly independent of the baseline system mode shapes
in a manner similar to MacNeal’s residual vectors, as follows:

h i   
T
‰ 0 MO ˆOL D ‰ T IOL MO ˆOL ˆTOL ŒMO ˆOL 
 

D ‰ T MO ˆOL MO ˆOL ˆTOL MO ˆOL  Œ0 (7.23)
h i   
T
‰ 0 KO ˆOL D ‰ T IOL MO ˆOL ˆTOL ŒKO ˆOL 
 
D ‰ T ŒKO ˆOL MO ˆOL OL   Œ0 : (7.24)

While the “purified” trial vector set has the above property, it includes an unnecessarily
large number of vectors. An appropriate, substantially smaller set of residual vectors is identified
by singular value decomposition of the generalized mass matrix,
 
ŒA D ‰ 0T MO ‰ 0 : (7.25)
The singular value decomposition process involves solution of the eigenvalue problem,
    
ŒA ' D '  ; 1  2  3  : : : : (7.26)
The cut-off criterion, noted below employed to define suitable reduced trial vector set, is

N N
 tol D 10 (where N  4–6 is usually adequate): (7.27)
1
The augmented trial vector set (replacing the reduction transformation of Equation (7.5))
is
   
N̂ OL D ˆOL ‰ 0 ' : (7.28)
The form of the resulting Ritz, multi-parameter sensitivity model (for selected values of
the scaling parameters) is
128 7. RECONCILIATION OF FINITE ELEMENT MODELS AND MODAL TEST DATA

" N
# " N
#
X X
kO C pi Œki  Œ' mO C pi Œmi  Œ' Œ D Œ0 ; (7.29)
iD1 i D1

   
ŒkO  D N̂ TOL KO N̂ OL ; ŒmO  D N̂ TOL MO N̂ OL ;
   
Œki  D N̂ TOL Ki N̂ OL ; Œmi  D N̂ TOL Mi N̂ OL : (7.30)

Recovery of mode shapes in terms of physical DOF is accomplished with

 
Œˆ D N̂ OL Œ' : (7.31)

7.1.5 RMA SOLUTION QUALITIES

Since its introduction in 2001, RMA has exhibited the capability to accurately follow modal
sensitivity trends over an extremely wide range of parametric variation. The simple cantilevered
(planar) beam example, provided in Figure 7.1, demonstrates typical RMA performance (“100%”
is baseline). Actual cross-orthogonality checks are also excellent.

Support EI1 EI2 EI3

Baseline: EO is Uniform
Illustrative Cantilevered Beam Modal Frequency Sensitivity
104
Modal Frequency (Hz)

103

102

101
Exact Reference Perturbed Modes
Exact Extreme Perturbed Modes
Augmented Mode-Based Sensitivity
100 -1
10 100 101 102
Percent of Baseline “EI2”

Figure 7.1: RMA sensitivity performance for a cantilevered beam example.

 7.1. PART 1: FINITE ELEMENT MODEL MODAL SENSITIVITY 129
7.1.6 ILLUSTRATIVE EXAMPLE: ISPE MODAL TEST
In 2017, an early finite element model of the ISPE test article, provided by Dr. Eric Stewart of
NASA/MSFC, was employed for an RMA convergence study. Details of the ISPE model with
parametric sensitivity regions are summarized in Figure 7.2.

Negligible
Sensitivity

Figure 7.2: ISPE finite element model and parametric sensitivity regions.

In response to concerns brought up by Dr. Eric Stewart of NASA/MSFC regarding RMA

solution convergence, an investigation of the matter was conducted. Specifically, the role of
the SVD tolerance parameter .tol/, defined in Equation (7.26), was evaluated. An objective
convergence criterion was developed based on comparison of parametric alterations resulting
from the solution of the exact modal equation,
" # " #
X X
KO C pi Ki Œˆe  D MO C pi Mi Œˆe  Œe  ; (7.32)
i i

and the approximate modal equations (see Equations (7.28)–(7.30), developed for a specific
value of “tol”),
130 7. RECONCILIATION OF FINITE ELEMENT MODELS AND MODAL TEST DATA

