A W, J W: Lbert Eckenmann Ohannes Eickmann
A W, J W: Lbert Eckenmann Ohannes Eickmann
A W, J W: Lbert Eckenmann Ohannes Eickmann
University Erlangen-Nuremberg
Chair Quality Management and Manufacturing Metrology
Naegelsbachstr. 25, 91052 Erlangen, Germany
e-mail: weckenmann@qfm.uni-erlangen.de
1. INTRODUCTION
Modern sheet metal forming technology allows the production of highly sophisticated free
form sheet metal components, allowing great flexibility in design matters and manufacturing
processes across a wide range of industries. An example is the automobile industry where
engineers are constantly trying to realize more extraordinarily-looking car bodies.
Due to residual strains, the workpiece can deform after a deep drawing process. During
assembly the workpiece has to endure forces that can distort its shape according to its
structure. In order to set the workpiece in its assembled state, the workpiece has to be fixed or
clamped during the measurement process with conventional measuring systems (e.g. tactile
coordinate measuring systems). Therefore a fixation device in which the workpiece can be
measured must be designed. The inspection process is laborious, time consuming and some
actions (e.g. fixation/clamping) cannot be automated, [1]. For this reason 100% testing of all
parts in mass production is impossible. Only sampling inspection (statistical control) is
practicable with this conventional approach.
Only optical measurement procedures can satisfy the restrictive conditions (time and
automation) for an in-line measurement process. Nowadays optical coordinate measuring
systems are fast, suitable for automation and robust enough to be used in an industrial
environment. Systems that use fringe projection technique allow a fast, parallel and noncontact acquisition of point clouds of the workpiece surface. The fringe pattern is projected on
the distorted workpiece and one or more cameras acquire the data. Fringe projection systems
are based on the triangulation principle between the camera and projector. From the measured
point cloud a triangle mesh of the surface can be generated, which could be used in Finite
Element Method (FEM) simulations of the fixation or clamping process. By using this virtual
fixation method, the inspection process chain can be significantly shortened, reduced in time
(e.g. for the measurement of a car door from 1 day to a few minutes) and automated. The
measurement time amounts approximately to 10-15 seconds. All other steps can be performed
by a PC-system, meanwhile the workpiece can continue to run through the production
process. This method provides a 100%-test of the production with simultaneous increase of
control over sheet metal production and a decrease of inspection costs, [2], [3].
CCD Camera
Laser
x
z
tg() =
Reference
plane
In this research project the fringe projection system MacroSPS from GFM, consisting of two
cameras and a DMD projector (Digital mirror device, [6]), is used. It was designed for
measuring large sheet metal parts as used in the automotive industry.
Technical specifications:
Measuring field:
400 x 700 x 900 mm
Collected points:
1400 x 1800
Angle of triangulation:
23
Lateral resolution:
500 m
Transversal resolution:
200 m
Deviations according to VDI/VDE 2634
Distance between sphere:
< 620 m
Spherical shape:
< 126 m
Flatness:
< 225 m
Repeatability:
< 10 m
Contrast ratio:
500 : 1
Measurement period:
< 3 min
3. MEASURING PROCESS
Fig. 4. Process chain: The green boxes depict necessary information and tools.
As shown in Fig. 3, the workpiece is measured in distorted state without a fixation device.
Afterwards the data set (including information about holes an edges) is processed (see Fig. 4)
and with the information about fixation (e.g. position of holes) and material properties a FEanalysis representing the virtual fixation is performed. The last step is a comparison with
the CAD-model or the fixed workpiece. The workpiece is indispensable only for the
measurement (which lasts 15-30 seconds). The other steps are realised by a PC. There is no
need to stop production for testing a workpiece. Thus this method is more suitable for an inline test than a conventional tactile measuring system. The most important steps are described
in the following chapters.
