Free-Form Surface Models Generation Using Reverse Engineering

Technique - An Investigation
M. Salman A. Mansor 1,2

1School ofMechanical, Aerospace and Civil Engineering, Faculty ofEngineering and Physical Sciences,
The University ofManchester, PO Box 88, Manchester M60 1QD, United Kingdom.
2School ofMechanical Engineering, Universiti Sains Malaysia, Engineering Campus, 14300 Nibong Tebal,
Seberang Perai Selatan, Pulau Pinang, Malaysia.
Email: [email protected]@yahoo.com

Abstract
This paper presents an investigation ofreverse engineering technique to generate free-form surface! models
in a computer-aided design system from physical objects. This technique involves data acquisitionjrom the
physical objects, registration of the data acquisition, data pre-processing, polygon mesh, segmentation,
surface fitting and surface model generation. From surface modelling perspective, the applicationS of this
technique can be widely used in many areas ofresearch and industry such as surface models generationfor
space suits in aerospace industries, prostheses in medicine, antique parts in archaeology, car blJdies in
automotive industries, turbine blades in nuclear power generators etc. Furthermore, the surjacemodels
generated can be used thereafter for engineering analysis, inspection in metrology, solid modelling,: design
and manufacture ofnew products. The advantages ofemploying this technique lead to reduction ofproduct
design times, easiness ofproduct design modifications and rapidproduct developments.

Keywords: Surface modelling; Reverse engineering; Computer-aided design; CAD.

1. Introduction

In modern manufacturing industries, a new product is generally manufactured according to the

computer-aided design (CAD) model that has been created by engineers. This process is known
as conventional engineering [18,20,23,25]. The opposite of that is termed as reverse engirteering
[7, 18,23,25] which is the process of existence product (i.e. physical object) is transformeq into a
CAD model. Several example of the physical object that can be used for reverse engineer~ng are
such as casing of mobile phone, human bone, train body, aeroplane compartment, turbine blade,
teapot etc.
In practice, a CAD model is designed as a solid or surface model which is depends on the
manufacturer requirement. A solid model can be created directly by using solid modeller tool or
can be produced by merging a complete set of surface models. To create a solid model fr~m the
surface models, a closed volume model must be formed by the surface models befor~ it is
converted to the solid model. Since the surface models are independently employed to create the
CAD model, hence only the surface models are concentrated in this work. On top of that, the
surface models generation by using a reverse engineering technique is focused because reverse
engineering is a fairly new technology that recently becomes very attractive and it is incredibly
demanded in the latest manufacturing world. .
From surface modelling perspective, the applications of reverse engineering can be Widely
used in many areas of research and industry such as surface models generation for space s~its in
aerospace industries, prostheses in medicine [13,24], antique parts in archaeology, car bodies in
automotive industries [21], turbine blades in nuclear power generators [8] etc. Furthermote, the
surface models generated can be used thereafter for engineering analysis, inspection in metrblogy,
solid modelling, design and manufacture of new products. The advantages of emplo~ this
technique lead to reduction of product design times, easiness of products design modifiqations
and rapid product developments. .

379
 As an impact and awareness of all the circumstance that have been mentioned before, hence an
objective of this work has been setup i.e. to investigate a versatile reverse engineering technique
that has been used for generating free-form surface models from physical objects. This
investigation is concentrated on basic principles, common issues and benefits of using the reverse
engineering technique. Thus, the investigation of the reverse engineering technique is presented
in Section 2 and finally it is followed by Section 3 for the conclusions.