" # " #
X X  
kO C pi ki Œ'a  D mO C pi mi Œ'a  Œa  ; Œˆ D N̂ OL Œ' : (7.33)
i i

The metric for evaluation of approximate solution convergence is the cross-orthogonality

matrix associated with exact and approximate (RMA) modal sets, specifically,
 
ŒCORea  D ˆTe ŒMO  Œˆa  : (7.34)
Convergence of the approximate (RMA) modal set is therefore judged on the basis of how
close the (absolute value) cross-orthogonality matrix is to an identity matrix. In addition, the
difference between exact and approximate corresponding modal frequencies is also employed as
part of convergence evaluation.
Before engaging in the actual RMA convergence study, a preliminary evaluation of modal
sensitivities for each of the 28 parametric variations (parameter change set to a value of pi D 1)
was conducted, wherein the cross-orthogonality between the baseline modes and exact perturbed
modes,
 
ŒCOReO  D ˆTe ŒMO  ŒˆO  ; (7.35)
and corresponding modal frequencies were evaluated. Results of that exercise, summarized in
Table 7.1 (note DF represents modal frequency change), indicate that 13 of the total of 28
parametric variations were insignificant.
It should be noted that the numerical values provided in the above table are the peak
frequency and cross-orthogonality alterations associated with the lowest 60 normal modes of
each “unit” parametric variation. As a result of this finding, only the “sensitive” 16 parameters
were evaluated in the RMA convergence study.
Results associated with the RMA convergence study, which used “unit” parametric vari-
ations and values of tolerance .tol/ set to 1e-4, 1e-5, and 1e-6, respectively, are summarized in
Table 7.2.
It is clear from the above results that tol D 1e-6 produces highly converged RMA modes
for the ISPE.

7.1.7 CLOSURE
Alteration of a structural dynamic model for the purpose of reconciliation with respect to mea-
sured data typically requires moderate to large variation in (stiffness and/or mass) parameters.
Moreover, even when small parametric variations of parameters are required, close spacing of
system modes produces large variations in modal vectors. Therefore, modal derivatives are not
well suited for tracking of parametric sensitivities of structural dynamic modes. A more robust
strategy for approximate modal sensitivity analysis employs the Ritz method. Specifically, SDM
 7.1. PART 1: FINITE ELEMENT MODEL MODAL SENSITIVITY 131

Table 7.1: Evaluation of the significance of ISPE model parametric variations

Case DF (%) 100-Cor (%) Class Typical Sensitive Case

Baseline File: 1.1c.mat, Off-Nominal File: 1.2c.mat
1.2c.mat 4 31 5
1.3c.mat 5 30 4
3
1.4c.mat 9 17 Sensitive

DF (%)
2
1.5c.mat 13 97 1
1.6c.mat 13 90 0

1.7c.mat 2 35 -1
10 20 30 40 50 60
1.8c.mat 1 3
1.9c.mat 1 7 100

1.10c.mat 0 0 Insensitive 80

100-Cor (%)
1.11c.mat 1 7 60

1.12c.mat 1 10 40

20
1.13c.mat 2 81
Sensitive
0
1.14c.mat 4 22 10 20 30 40 50 60

1.15c.mat 0 0 Mode

1.16c.mat 0 1 Insensitive
Typical Insensitive Case
1.17c.mat 2 8 Baseline File: 1.1c.mat, Off-Nominal File: 1.8c.mat
1.18c.mat 23 95 Sensitive 5
4
1.19c.mat 1 6 Insensitive
3
DF (%)

1.20c.mat 6 93 2
1.21c.mat 3 89 1
0
1.22c.mat 2 15
-1
1.23c.mat 23 100 Sensitive 10 20 30 40 50 60

1.24c.mat 16 100
100
1.25c.mat 8 2
80
1.26c.mat 28 100
100-Cor (%)

60
1.27c.mat 0 0
40
1.28c.mat 0 0
Insensitive 20
1.29c.mat 0 0
0
10 20 30 40 50 60
1.30c.mat 0 0
Mode
132 7. RECONCILIATION OF FINITE ELEMENT MODELS AND MODAL TEST DATA
Table 7.2: Summary of RMA convergence study results

