Medical Engineering & Physics 27 (2005) 189–193

Technical note

Reverse engineering in CAD model reconstruction

of customized artificial joint
Yan-Ping Lina,∗ , Cheng-Tao Wanga , Ke-Rong Daib
a School of Mechanical and Power Enginnering, Shanghai Jiaotong University, No. 1954 Huashan Road, Shanghai, China
b Department of Orthopaedics, Shanghai Ninth People’s Hospital, Shanghai, China

Received 8 October 2003; received in revised form 6 September 2004; accepted 17 September 2004

Abstract

In order to meet the requirements of customized artificial joint design, and to reduce the production cycle and cost, we present a method to
generate the complex surface of an artificial knee joint by co-ordinate measuring machine (CMM) from the normative prosthesis, and form
the model data base. First, this paper describes how to plan the measure method to get the better data points and how to deal with the point
cloud data. Then, the free-form surfaces are constructed from the point cloud data using the reverse engineering software—Surfacer. Lastly,
the solid CAD model of the artificial knee joint is created from the surfaces by extension, intersection and so on. These models formed the
data base of the prosthesis, in which we can select a suitable kind of artificial knee joint model to customize for the patient. That is, we only
need to change the local data of the corresponding CAD model to meet the different requirements of the patient.
© 2004 Published by Elsevier Ltd on behalf of IPEM.

Keywords: Artificial joint; Reverse engineering; CMM; Surfacer; CAD

1. Introduction Reverse engineering (RE) is an important branch of

Total knee arthroplasty (TKA) is an effective method to
cure late osseous arthritis, atrophic arthritis and other types
of arthritis. It can alleviate the patient’s pain effectively and
improve the knee joint function greatly. More and more
TKA cases occur with the enhancement of science and tech-
nology and the improvement of the quality of life. More-
over, CAD/CAM have been applied in clinics along with the
progress of computer technique. Designing the customized
prostheses that best matches the human anatomy, accord-
ing to different patient’s requirements, is becoming current
[1]. We use the free-form surface of the normative artifi-
cial knee joint, and apply reverse engineering in the cus-
tomized design. By doing this, we reduce the period and
the cost greatly. According to the patient’s situation, we
can manufacture the customized artificial knee joint in 3–7
days.
∗ Corresponding author. Tel.: +86 21 629 32905.
the mechanical design and manufacture application field,
and this technique has been widely recognized as a cru-
cial step in the product design cycle. In a normal auto-
mated manufacturing environment, the operation sequence
usually starts from product design via computer-aided de-
sign (CAD) techniques, and ends with generation of ma-
chining instructions required to convert raw material into a
finished product. In contrast to this conventional manufac-
turing sequence, reverse engineering represents an approach
for the new design of a product that lacks an existing CAD
model.
In the process of the product design and research, the
use of RE will largely reduced the production period and
costs. Artificial knee joints used in the joint replacement
surgery mostly are mostly imported products, which are ex-
pensive and cannot meet the customized requirement. Cus-
tomized design customizes the products depending on the
patients’ different demands to design the artificial joint with
suitable dimension, such as the length and the size of the
stem. The artificial knee joint has a particular shape and