2. An investigation into reverse engineering technique

From surface modelling point of view, the main purpose of reverse engineering is to redesign
existence surface of a product so that similar or new surface models of the product can be
reproduced. Traditionally, reverse engineering technique that has been used for surface model
generation requires a lot of manual process by experienced engineers. For instance, a combination
of ruler, measuring tape or calliper is required to measure dimensions of a physical object for data
acquisition. Then, the obtained data will be used in conventional engineering way to create the
surface model in any CAD systems. This technique gives disadvantages which are lack of
accuracy and uncertainty because of human interaction, it is very time consuming to acquire the
data and the work is very tedious especially when dealing with free-form surfaces.
Therefore, new techniques of reverse engineering have been introduced few years ago in order
to overcome the disadvantages. In principle, a complete task for reverse engineering system
consists of data acquisition, registration, data pre-processing, polygon mesh, segmentation,
surface fitting and surface model generation [19, 23, 25]. To implement the reverse engineering
system, combination of hardware and software are required. Currently, there is a lot of hardware
and software are being developed by researchers and commercialised by well-established
companies to cooperate with reverse engineering systems. For examples, the available hardware
for reverse engineering systems such as 3D laser scanner [5], Magnetic Resonance Image (MRI),
Computer Tomography (CT) scanner [17, 24], Coordinate Measurement Machine (CMM) with
mechanical probe [7, 9] etc. While the available software for reverse engineering systems such as
Paraform, CopyCAD, RapidForms, Geomagic Studio etc. The consequence of that, reverse
engineering nowadays become more versatile technology in the advanced manufacturing world.
However, the latest techniques to generate surface model are some how in many different
ways. For example, techniques to acquire data from the physical object requires hardware which
can be obtained by using contact method, non-contact method or a hybrid of contact and non-
contact method [19, 25]. Another example is the free-form surface model can be represented by
parametric form surface representations such as Non-Uniform Rational B-Spline (NURBS), B-
Spline or Bezier surface depend on the software whether it has been employed these type of
surface representations or not. Therefore, a versatile technique that can be used for surface
modelling is investigated and presented as follow:
In this technique, there are six stages will be investigated based on the principle that has been
mentioned earlier i.e. (i) data acquisition, (ii) registration, (iii) data pre-processing, (iv) polygon
mesh, (v) segmentation, (vi) surface fitting and surface model generation. All the stages will be
described in Section 2.1 to 2.6 including the related basic principles and some common issues
regarding each stage are highlighted.

2.1. Data acquisition

Data acquisition, which begins the process of reverse engineering, requires hardware which is
generally referred as a 3D scanner to scan the physical object. The scanning methods can be
classified into two i.e. contact (e.g. CMM with mechanical touch probe) and non-contact (e.g. 3D
laser scanner, MRI, CT scanner, ultrasonic, digital camera etc). The contact methods involve
mechanical contact between the physical object and the probe during the scanning operation.

380
While non-contact methods, there is no mechanical contact between the physical object and probe
during the scanning operation. A good overview and classification of 3D scanners can be ~ound in
the works of Motavalli [19] and Varady et al [25]. Compared to contact methods, non.contact
methods have short scanning time, can obtain thousands of measurement points simultarieously,
can capture the complete data point of the component and have long measurement range ~ut they
are less accurate (0.3 mm compared to O.Olmm for contact method), more error prone when the
physical objects have shiny surfaces and are more costly. Therefore, a laser scanner Is more
preferable for this technique because in stead of vast capability from the comparison mentioned
before, it is also widely used in industries [5], it is relatively a small device, easy to handll'l it and
it has more flexibility which can attach to robot arm or CMM so that it can have freedom of
movement to scan the object from multiple views. '
Clark [11] and Varady et al [25] have identified common issues that occur in data acq~isition.
These include accuracy of data point, accessibility of the scanner, occlusion due to shadowing,
obstruction or fixturing of the physical object, effect of object's surface such as shining object.
Some solutions to these issues have been reviewed by Varady et al [25] such as calibrating any
sensing device in order to get accurate data, use of multiple scanning devices and multiple views
to solve problems arising from accessibility and occlusion. Clark [11] suggests coating th!;l target
object in powder or paint to prevent high reflectivity of optical light from shiny object~ which
will generate noisy data. .

2.2. Registration

Since the data are acquired from different views, the data need to be registered in order to
merge them into a complete set of points in a single view. To achieve that, a single view (i.e.
fixed view) represents one set of points and the other sets of points which are represented as
multiple views (i.e. floating view) have to be merged. This process involves transformation sets
of data from the floating views to the set data of a fixed view [9]. A set of points coordinate Ph in
a floating view b can be converted to the equivalence set of points coordinate Pain a fixed view
a by using equation 1.
Po = p. T M (1)
where T is the transformation matrix for a translation set of points from b to a, while 'M is a
rotational matrix about an arbitrary axis and can be obtained by a translation set of points $ to the
origin followed by rotation Rx, Ry and Rz (with the angles of a, f3 and () respectively) aboUt the x,
y and z axes and finally inverse transformations Ry-I, Rx- I and Sl. Thus M can be written as
equation 2.
M = 8 Rx(a) Ry(P) Rz(8) Ry-I (p) Rx- I (a) 8-1 (2)

2.3. Pre-processing

After registration, pre-processing takes place to remove the noisy points and remain the: points
required. To achieve that, noise in the acquired data is filtered. It is important to remove th¢ noisy
points to prevent error in the surface generation in the subsequent stages. The noisy points Ican be
removed manually if the noisy data is already known. Otherwise, there are different filtering
approaches that can be used presented by Varady et al [25] such as dimensionality ~ltering
applied by Benko et al [3] for noisy data.