Tolerance 1.00E-04 1.00E-05 1.00E-06

Residuals 65 160 291
Case |∆f| (%) |∆C| (%) |∆f| (%) |∆C| (%) |∆f| (%) |∆C| (%)
1.2c.mat 1.5 11 1.2 10 0.1 3
1.3c.mat 2.3 16 1.9 16 0.0 1
1.4c.mat 3.2 33 1.6 29 0.1 1
1.5c.mat 1.0 16 0.5 1 0.2 0
1.6c.mat 1.3 1 1.0 3 0.1 0
1.7c.mat 0.4 8 0.4 4 0.1 0
1.13c.mat 0.3 12 0.1 2 0.0 0
1.14c.mat 2.5 16 0.6 5 0.0 0
1.18c.mat 0.7 19 0.1 0 0.0 0
1.20c.mat 3.0 20 1.4 7 0.1 2
1.21c.mat 0.7 15 0.3 1 0.0 1
1.22c.mat 0.4 7 0.2 1 0.1 1
1.23c.mat 1.0 4 0.5 3 0.1 0
1.24c.mat 0.1 0 0.1 0 0.0 0
1.25c.mat 1.7 99 0.6 6 0.1 1
1.26c.mat 0.9 5 0.2 2 0.0 0
Notes:
1. |∆f| (%) = [approximate−exact frequency]/[exact frequency] (%)
2. |∆C| (%) = 100% - [Cross-Othogonality] (%)
3. |∆f| and |∆C| are the envelopes associated with the lowest 60 system modes

employs a truncated set of baseline system modes as trial vectors to define reduced order mass
and stiffness matrices. SDM is a mathematically stable method for approximate parametric sen-
sitivity analysis; however, the baseline system modes are often not adequate for accurate tracking
of sensitivities.
RMA is a methodology that defines trial vectors that augment baseline structure modes,
resulting in substantially improved the ability of SDM to efficiently track parametric sensitivities
of structural modes. RMA employs (a) system modes associated with large reference parametric
alterations and (b) SVD to define a reduced set of residual vectors that are rich in dominant
geometric changes experienced by the subject structural system. Since its introduction in 2001,
RMA has been successfully employed during several modal tests and a variety of mathematical
 7.2. PART 2: TEST-ANALYSIS RECONCILIATION 133
model studies. Most recently, RMA was employed to investigate parametric sensitivities for the
ISPE structure. As a result of this work, two enhancements of the method have been introduced.
a. Preliminary screening of candidate parametric sensitivities based on reference frequency
shifts and cross-orthogonality metrics was defined to differentiate “sensitive” from “insen-
sitive” cases (eliminating “insensitive” parametric variations from further consideration).
b. A frequency and cross-orthogonality (SVD) convergence metric for determination of an
augmented residual trial vector set that satisfies RMA accuracy requirements.

7.2 PART 2: TEST-ANALYSIS RECONCILIATION

7.2.1 INTRODUCTION
Reconciliation of a test article’s finite element model with experimental modal data, if conducted
in an objective and systematic manner, requires minimization of a cost function. A variety of
modal cost functions are employed by many investigators. The present discussion describes a
particular cost function that describes a balanced modal frequency and mode shape “error” rela-
tionship. Minimization of the modal cost function’s error norm employing gradient based and
Monte Carlo strategies are evaluated.

7.2.2 MODAL COST FUNCTION .C / AND OPTIMAL TEST-ANALYSIS

RECONCILIATION
Consider the standard expression for the undamped structural dynamics eigenvalue problem,

ŒK Œˆ ŒM  Œˆ Œ D Œ0 : (7.36)

When modal test data is substituted into the above expression,

ŒK Œˆt  ŒM  Œˆt  Œt  D ŒR ; (7.37)

there is a residual error, ŒR, due to (a) differences between the FEM and modal test data and
(2) measurement error.
Pre-multiplication by the FEM mode shapes results in
 T     
ˆ Kˆt ˆT Mˆt Œt  D ˆT R : (7.38)
By substituting the transpose of ŒKŒˆ from Equation (7.36), the above relationship be-
comes
     
Œ ˆT Mˆt ˆT Mˆt Œt  D ˆT R : (7.39)
Finally, pre-multiplication of the above result by the inverse of the FEM eigenvalues de-
fines the modal cost function ŒCˆ  as
134 7. RECONCILIATION OF FINITE ELEMENT MODELS AND MODAL TEST DATA

 1
 1
ŒCˆ  D  ˆT R D ŒCOR Œ ŒCOR Œt  ; (7.40)
where the cross-orthogonality matrix, ŒCOR, associated with test and FEM modes is
 
ŒCOR D ˆT Mˆt : (7.41)
When there are rigid body and/or zero-frequency mechanism modes present, a shift op-
erator, s , is employed to avoid numerical problems. The shifted modal cost function is

1
ŒCˆ  D ŒCOR Œ C s  ŒCOR Œt C s  : (7.42)
The modal cost function is defined for any corresponding sets of truncated FEM and test-
based modal data. It should be noted that when rigid body and/or mechanism modes are not
explicitly measured, the set of test modes should be augmented with the corresponding FEM
modes.
The FEM may be efficiently altered employing the RMA method discussed in Part 1 of
this chapter, for specific values of selected parameters. A reconciled FEM is therefore defined
by the combination of selected parameter alterations, which minimize the norm of the modal
cost function, ŒCˆ . This may be accomplished by use of a random, Monte Carlo [8] parameter
search strategy or the Nelder–Meade Simplex algorithm [9].