1350-4533/$ – see front matter © 2004 Published by Elsevier Ltd on behalf of IPEM.
doi:10.1016/j.medengphy.2004.09.006
190 Y.-P. Lin et al. / Medical Engineering & Physics 27 (2005) 189–193
Fig. 1. The process of reverse engineering.
complex surface, and the form of the surface indirectly af-
fects its work performance. Only after a great deal of de-
signs and experiments, can a better surface come forth. In
this paper, we employ RE technology to gain the complex
surface, then change part of the data to suit for the patient’s
requirement. In this way, we greatly reduce the production
period.
The typical process of RE begins with digitization from
the surfaces of an existing part (Fig. 1). The CAD model
of this part is then created by smoothly fitting these digi-
tized data points. Digitization of a part surface in RE can be
achieved by utilizing either contact probing or non-contact
sensing techniques [2]. A coordinate measuring machine
(CMM) represents a typical example of contact probing de-
vices that incorporate highly precise linear transducers and a
microcomputer-based controller to provide an accurate, sen-
sitive and reproducible indication when the part surface is
touched. Contact probing devices are generally more accurate
but slow in data acquisition, and vice versa for non-contact
type devices. Non-contact devices, such as a laser scanning
machine, CT scanning, MRI, and CGI, are recently devel-
oped with the wide use of optics and electron elements. Non-
contact sensing techniques project a laser beam onto the part
surface and the reflected beam is received by optical sensor.
By applying a triangulation procedure, the 3-D coordinates
of data points on the part surface are calculated at a consid-
erably greater speed compared to contact probing techniques
[3,4]. As laser sensing is sensitive to shiny and dark surfaces,
other techniques may be required, such as complicated light-
ing techniques, and spray coating. For these reasons, contact
probing devices, such as CMM, are widely accepted as tools
for accurate digitization of a part’s surface in reverse engi-
neering even though they require comparatively protracted
measurements.
2. Materials and methods
2.1. Digitization from the surface
In this paper, we introduce the IOTA0101DH/T-P CMM
machine made by Institute of Aviation Precision Mechan-
ics of China. It has the low system error and high measuring
precision, orientation precision is 3.5 ␮m, and measuring pre-
cision is 4 ␮m at best.
The CMM scans the surface from one point to another
point, namely by picking the points through the scanning
line. The scanning line is the intersection of the scanning
plane and the workpiece (shown in Fig. 2). There are two
Fig. 2. A sketch of the CMM scanning process.
main scanning processes: radial scanning and axial scan-
ning.
Measurement path planning is concerned with determin-
ing the positions of measurement points, the directions of
probing vectors and a collision-free path. In order to achieve
the best point clouds, it is important to plan the measurement
path properly. In this project, the most important surface is
the free-form surface such as the inside and outside condy-
lus articular surface, the patella sliding chute surface and the
tibia table surface. Selection of probe orientation plays an
important role in the system, because improper probe orien-
tation may make a reachable point become unreachable. We
can adjust the CMM in different poses, at the same time re-
duce the step size to get more most points in this free-form
surface.
2.2. Preprocessing of the point clouds
Preprocessing of the point clouds is the key step of the
RE process, its result will directly impact the quality of the
model reconstruction. In order to improve the rationality and
integrity of the data, the point clouds generally need some
preprocessing such as registration of point clouds into one
coordinate system, and remove the outliers [5]. While the
CMM is measuring the artificial knee joint, we change the
measuring orientation in response to the feature of the ar-
tificial knee joint, measure the physical model at different
positions such as 0◦ , ±45◦ , ±90◦ , ±135◦ , ±180◦ , and so on.
We finish the data collection at one fixed position. Thereby,
the error, which comes from the registration of point clouds
into one coordinate system, is avoidable. Whether you adopt
the contact or non-contact probing devices, there always are
 Y.-P. Lin et al. / Medical Engineering & Physics 27 (2005) 189–193
Fig. 3. The point cloud before (a) and after (b) preprocessing.
some outliers in the original point clouds with the limitation
of the measuring tool and the measuring manner. These out-
liers are interferential to the reconstructed 3-D model. There-
fore, in the RE process, we should diagnose the point clouds
and remove the outliers to assure the accuracy and the pre-
cision. Many researchers have focus on removing outliers
and have obtained good results. One advantage of the CMM
measure is less outlier. In this paper, we remove the obvious
outliers through interactive action directly in view of conve-
nience, at the same time the accuracy is insured in this way
(Fig. 3).
2.3. Surface fitting
In this paper, we adopt Imageware Surfacer package to
reconstruct the 3-D model of an artificial keen joint. The
Surfacer allows modeling of a high quality surface model
with scanned-in point data from a physical model, and also
lets the user compare and modify CAD models with point
data, and quickly produces various design iterations. With the
digitized data, we can create the 3-D model by going from
point to curve to surface to solid, or point to surface to solid
directly, according to every surface feature. In the processing
of the points, curves and surface, we can use many kinds of
diagnosis methods to ensure the precision.
2.3.1. Point processing
Since the knee joints have complex contour surfaces, and
the whole knee joint model cannot be approximated by a
single mathematical surface, the preprocessed point clouds
generally need to be subdivided into many sections, each of
which can be fitted to a surface: plane, cylinder, sphere, cone,
free form surface and so on. Firstly, we divided the point
191
Fig. 4. Point cloud cross section by curved aligned.
clouds according to the feature into condylus and patella sur-
face blocks, inside/outside thighbone and condylus surface
block, condylus glenoid surface block, joint’s handle surface
block, and so on. Every block will be fitted to different free
form surfaces, these surfaces intersect or transit each other
to construct on solid, then the CAD model of the artificial
knee joint is formed. By using subdivided data, the problem
of dealing with the point cloud data can be simplified, the
precision can also be improved.
Generally, in order to fit the points to curves optimally, the
blocks that are in surface-state should be cut into many linear
point clouds. For example, in this paper we adopt the “curve
aligned” cross section method to divide the condylus and
patella surface block into many linear point clouds. In Fig. 4,
the base curve of the cross section operation is based on the
original condylus and patella surface, and it characterizes the
surface feature correctly. If the cut planes are aligned with
this curve, better linear point clouds are obtained to fit the
curves.
2.3.2. Curve processing
The object of RE is to create the surface or curve with
both high precision and very smooth. However, precision
and the smoothness are contradictory, and we must sacri-
fice one to improve the other. As a result, we can change
the number of control points and the degree of the curves
to modify the curves. By increasing the number of con-
trol points and the degree of the curves, more accurate
shapes can be generated. Contrarily, by decreasing, more
smooth shapes can be generated. Through adjusting the
number of control points and degree, we can get curves
with the satisfying precision and smoothness. Moreover,
there are many methods to edit and diagnose the partial
curve.
Every curve directly created by points always has differ-
ent curvature in its endpoint. So the surface has bad smooth-
ness when extended. In order to get satisfying precision and
192 Y.-P. Lin et al. / Medical Engineering & Physics 27 (2005) 189–193