2.4. Polygon mesh

For polygon mesh, the points from the pre-processing stage are joined together to form a
triangular mesh. Polygon mesh is done to produce a rough surface of the outer skin defin~ng the
shape of an object. This polygon mesh is shown in form of facets after joining the points. In order

381
to generate the polygon mesh, several triangulation algorithms can be used, the most known of
which is Delaunay triangulation [3, 6, 26]. Two ways to apply Delaunay triangulation are the
Voronoi and the circumcircle approaches. The first approach begins with generating the Voronoi
diagram which is also known as Direchlet tessellation or Theissen tessellations of the points
acquired. The Voronoi diagram will have only one point in the middle of each polygon. The
Delaunay triangulation is then created by connecting all points. In the circumcircles approach,
circumcircles that do not contain any other points in their interior are created through three points.
The Delaunay triangulation is then created by joining the three points from the circumcircles.
While scanning a complex shape of physical object for the data acquisition, it may sometime
miss out some part of the scanned object. Thus the missing scanned data is arises a problem of
holes in the polygon mesh generated. If this problem is not solved at this stage, it will affect the
result in the next subsequent stages and definitely will generate a poor surface model which
contains the holes. Recently, one ofthe available solutions to solve this problem was presented by
lun [12]. He has proposed a methodology that can be used to fill the holes automatically with a
piecewise manner.
Since the polygon mesh is shown in form offacets, hence a stereolithography (STL) file format
can be generated which is the standard file used for rapid prototyping. Typically, the STL format
contains only the coordinates of triangle vertices and the normals of triangle facets. For instance,
Uu et al [1] was employed the STL format within their work for integrating a system of cross-
sectional imaging based reverse engineering and rapid prototyping for reproducing complex
objects.

2.5. Segmentation

Next, segmentation is done to divide the polygon mesh into surface patches. Segmentation is
carried out to partition the original point set into patches, one for each natural surface, so that
each patch contains just those points sampled from a particular natural surface. Basically, there
are two different methods that can be used for segmentation Le. face-based and edge-based
methods [2, 25]. Face-based methods infer connected regions of points with similar properties as
belonging to the same surface patch. For example, a group of points having the same normal
(using the normal to the triangles) is belongs to the same plane. Then, edges of each patch
(boundary of the patch) can be derived by intersection from the surface patches. While edge-
based methods, boundaries are determined in the point data representing edges (either sharp or
smooth) between surface patches. For instance, if sharp edges are being sought, places where the
normal of surface patch estimated from the point data suddenly change the direction are
identified. If smooth edges (tangent-continuous) are being considered, places where surface
curvatures or other higher derivatives have discontinuity are identified.

2.6. Surface fitting and surface model generation

In order to fit a surface to a set of points, several methods have been proposed [4, 7, 9, 10, 14,
15, 16,22,23], one of which is the least square method. Various types of surface models are used
to model the segmented data. To model standard engineering surfaces (i.e. analytical surfaces)
such as cylinder, conical and others, implicit polynomial functions are used. While to model free-
form surfaces, parametric representations such as NURBS, B-Spline and Bezier surfaces are used.
For fitting a parametric surface, the least square method is defined as equation 3.
min~) (f.(u i , w,),f/u p w,),f.(u" W,))T - Pi 11 2 (3)
;"'1

where Ix,/y andlz are functions defining x, y and z coordinates in terms of parameters u and w,
while n is the number of points and Pi is the actual data points from a scanning operation. For
example, when the type of surface is a Bezier surface,Ix,/Y andlz can be expressed as equation 4.