7.2.3 ILLUSTRATIVE EXAMPLE: RECTANGULAR PLATE SUPPORTED

BY CORNER POSTS
The model illustrated in Figure 7.3 is a TAM, defined for a 1386 DOF FEM. Note that the
heavy lines illustrate the regions of the model, which are tagged as uncertainties. Two stiffness
parameters map the uncertain components: (1) plate edge bending stiffness and (2) elastic sup-
port post stiffness. Simulated modal test data (12 modes) was computed based on plate edge
bending stiffness increased by 300% and elastic support stiffness reduced by 70%. The base-
line FEM of the structure was augmented by sensitivity matrices, which describe the uncertain
regions.
In this illustrative example, performances of Monte Carlo and Nelder–Meade Simplex
search strategies applied to the modal cost function were explored. Results of the Monte Carlo
search analysis are illustrated in Figure 7.4, and results of the Nelder–Meade Simplex search
analysis are illustrated in Figure 7.5. Both search strategies reliably estimated the two FEM
stiffness parameters.
The bar graph provided in Figure 7.6 illustrates effectiveness of RMA modal sensitivity
and the modal cost function (norm) minimization strategy for the present illustrative example.
 7.2. PART 2: TEST-ANALYSIS RECONCILIATION 135

Z
Y X

Figure 7.3: Rectangular plate with corner supports.

100 100
Error Norm

Error Norm

Random Search Values for P(2)

MinError Value for P(2)=0.696

10-1 Random Search Values for P(1) 10-1

MinError Value for P(1)=2.94

0.5 1 1.5 2 2.5 3 3.5 -0.8 -0.7 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1
Parameter (1) Parameter (2)

Figure 7.4: Rectangular plate Monte Carlo search analysis patterns.

136 7. RECONCILIATION OF FINITE ELEMENT MODELS AND MODAL TEST DATA

Search Values for P(1) Search Values for P(2)

1.2 Optimum Value for P(1)=2.96 1.2 Optimum Value for P(2)=0.7

1 1

0.8 0.8

Error Norm

Error Norm
0.6 0.6

0.4 0.4

0.2 0.2

0 0.5 1 1.5 2 2.5 3 3.5 -1 -0.9 -0.8 -0.7 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0
Parameter (1) Parameter (2)

Figure 7.5: Rectangular plate Nelder–Meade simplex search analysis patterns.

5 Pre-Reconciliation
Post-Reconciliation
Test Frequencies
4.5

3.5
Frequency (Hz)

2.5

1.5

0.5

0 1 2 3 4 5 6 7 8 9 10 11 12
Mode Number

Figure 7.6: Rectangular plate modal cost function (norm) minimization results.
 7.2. PART 2: TEST-ANALYSIS RECONCILIATION 137
7.2.4 ILLUSTRATIVE EXAMPLE: ISS P5 MODAL TEST
The ISS P5 modal test, conducted in 2001, was the first opportunity to employ RMA sensitivity
and modal cost function (norm) minimization methodology on a “production” modal test. Before
presenting results of this particular investigation, it is important to emphasize valuable “lessons
learned” from this first experience.
1. Due to “noise” in the modal test data, gradients associated with the modal cost func-
tion, ŒCˆ , were not sufficiently “smooth” for application of the Nelder–Meade Simplex
method. Therefore the Monte Carlo search analysis strategy was employed (in this test
and all subsequent modal tests since then).
2. Error norm “clouds,” which are 2D projections of the multidimensional modal cost func-
tion norm and individual parametric variations (all parameters are simultaneously varied
in the search process) indicate the relative significance of each parameter (as illustrated in
Figure 7.7).

8 Random Search Values for P 8 Random Search Values for P

MinError Value for P=0.00371 MinError Value for P=1

7 7

6 6
Significant Parameter Insignificant Parameter
Error Norm

Error Norm
5 5

4 4
Random Search Values for P(2)
MinError Value for P(2)=0.696
3 3

2 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 2 0.9998 0.9998 0.9999 0.9999 0.9999 0.9999 0.9999 1 1
Parameter Parameter

Figure 7.7: ISS P5 typical modal cost function error norms.

