Scribd
Upload
33 views

Lect Slides - Development of Empirical Models From Process Data

This document discusses methods for developing empirical models from process data to describe dynamic systems. It covers fitting first and second order models to step response data using graphical techniques. Methods like the tangent line method and Sundaresan & Krishnaswamy method are described to estimate parameters of first order plus time delay models from experimental step responses. The document also discusses fitting integrator models and using other approaches like regression and neural networks for model development. Examples and problems are provided to illustrate the different modeling techniques.

Uploaded by

Ayush Nagar
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
33 views

Lect Slides - Development of Empirical Models From Process Data

This document discusses methods for developing empirical models from process data to describe dynamic systems. It covers fitting first and second order models to step response data using graphical techniques. Methods like the tangent line method and Sundaresan & Krishnaswamy method are described to estimate parameters of first order plus time delay models from experimental step responses. The document also discusses fitting integrator models and using other approaches like regression and neural networks for model development. Examples and problems are provided to illustrate the different modeling techniques.

Uploaded by

Ayush Nagar
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 23

BITS Pilani

K K Birla Goa Campus

Development of Empirical Models from


Process Data
Topics to be covered

• Introduction to Black box (empirical models)


• Fitting first and second order models using step
tests
• Graphical techniques
• Fitting integrator model to step response data

21/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


Empirical (Black box)models
(Process identification)
• Process (Theoretical models) simulation vs Process
(system) Identification (empirical models)

• Simulation
• Model is known (from balances)
• Response (y(t) is then generated

• Identification
• Model is determined from
actual experimental data
3

21/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


Empirical models

• Steady state models


• Can be used for Instrument calibration, process optimization,
specific instances of process control

• Dynamic models
• Can be used for understanding the process behaviour during
upset conditions

21/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


Fitting first order model using step
tests
• Process reaction curve
• Output response of a process to a step change in input

• For the process to be approximated by a first or second


order linear model, model parameters need to be
obtained by inspecting process reaction curve

• First order model

• For step response


5

21/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


Fitting first order model using step
tests
• Graphical construction to estimate the model
parameters (Gain and time constant)
• Gain (K)
• Ratio of steady state output
to change in input

• Time constant (τ)


• Slope of a tangent line
(at t = 0)
• 63.2% method

21/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


Example 7.2

21/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


Limitations of approximating to first
order process
• Only very few experimental step responses exhibit first
order behaviour
• Output data are corrupted with noise
• Process input may change in an unknown manner
• Difficult to generate perfect step input

Departures from ideal response curve (Fig 7.3) are


common

To account for higher order dynamics, a time delay


term can be included
8

21/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


FOPTD model
• Fitting of FOPTD (First order process with time delay)

Parameters
• K = ratio of steady state change
in y to input step change M
• Intersection of Tangent line at
point of inflection and time axis
 Time delay
• Intersection of tangent line with
steady state response line
(y=KM), corresponds to time
(t=θ + τ)  Time constant can be
calculated 9

21/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


FOPTD model
• Tangent methods – uses single point to estimate time
constant
• Use of several points (i) from response can also be used to
determine time constant

• Semi-log response of output vs (ti – θ) will be a straight line


with slope (-1/τ)

• Major disadvantage of FOPTD model – It is difficult to find


the point of inflection (due to noise, etc.) 10

24/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


Sundaresan and Krishnaswamy
(S&K) method
• It avoids using point of inflection construction

• Two times t1 and t2 estimated from step response curve

• t1  35.3 % of response

• t2  85.3 % of response

• Time delay and time constant are calculated as

11

24/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


Graphical techniques for second
order models
• Fitting second order model to the data – relatively better
approximation

• Limiting cases

τ2/τ1 = 0  first order

τ2/τ1 = 1  critically
damped case
12

24/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


Smith’s method

• Times at which response reaches


20% and 60%

• Using Fig. 7.7, value of damping


coefficient and time constant
can be determined

13

24/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


Example 7.3

14

24/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


Fitting an integrator model to step
response data
• First order process to a step change

• For short times (t < τ), the exponential term can be


approximated

• Response

15

24/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


Fitting an integrator model to step
response data
• Step response of the integrating element

• Response (in time domain)

• First order model can be approximated as an integrating


element with a single parameter

Ramp-like response to a step change in input 16

24/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


Fitting an integrator model to step
response data
• Approximate short term
response to FOPTD
model

17

24/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


Other methods of model
development
• Development using linear and non-linear regression

• Neural network model

18

24/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


Using Excel Solver to Fit Transfer
Function Models
• use y (data) vs. y (predicted)
• column 1 is data (taken at different times), or y1
• column 2 is model prediction (same time values as
above), or y2
• target cell is S (y1 - y2)2 , to be minimized
• specify parameters to be changed in reference cells
(e.g. t1 = 1, t2 = 2)
• open solver dialog box to check settings
• click on < solve > (calls optimization program)
19

24/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


Problem 7.1

20

24/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


Problem – 7.2

24/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


Problem – 7.4

24/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus


Problem – 7.7

24/02/2020 AMOL DESHPANDE BITS Pilani, K K Birla Goa Campus

You might also like