4. DATA PROCESSING
Fringe projection systems deliver some millions of point coordinates (x, y, z) per
measurement as a so-called point cloud. To receive the results of the following steps in an
appropriate time, the amount of points has to be reduced to 30,000 50,000. The points of the
surrounding measuring field are deleted and the remaining point cloud is reduced by a
curvature based method where most points in flat areas are deleted and points in zones with
strong curvature are kept. Thus all geometrical properties are preserved. Because noise of the
measured surface would falsify the FE-simulation result, the maximum permissible
smoothing is elaborated (under the condition that surface properties are kept). This smoothing
is done by filters (e.g. median, Gaussian, [8]).
For creating the surface, a triangle mesh based on the point cloud is generated, Fig. 5.
There is no three-dimensional order contained in the list, for geometrical operations a
comparison of all points is required to find the point's neighbourhood relations. A linear
approximation to the surface is a triangle mesh, in which the space between the adjacent
points is approximated by triangles, Fig. 5. Methods like Delaunay triangulation allow the
automated determination of a unique triangle patch representing the measured surface [4].
The processing time for the realisation of the triangle mesh is approximately one minute.
A
B
n
C
5. VIRTUAL FIXATION
The measured, smoothed and diminished triangle mesh of the distorted workpiece is the
base for the FEM model. Each triangle is defined as a shell element with a given thickness.
Boundary conditions have to be identified and fixed when modelling the simulation [9].
Below, the boundary conditions for the example depicted in Fig. 6 are given:
The material properties are determined as follows: Youngs modulus E = 210,000 N/mm2,
Poissons ratio = 0.4 and density = 7.87 kg/dm2.
In the given example the average wall thickness for the whole workpiece is 0.8 mm.
Crucial boundary conditions are the Degrees of Freedom (DoF) at the fixation points
including impressed displacements, rotations, forces and moments. These points are taken
from CAD-models or technical drawings (e.g. holes) and transferred to the FE-model in a
realistic manner. Detection of holes and edges can be automated by the use of algorithms
[4] [10]. In the given example the following conditions were assumed: Points A1 A4
belong to the CAD-model, points B1 B4 belong to the virtual fixed data. Points A1 and
A2 coincide with the analogue points B1 and B2. For the points B3 and B4 there is a
displacement in the direction of the violet arrows. At the end of the simulation the points
B3 and B4 coincide with the respective A3 and A4.
In the presented example the workpiece was measured in relaxed, distorted state and in
fixed state. By virtual fixation that means removing the distortion virtually , the unfixed
state was transformed into the fixed state. At the end of the process a new file (e.g. .stl-file)
with a new form of the surface is obtained, which considers the boundary conditions. The
processing time depends on the model and varies between a few seconds and a few minutes.
B4
A4
B3
A3
implementation
implementation
A 1,B 1
A 2,B 2
The CAD-model and the virtual fixed shape (both are colour coded) are being overlapped.
Taking the surface of the CAD-model as reference, it is possible to characterize positive or
negative deviations based on the visualized colour in a the specific area of the work piece,
which is under consideration, Fig. 7.
The two data sets have two different coordinate systems and therefore different origins and
orientation. For this reason a transformation (= translation + rotation) of the workpiece
coordinate system is indispensable (Registration)[3]. The translation can be accomplished
in different ways:
With definite points: the matrix of the transformation is defined with the coordinates of
some characteristic points.
Best fit method: the CAD-model and the virtual compensated shape are overlapped in the
best possible way (Gaussian or minimum method).
The processing time for the transformation is approximately 35 seconds.
The crucial point of this work is a quantitative comparison of the virtual fixed shape with
the CAD-model (= reference). Figure 8 illustrates the method used:
Overlapping of the CAD-model and the virtually fixed model (consecutively called FEMprofile or FEM-model);
Section plane areas used for creating comparable contours of the two surfaces;
The comparison is made at critical points (e.g. at a radius or an edge) where the deviation
from a nominal position is calculated. Distances between holes can also be measured:
These distances can be compared with the nominal dimension, which is given in a CADmodel or an engineering drawing.