More than 25 FEM parameters were initially considered in the modal test-analysis recon-
ciliation exercise. It was initially concluded that nine FEM parameters, when adjusted, effected
satisfactory reconciliation for TSS2, TSS4, and TSS17 mode sets. The adjusted parameters are
summarized in Table 7.3. Note that only the left forward-mid and mid-aft strut end stiffness
(“LT FM Struts” and “LT MA Struts”), corresponding to joint play were altered from test to
test to effect reconciliation.
A summary of TSS2, TSS4, and TSS17 mode set reconciliations indicating compliance
with NASA [10] modal frequency and cross-orthogonality goals is provided in Table 7.4.
The previous model reconciliations were judged to be in appropriate compliance with
NASA-STD-5002 [10] goals for this challenging modal test (due to significant nonlinear dy-
namic behavior), by the test team and NASA management. The test-analysis reconciliation ex-
ercise was completed within two weeks of the completion of laboratory activities.
138 7. RECONCILIATION OF FINITE ELEMENT MODELS AND MODAL TEST DATA
Table 7.3: Summary of FEM parameters adjusted for reconciliation
TSS02 TSS04 TSS17
Parameter Description Original Value Update Value % Change Update Value % Change Update Value % Change

Table 7.4: ISS P5 data sets TSS2, TSS4, and TSS17 test to updated FEM NASA-STD-5002
compliance status
Pre-Test TSS02 TSS04 TSS17
Comments
Mode Freq (Hz) Post-Test (Hz) Test (Hz) ∆F (%) COR (%) Post-Test (Hz) Test (Hz) ∆F (%) COR (%) Post-Test (Hz) Test (Hz) ∆F (%) COR (%)

Satisfies NASA STD-5002 Modal Frequency

and On-Diagonal Cross-Orthogonality Goals

Frequency Band associated with Reduced

MI/MO Coherence (Strong Nonlinearity)
Satisfies NASA STD-5002 Modal Frequency
and On-Diagonal Cross-Orthogonality Goals

7.2.5 ILLUSTRATIVE EXAMPLE: ISPE MODAL TEST

While ISPE FEM-test reconciliation activities have been the responsibility of NASA/MSFC,
it is of interest here to engage in an independent hypothetical exercise that employs RMA sen-
sitivity and modal cost function (norm) minimization methodology tools. A hypothetical pair
of single parameter mass and stiffness sensitivity matrices, which are “rigged” to effect perfect
test-analysis correlation (ISPE modes in the 0–65 Hz band) for a unit value of the sensitivity
parameter, p , are defined. While the ISPE modes estimated with SFD-2018 are complex, the
target test modes for the present exercise are the real parts of those modes.
Migration of ISPE modal frequencies, presented in Figure 7.8, illustrates the capability of
RMA to track substantial modal frequency changes associated with large parametric variation
(as previously noted in Figure 7.1). Large modal frequency variations with “cross-overs” cannot
be adequately tracked through the use of exact and approximate modal derivatives.
Results of Monte Carlo based minimization of the single-parameter-based modal cost
function search analysis, and a summary of ISPE original TAM, reconciled TAM, and SFD-
2018 test modal frequencies, respectively, are summarized are illustrated in Figure 7.9.
Displays of SFD-2018 test mode orthogonality, TAM-test mode cross-orthogonality, and
TAM-test modal coherence, for the original and reconciled TAMs are illustrated in Figure 7.10.
Please note that the above ISPE model correlation and reconciliation exercise represents
an idealized hypothetical process. Official ISPE model correlation and reconciliation results,
performed by NASA/MSFC, are not presented in this book.
 7.2. PART 2: TEST-ANALYSIS RECONCILIATION 139

ISPE Modal Sensitivity: 15–40 Hz Band ISPE Modal Sensitivity: 40–65 Hz Band
40 65

Modal Frequencies (Hz) 35 60

Modal Frequencies (Hz)

30 55

25 50

20 45

15 40
0.985 0.99 0.995 1 1.005 1.01 1.015 0.985 0.99 0.995 1 1.005 1.01 1.015
Parameter (p) Parameter (p)

Figure 7.8: Migration of ISPE modal frequencies due to parametric variation.