382
 F(u,w)= t!Bim(u) Qij Bjn(w) (4)
i_O / .. 0

where F(u, w) = {fx(u,w),fy(u, w),j.(u, w)}, m+l and n+1 are the number of control points, lJ;m(u) is
blending function in u direction Le. (0 :5 u:5 1), Qij position of the control points and Bj~(w) the
blending function in w direction Le. (0 :5 w :5 1). To fit a bicubic Bezier patch, equation 3 can be
written as equation 5.
N

E(F) = LIIF(u,w) -
;=0
pJ (5)
where the function E of Q/l, Q12, .. " Q44 will have a minimum when 8ElaQij = 0 for i = 1,2,3,4
and j = 1,2,3,4. By solving that, all the control point Qij can be determined. To simptify the
calculation, the values of u and w, which are input values, are assumed to have equal increments
and the grid lines coincide with scan paths. Thus if there are 11 grid lines in the u direction, the
values of u will be 0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9 and 1. The assumption that q and w
values are equally spaced is based on a further assumption that the spacing between the ~canned
data is uniform and the slope of the surface does not change drastically. Therefore, once the
control points are obtained and the grid are determined, hence the x, y and z coordinates for every
point on Bezier patch can be determined by using equation 4, where F(u,w) = {fx(u,w),{y(u,w),
!z(u,w)} = {x,y, z}. As a result, a surface model with Bezier representation is generated.
Unfortunately, the local control of surface shape is not possible for the Bezier surfaqe [23].
Unlike B-Spline and NURBS surfaces, the Bezier surface does not have ability to add control
point without increasing the degree of the surface. Consequently, it limits the availability of the
surface fitting optimisation and also because of that ability, B-Spline and NURBS have been the
standard for representing free-form surfaces in current commercial CAD systems. However, the
NURBS surface is differs from B-Spline surface because it has a weight and denominator to
represent smoother surface. Similarly, the basic concept presented before can be applied: for B-
Spline or NURBS surface. Lin et al [16] employed an existence B-Spline surface fitting as
comparison to their method i.e. surface lofting method. Their method begins with fitting e~h row
of data points to obtained B-spline curves and then the section curves are fitted into 1Ii lofted
surface. The B-Spline surface,fx,J;, and!z used can be expressed as equation 6.
B(u, w) = !! Pij NiP (u) N jq
;=0 j_O
(w) (6)

where B(u,w) = {fx(u,w),J;,(u,w),!z(u,w)}, Pij are position of the control points, m+l and n+1 are
the number of control points, N;p(u) and ~q(w) are the normalized B-Spline basis functions of
degree p and q in two parametric directions which are in u i.e. (0:5 u :5 I) and w i.e. (0:5 w :5 1)
directions respectively. While Chivate et al [10] employed NURBS surface fitting using a least
squares data approximation method in which the NURBS surface, fx,/y and!z can be expressed as
equation 7.

S(u w) =
!! vJ)ijN,p (u)N
i.O j.o
Jq (w)
(7)
, !!vifNiP(u)Njq(w)
jaO j.a

where S(u,w) = {fx(u,w),J;,(u,w),!z(u,w)}, vij represent the weight, Pij are position of the ~ontrol
points, m+1 and n+1 are the number of control points, Mp(u) and N.iq(w) are the normalized B-
Spline basis functions of degree p and q in two parametric directions which are in u i.e. (0 :5 u :5
1) and w i.e. (0 :5 w :5 1) directions respectively. Since NURBS surface has more advantages,
therefore it is more favoured to be used for this technique.

In addition, the different between this technique and the others are one or more stages have
been skipped except data acquisition, surface fitting and surface model generation. The
consequent of that is the generated surface model may not have a very good quality, esp~cially

383
when dealing with a complex physical object which requires segmentation to define the patches.
Also, there are feedbacks from each stage so that some of the stages can be repeated to edit and
modify previous results such as point cloud, polygon mesh, segmentation or surfaces model
generation. For example, a feedback to data acquisition may be necessary if a surface fitting stage
indicates that there are holes, error or noisy points in the data set.

3. Conclusions

Although there is a number of reverse engineering techniques currently available in the

advanced manufacturing world, this investigation has given an attention to a versatile technique
of reverse engineering for generating free-form surface models from physical objects.
To have a very good reverse engineering system, the system must have all the presented stage
Le. data acquisition, registration, data pre-processing, polygon mesh, segmentation, surface fitting
and surface model generation.
The output from polygon mesh Le. data in STL file format can be used for rapid prototyping
which leads to the rapid product development.
Whilst the result from surface fitting and surface model generation can be used for surface
modelling and solid modelling which thereafter the design of products can be easily modified.
Then, it can be used for other purposes.
Since the surface models can be generated from the physical objects by obtaining the digitised
data rapidly, hence, the product design times can be definitely reduced.