Fig. 8. Quantitative analysis: unfixed (blue), fixed (red) and virtually fixed (green) datasets.
Figure 8 shows as sample object a cylindrical formed metal sheet. The comparison
between the virtually fixed and the unfixed shape shows the success of virtual fixation, which
is carried out by means of FEM analysis. For this purpose the relative deviation of the critical
edge in area 1 to the corresponding reference area of the real shape in fixed state is calculated.
Maximum deviation from real shape in fixed state in the considered edge area 1 is reduced
from 15.88 mm to 1.20 mm. The average deviation in the depicted section was reduced by
virtual distortion compensation to a value of < 0.6 mm.
If the focus of the comparison lies on the overall shape of a workpiece, a holistic false
colour rendering is used. Below, a complex reinforcement structure of a car body is used as an
example, Fig. 9. Figures 10 and 11 show the success of virtual fixation, as the average
deviation from the reference CAD-model is reduced from 5.4 mm to 1.4 mm. Please note in
Figure 10 that the used reengineering software is unable to depict the whole compared
structure, as the deviation before performing virtual fixation is too large in some areas. Thus
the software is not able to assign all measured points to equivalent reference points. After
performing virtual fixation and distortion compensation (Fig. 11) the software can handle the
comparison of the whole shape.
CAD Model
Distorted shape
Virtually fixed shape
The X-crosssection in Fig. 12 shows that through virtual fixation the areas between the
fixation (or clamping) points converge towards the reference shape. Still the method could be
enhanced. Because this virtual fixation and distortion compensation is done by means of
FEM-analysis, special focus has to be set on optimizing the boundary conditions. Especially
the degrees of freedom at the fixation points and the local wall thickness are subjects where
improvement is still possible. For that purpose these parameters will be optimized in the
context of further research.
The performance of virtual fixation is evaluated by an indirect comparison between the
virtually fixed data and the measured data set of the fixed (= real distorted) workpiece, using
the CAD-model as reference. A proprietary developed fixation device is used. The positions
of the fixation points are based on the CAD-model. Figure 13 shows the deviation of
measured sample points to equivalent points on the CAD-model.
The results validate that the used fixation device and the used workpiece meet the drawing
specifications with good approximation and are suitable for verifying the results of the virtual
fixation. A fixation device could also be used for tactile and optical control surveys in
production runs. The maximum deviation to the CAD-model is 1.2 mm [3], so the aim is to
receive equivalent results by using virtual fixation.
To sum up: already at this early stage of research the achieved results with the method of
virtual fixation are very precise and reflect the real workpiece's shape very well. The
maximum deviation of the shape is satisfactorily accurate for most applications.
8. MEASUREMENT UNCERTAINTY
The accuracy of the virtual fixation method is limited by measurement uncertainty, which
is the object of further analysis. The measurement uncertainty is the sum of the uncertainty of
the fringe projection system, the uncertainty of FEM simulation (e.g. wall thickness
assumption), uncertainty in the simulation of the fixation system and the uncertainty of the
evaluation of point coordinates between which the distance is calculated as an indicator for
deviation. With the current configuration of the MacroSPS, the uncertainty of virtual fixation
is calculated to 0.7 mm [3]. In comparison, the measurement uncertainty of CMMs for the
same tasks is added up to 0.1 - 0.2 mm [7].
Figure 14 shows the Ishikawa diagram for fringe projection systems. As depicted,
operators and measurement strategy are the main factors that influence the quality of
measurement results. Therefore it is crucial to find the ideal configuration of the measuring
system, e.g. the geometrical relation between projector, camera(s) and object. It is also
necessary to consider the quality of components, as the resolution of measuring systems
depends on the CCD-Matrix in the camera, for example [3]. For the analysis of these effects,
the measurement will be simulated by a virtual and flexible measuring system. Thus it will be
possible to optimize the measurement situation and to compile the acquired knowledge in an
operator assistance system.