35
60
30
50
25
Frequencie (Hz)
Error Norm

40
20

30
15 Original TAM
Reconciled TAM
20 SFD-2018 Test
10

Trials
5 Optimum 10

0 0
0.985 0.99 0.995 1 1.005 1.01 1.015 1.02 0 10 20 30 40 50 60
Parameter Parameter

Figure 7.9: ISPE modal cost error norm and ISPE modal frequency summary.
140 7. RECONCILIATION OF FINITE ELEMENT MODELS AND MODAL TEST DATA
Original TAM-Test Correlation
Original Test Mode Orthogonality Original TAM-Test Mode Cross-Orthogonality Original TAM-Test Modal Coherence
100
90 90
90
10 10 80 10
80
80
70 70
20 70 20 20
60 60
60
30 30 50 30
50

Test

TAM

Test
50
40 40 40
40 40 40
30 30 30
50 50 20 50
20 20
10 10 10
60 60 60

10 20 30 40 50 60 10 20 30 40 50 60 10 20 30 40 50 60
Test Test Test

Reconciled TAM-Test Correlation

Final Test Mode Orthogonality Final TAM-Test Mode Cross-Orthogonality Final TAM-Test Modal Coherence
100
90 90 90
10 10 10
80 80 80
20 70 20 70 20 70

60 60 60
30 30 30
50 50
Test

TAM

Test
50
40 40 40 40 40 40

30 30 30
50 50 20 50 20
20
10 10 10
60 60 60
10 20 30 40 50 60 10 20 30 40 50 60 10 20 30 40 50 60
Test Test Test

Figure 7.10: Original (pre-test) and reconciled TAM-Test correlation matrices.

7.2.6 ILLUSTRATIVE EXAMPLE: WIRE ROPE TEST ARTICLE

While the wire rope test article, discussed in Chapter 4, is not a modal test-correlation and
reconciliation application, it provides an example of detailed nonlinear system identification
for a dynamic system that possesses strong hysteretic behavior. It is instructive at this point
to recall results of preliminary data analysis and nonlinear MI/SO spectral analysis that were
presented in Chapter 4. Preliminary data analysis of the estimated isolator deflection and isolator
internal force probability density functions is summarized, along with a photograph of the test
configuration, in Figure 7.11.
The hypothesized “algebraic” nonlinear system composed of measured time histories (ap-
plied force and acceleration response) and synthesized “measured” time histories (cubed dis-
placement and velocity
jvelocityj) depicted in Figure 7.12 was subjected to MI/SO analysis.
The cumulative coherence plot, shown in Figure 7.13 indicates that incorporation of the
two nonlinear terms produces a nearly unit value cumulative coherence (red curve), while the
ordinary coherence associated with a linear model (blue curve) indicates reduced coherence.
While the above results provide clear evidence that the behavior of the wire rope isolators
is nonlinear, closer examination of isolator response to swept-sine excitation suggests the wire
 7.2. PART 2: TEST-ANALYSIS RECONCILIATION 141

Enidine WR2-200-108

Wire-Rope Isolators

Figure 7.11: Wire rope isolators subjected to broadband random excitation.

142 7. RECONCILIATION OF FINITE ELEMENT MODELS AND MODAL TEST DATA

F(t) H1

F (U(t))3 H2 Ü(t)

U
U̇(t) ∙ |U̇(t)| H3

Figure 7.12: Hypothesized MI/SO nonlinear system.

File: WireNonR…MI/MO: Navg=29, df=0.488, MI/SO Cumulative Coherence

1

0.9
Due to Force
0.8 Due to Force + U2Z 3
Due to Force + U2Z 3 + V 3Z*|V 3Z|
0.7
Cumulative Coherence

0.6

0.5

0.4

0.3

0.2

0.1

0
10 15 20 25 30 35 40 45 50
Hz

Figure 7.13: Hypothesized nonlinear system cumulative coherence.

 7.2. PART 2: TEST-ANALYSIS RECONCILIATION 143

Linear Spring Iwan Model

Figure 7.14: Hypothesized nonlinear wire rope model.

Optimum Wire Rope Fit: K1=420, F0=2.37, KF=2184 Optimum Wire Rope Fit: K1=420, F0=2.37, KF=2184
300 200
280 190
260 180
240 170
220 160
ERR

ERR
200 150
180 140
160 130
140 120
120 110
100 100
2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3 350 400 450 500
F0 K1

Optimum Wire Rope Fit: K1=420, F0=2.37, KF=2184

300
280
260
240
220
200
ERR

180
160
140
120
100
2100 2120 2140 2160 2180 2200 2220 2240 2260 2280 2300
K

Figure 7.15: Wire rope cost function error norm projections.