Fig. 5. Construct a curve by fitting a line at the end of the curve.

smoothness after the surface extends, we use the method of

fitting the two ends of the point cloud with a line instead of a
curve (Fig. 5).
(1) Divide the cloud into three parts: D1, D2 and D3. D1 and
D2 contain a few points at the end of the cloud and will
be fit as a line, the others, D3, will be fit as a free-form
curve.
(2) Fit every part respectively to line, L1 and L2, and to curve
C1.
(3) Bridge the curves, between L1 and C1, L2 and C1, ac-
cording to the principle of curve tangency, to a whole
curve.
(4) Edit, modify and diagnose the curve to have better pre-
cision and smoothness.
In this way, we can obtain the curves that almost have the
same curvature in the endpoint. As a result, the corresponding
surface has the better smoothness, and the solid model is easy
to construct.
2.3.3. Surface processing
The knee joint requires high precision and smoothness
(Class A surface). In order to form a Class A surface the
surface must have consecutive curvature, and the surface’s
control points must be aligned very evenly. Therefore, it is
time-consuming work to create a Class A surface. During
the construction of the surface, we can modify and diagnose
it to be better dynamically and in real time. The difference
between the point cloud data and the created surfaces can
be shown using an error map in which the amount of devi-
ation is represented by colored regions. Fig. 6 is the diag-
nosis map between the point cloud data and the condylus
and patella surface. In this map, the maximal error value is
0.451 mm (individual) and most of the tolerance is less than
0.25 mm.
Fig. 6. Diagnosis map between point clouds and surface.
Fig. 7. The CAD model and the production by NC machining.
3. Results
After having created each surface, the following task is
to construct the knee joint solid according to these surfaces.
Generally, the surface fitting with the original point cloud data
is not big enough to construct the solid. Therefore, we need
extend or enlarge the surfaces. With the enlarged surfaces,
we can intersect them, cut them, facet them, sew them as a
whole surface and so on. Then, we can customize the knee
joint, such as the shape and dimension of the stem, according
as the patient’s requirement. Up to now, a 3-D solid of knee
joint is created. This 3-D CAD model will be manufactured
by numerical control machine to form the product, which
will be applied for surgery. Fig. 7 shows the CAD model of
the artificial knee joint and the production applied for clinic
medicine.
4. Discussion
The CAD model formed the data base of customized ar-
tificial joints applied in clinic medicine, in which we can
select a suitable kind of artificial knee joint model to cus-
tomize for the patient. The basic free-form surfaces, that is,
the inside and outside condylus articular surface, the patella
sliding chute surface and the tibia table surface, will be filed
as template prototype. For different patients, we just need to
change the local data of the corresponding CAD model to
meet the different requirements. For instance, according to
the position of the osteotome, the dimension and shape of
the patient joint, we can customize the length of the pros-
thesis stem, the size of the prosthesis to satisfy the patient
at most. For the same kind of prosthesis, we use the same
basic free-form surface, only change the contour to construct
different sized models. Therefore, through reverse engineer-
ing, we found the data base of different kinds of prosthesis
free-form surfaces, not only establish the basic theoretic base
for prosthesis clinic application, but also largely reduce the
production period and costs.
 Y.-P. Lin et al. / Medical Engineering & Physics 27 (2005) 189–193
Acknowledgements
The authors would like to acknowledge Shanghai Si-ai
High Technology Ltd. for their kind supports, Professor Feli-
cia Brittman for her kind edit. This research study was granted
by Shanghai Ninth People’s Hospital.
References
[1] Lv HS. Artificial joint surgery. Beijing: Science Press; 2001.
p. 268.
193
[2] Ke YL, Xiao YX, Li JX. The research of reverse engineering
CAD modeling technique. J Comput aid Design Graph 2001;13(6):
570–5.
[3] Gao GJ, Chen KN, Zhang SS. The technique research of re-
verse engineering of free-form surface. Mech Sci Tech 2000;19(1):
94–6.
[4] Paulo C-M, Tim K, Jim D. CMM verification: a survey. Measurement
1996;17:1–16.
[5] Woo H, Kang E, Wang S, et al. A new segmentation method
for point cloud data. Int J Machine Tools Manuf 2002;42:167–
217.