9. CONCLUSION
Fringe projection systems are suitable for fast and contact free measurement of formed
sheet metal parts without fixation or clamping. The measurement result can be used to extract
features on the workpiece, like holes or edges. Some of these are relevant for the assembly
process and subject to further inspections. Thanks to the information about the deviation of
the assembly features from their actual position to their nominal position, virtual distortion
compensation which means virtual fixation can be used to calculate feature parameters of
the distortion compensated shape. Thus the shape in assembled state can be simulated and the
inspection process chain can be shortened and automated simultaneously. Procedures for
automating the process must be developed.
The comparison between the measurement with the fixed work piece and the virtually
fixed data shows that the method could be improved. For example boundary conditions to
simulate a real model of the fixation system. Furthermore, the geometric conditions can be
improved: for example thickness was an average value for all the workpieces. In further
analyses variable wall thickness (as in the real workpiece) will be applied. Another object of
research will be a comparison between optical measurement, tactile measurement and virtual
fixation. Tactile measurement is state of the art in quality inspection of formed sheet metal
parts and thus will be used for verifying results.
The actual limitation of the method is the measurement uncertainty and uncertainty of the
accuracy of the FEM simulation. Because of this, further investigations aimed at optimizing
the measurement situation and reducing the uncertainties will be performed. The results of
these investigations will be introduced to an assistance system for guiding the operator.
ACKNOWLEDGEMENTS
The underlying research is gratefully funded by the German Research Foundation (DFG).
REFERENCES
1. Weckenmann A., Gall P., Ernst R.: Virtuell einspannen, Prfprozess flchiger Leichtbauteile verkrzen.
Qualitt und Zuverlssigkeit, no. 49, 2004, pp. 49-51.
2. Weckenmann A., Gall P., Gabbia A.: 3D surface coordinate inspection of formed sheet material parts using
optical measurement systems and virtual distortion compensation. 8th International Symposium
LaserMetrology, 2005. Proc. of the 8th International Symposium, Merida, Mexico, 12th-18th February 2005.
3. Weckenmann A., Gabbia A.: Testing formed sheet metal part using fringe projection and evaluation by
virtual distortion compensation. Fringe 2005, Proc. of 5th International Workshop on Automatic Processing
of Fringe Patterns, Stuttgart, Germany, 12th-14th September 2005.
4. Weckenmann A., Gall P., Hoffmann J.: Inspection of holes in sheet metal using optical measuring systems.
Proc. of VIth International Science Conference Coordinate Measuring Technique, Bielsko-Biala, Poland, 21st24th April 2004, pp. 339-346.
5. Stockinger F.: Kalibrierung von optischen 3D-Sensoren. Erlangen, Germany, Dissertation 1997.
6. Frankowski G., Chen M., Huth T.: Real-time 3D Shape Measurement with Digital Stripe Projection by Texas
Instruments Micromirror Devices DMD. Proc. of SPIE, vol. 3958, San Jose, USA, 22nd .-28th January 2000,
pp. 90-106.
7. Weckenmann A., Knauer M., Killmaier T.: Uncertainty of Coordinate Measurements on Sheet Metal Parts in
Automotive Industry. Sheet Metal 1999, Proc. of the 7th International SheMet, Erlangen, Germany, 27th- 28th
September 1999, pp. 109-116.
8. Karbacher S., Husler G.: A new approach for modelling and smoothing of scattered 3D data. Ellson R.N.,
Proc. of SPIE, vol. 3313, San Jose, USA, 24th -30th January 1998, pp. 168-177.
9. Steinke P.: Finite-Elemente-Methode. Berlin, Heidelberg, New York, Springer 2004.
10. Sun Y., Page D., Paik J.K., Koschan A., Abidi M.A.: Triangle Mesh-Based Edge Detection and its
Application to Surface Segmentation and Adaptive Surface Smoothing. Proc. of IEEE 2002, International
Conference on Image Processing (ICIP 2002), Rochester, New York, USA, 22nd - 25th September 2002,
pp. 825-828.