144 7. RECONCILIATION OF FINITE ELEMENT MODELS AND MODAL TEST DATA
rope isolator exhibits “hysteretic” nonlinear behavior. The hypothesized wire rope model consists
of the linear spring and Iwan friction slip [11] element depicted in Figure 7.14.
Unknown parameters for the model consist of (a) linear spring stiffness, K1, (b) total
Iwan stiffness, KF , and (c) total Iwan critical slip force, F 0; the Iwan parameters are uniformly
distributed among n D 5 sub-elements.
Estimation of the three unknown parameters was accomplished by minimization of the
error norm defined by the square of the absolute difference between measured (swept-sine) and
model force time histories. A Monte Carlo search strategy was employed to determine optimum
values for the three unknown parameters. Error norm projections for the three parameters are
illustrated in Figure 7.15.
Effectiveness of the wire rope nonlinear system identification process is illustrated in com-
parison of the measured and fitted model load-deflection plots provided in Figure 7.16. The
dashed and solid black lines indicate stiffness asymptotes associated with non-slip and fully
slipped behaviors.
While the majority of this book has been devoted to the integrated test analysis process
for linear structural dynamic systems, this final illustrative example offers a glimpse of promising
future developments that may expand the process to include systems with quite general nonlinear
features.

Optimum Wire Rope Fit: K1=429, F0=2.36, KF=2187

8

2
F (lb)

-2

-4 Test Data
Fitted Model
K1*DU
-6 (K1+KF)*DU

-8
-0.015 -0.01 -0.005 0 0.005 0.01 0.015
DU (in)

Figure 7.16: Results of the wire rope hysteretic nonlinear model system identification.
 7.2. PART 2: TEST-ANALYSIS RECONCILIATION 145
7.2.7 CLOSURE
Efficient and accurate reconciliation of finite element models with experimental modal data is
first of all dependent on high quality data sets associated with models and measured data sets,
assured by adherence to all steps of the integrated test analysis process presented in this book.
Careful formulation of predictive finite element models of a subject system’s anticipated oper-
ational and test article configurations, taking full advantage of modern CAE/FEM resources,
is essential. Specifically, finite element models must fully conform with design drawings and
supporting data, paying special attention to joint and interface details (to facilitate appropri-
ate parametric sensitivity analyses). Objective assessment of unexpected test results followed by
repeated diagnostic tests and evaluations, as illustrated in the ISS P5 modal test program, are
essential in many situations. Success of the integrated test analysis process is highly dependent
upon willingness of the management and technical staff members to ponder unexpected and in-
convenient experimental data, rather than force/adjust results in conformity with preconceived
notions.
On the assumption that the above prerequisite guidelines are followed, the two essential
tools for modal test analysis reconciliation are (1) accurate, efficient finite element sensitivity
models and (2) robust cost function definition and optimizations. This book features an effective
RMA procedure that appears to eliminate inaccuracies and ambiguities inherent in techniques
based on modal derivatives and truncated modal data sets, aka SDM. Indeed, RMA, which is
gaining a high level of acceptance in the U.S. aerospace technical community, provides a valued
enhancement for SDM. A variety of projects over the past 18 years, beginning with the ISS P5
modal test, have pointed to advantages of Monte Carlo over Nelder–Meade optimization due
to the presence of non-smooth gradients in test-analysis cost functions owing to “noise.” While
this author has employed one specific cost function definition for test-analysis reconciliation,
the reader should be aware of other cost function definitions and optimization strategies used
by experienced technical organizations.
While the primary emphasis in this book centers on modal test applications, the wire rope
test and system identification example, points to strategies aimed at objective characterization
of nonlinear dynamic phenomena.
Finally, it is recommended that serious attention should be given to the “roadmap for
a highly improved integrated test analysis process” presented in Chapter 6 of this book. The
“roadmap” offers a way around (1) dependence on a pre-test model based TAM mass matrix
(assuring perfectly orthogonal experimental modes) and (2) limitations associated with the un-
realistic “real-modes” approximation for closely spaced modes and the “many modes” problem.
In order to properly exploit the latest SFD-2018 methodology, test analysis reconciliation strate-
gies must be defined and evaluated to address the “reality” of complex experimental modes.
146 7. RECONCILIATION OF FINITE ELEMENT MODELS AND MODAL TEST DATA
7.3 REFERENCES
[1] R. I. Fox and M. P. Kapoor, Rates of change of eigenvalues and eigenvectors, AIAA Jour-
nal, 6, 1968. DOI: 10.2514/3.5008. 123
[2] R. B. Nelson, Simplified calculation of eigenvector derivatives, AIAA Journal, 14, 1976.
123
[3] W. Ritz, Über eine neue methode zur Lösung gewisser variationsprobleme der math-
ematischen physik, Journal für die reine und angewandte Mathematik, 135, 1908. DOI:
10.1515/crll.1909.135.1. 123
[4] Twenty years of structural dynamic modification—a review, IMAC 20, 2002. 124
[5] R. MacNeal, A hybrid method for component mode synthesis, Computers and Structures,
1, 1971. DOI: 10.1016/0045-7949(71)90031-9. 124, 125
[6] R. N. Coppolino, Structural mode sensitivity to local modification, SAE Paper 811044,
1981. DOI: 10.4271/811044. 124
[7] R. N. Coppolino, Methodologies for verification and validation of space launch system
(SLS) structural dynamic models, NASA CR-2018–219800, 1, 2018. 124
[8] N. Metropolis and S. Ulam, The Monte Carlo method, Journal of American Statistical
Association, 44(247):335–341, 1949. DOI: 10.2307/2280232. 134
[9] J. Nelder and R. Mead, A simplex method for function minimization, Computer Journal,
7(4):308–313, 1965. DOI: 10.1093/comjnl/7.4.308. 134
[10] Load analyses of spacecraft and payloads, NASA-STD-5002, 1996. 137
[11] W. Iwan, On a class of models for the yielding behavior of continuous composite systems,
Journal of Applied Mechanics, 89:612–617, 1967. 144
 147