References

[1] Liu Z, Wang L, Lu B. Integrating cross-sectional imaging based reverse engineering with
rapid prototyping. Computers in Industry 57 (2006) 131-140.
[2] Woo H, Kang E, Wang S, Lee KH. A new segmentation method for point cloud data.
International Journal of Machine Tools & Manufacture 42 (2002) 167-178.
[3] Benko P, Marin RR, Varady T. Algorithms for reverse engineering boundary representation
models, Computer -Aided Design 33 (11) 839-851, 200l.
[4] Bolle RM, Vemuri BC. On Three-Dimensional Surface Reconstruction Methods. IEEE
Transaction On Pattern Analysis And Machine Intelligence 13 (1) (1991) 1-13.
[5] Son S, Park H, Lee KH. Automated laser scanning system for reverse engineering and
inspection. International Journal of Machine Tools & Manufacture 42 (2002) 889-897.
[6] Chen L-C, Lin GCI. A Vision-aided reverse engineering approach to reconstructing free-
form surfaces. Robort & Computer-Integrated Manufacturing 13 (4) (1997) 323-336.
[7] Chen L-C, Lin GCI. An integrated reverse engineering approach to reconstructing free-form
surfaces. Computer Integrated Manufacturing Systems 10 (1) (1997) 49-60.
[8] Chen L-C, Lin GCI. Reverse Engineering in the design of turbine blades - a case study in
applying the MAMDP. Robotics and Computer Integrated Manufacturing 16 (2000) 161-
167.
[9] Chivate PN, Jablokow AG. Solid-model generation from measured point data. Computer-
Aided Design 25 (9) (1993) 587-600.
[10] Chivate PN, Puntambekar NV, Jablokow AG. Extending surfaces for reverse engineering
solid model generation. Computers in Industry 38 (1999) 285-294.
[II] Clark J. Implementing Non-Contact Digitisation Techniques in a TCT Environment. TCT
conference (1999).
[12] Jun Y. A piecewise hole filling algorithm in reverse engineering, Computer-Aided Design 37
(2005) 263-270.
[13] Lin YP, Wang CT, Dai KR. Reverse engineering in CAD model reconstruction of
customized artificial joint. Medical Engineering & Physics 27 (2005) 189-193.

384
[14] Lancaster P, Salkauskas K. Curve And Surface Fitting, An Introduction. Academic Press Ltd
(1986).
[15] Lin AC, Lin S-Y, Fang TH. Automated sequence arrangement of 3D point data for surface
fitting in reverse engineering. Computer in Industry 35 (1998) 149-173. .
[16] Lin C-Y, Liou C-S, Lai J-Y. A surface-lofting approach for smooth-surface reconstruction
from 3D measurement data. Computer in Industry 34 (1997) 73-85. .
[17] Liu S, Ma W. Seed-growing segmentation of3-D surfaces from CT-contour data. C~mputer­
Aided Design 31 (1999) 517-536.
[18] Mills BI, Langbein FC, Marshall AD, Martin RR. Approximate Symmetry Deteqtion For
Reverse Engineering. Sixth ACM Symposium on Solid Modelling and Applications, Ann
Arbor, Michigan, June 4 - 8, 2001.
[19] Motavalli S. Review of reverse engineering approaches. Computers Industrial Engineering
35 (1-2) (1998) 25-28.
[20] Peng Q, Loftus M. Using image processing based on neural networks in reverse engIneering.
International Journal of Machine Tools & Manufacture 41 (2001) 625-640.
[21] Sansoni G, Docchio F. Three-dimensional optical measurements and reverse engineering for
automotive applications. Robotics and Computer-Integrated Manufacturing 20 (2004) 359-
367.
[22] Soderkvist I. Introductory Overview of Surface Reconstruction Methods. Lulea University of
Technology, Department of Mathematics, Research report 10 (1999).
[23] Tai C-C, Huang M-C. The processing of data points basing on design intent in reverse
engineering. International Journal of Machine Tools & Manufacture 40 (2000) 1913-'1927.
[24] Viceconti M, Zannoni C, Pierotti L. TRI2S0LID: an application of reverse eng~neering
methods to the creation of CAD models of bone segments. Computer Methods and Programs
in Biomedicine 56 (1998) 211-220.
[25] Varady T, Martin RR, Cox J. Reverse engineering of geometric models - an introduction.
Computer-Aided Design 29 (4) (1997) 255-268.
[26] Zeid I. CAD/CAM Theory and Practice. McGraw-Hill, Inc (1991) 932.

385