Author’s Biography

ROBERT N. COPPOLINO

Robert N. Coppolino received his formal education in New York City having graduated from
Stuyvesant High School and earned a B.S. Aerospace Engineering, and an M.S. and Ph.D. in
Applied Mechanics at the Polytechnic Institute of Brooklyn. As a structural dynamics engineer
at Grumman Aerospace Corporation (1967–1975), he supported the Lunar Module, Skylab,
and Space Shuttle programs, and contributed to development of techniques for launch vehicle
pogo (instability) suppression and propellant tank hydroelastic modeling and test-analysis corre-
lation. From 1975–1983, Dr. Coppolino was at The Aerospace Corporation, ultimately holding
the post of manager of the Engineering Dynamics Section. While at Aerospace, he supported
NASA’s Space Shuttle Pogo Integration Panel, developed the Simultaneous Frequency Domain
(SFD) method for experimental modal analysis, and defined innovative residual vector tech-
niques the were successfully employed in Space Shuttle nonlinear payload interface flight loads
predictions and detection of structural damage on offshore jacket platforms. From 1983–1987,
he served as manager of the Advanced Methods and Development branch at the MacNeal–
Schwendler Corporation (MSC). During his stay at MSC he led a consortium composed of
MSC, the J. S. Bendat Company, and Synergistic Technology Incorporated, which developed
technologies that were ultimately called the Integrated Test Analysis Process (ITAP). From 1987
to the present, Dr. Coppolino has held various senior posts at Measurement Analysis Corpora-
tion, currently serving as Chief Technology Officer for Torrance Operations. During this period
he collaborated closely with the late Julius S. Bendat on development of nonlinear spectral tech-
niques for identification of nonlinear systems. In addition, Dr. Coppolino continued develop-
ment of ITAP components including efficient procedures for modal test-analysis reconciliation.
The full suite of ITAP capabilities experienced its first end-to-end, “prime-time” application
on the International Space Station P5 Short Spacer modal test, conducted at NASA/MSFC
in 2001.
In 2016, Dr. Coppolino was invited to join the NASA Engineering Safety Center, Loads
and Dynamics Technical Discipline Team (LDTDT) as an industry member. Since that time, he
has provided independent review of NASA’s SLS and Orion programs, and as part of this work
developed procedures aimed at dealing with the “many modes” problem. A highlight of these
activities is development of SFD-2018, which appears to provide the impetus for a Roadmap
for a Highly Improved Modal Test Process.
148 AUTHOR’S BIOGRAPHY
Robert Coppolino is married to Catherine (Stafford) Coppolino, a person of outstanding
courage and love for others. Together they have five children, Michael, Melisssa, Kenneth, Peter,
and James, who have blessed them with seven wonderful grandchildren, Hannah, Hailey, Jakob,
Robert, Ryder, Drew, and Tyler.