T
his article describes a new solution for precise automated digitalization of large objects. The practical application is carried out with the help of a standard (imprecise) industrial robot, a near range on-line photogrammetric system and a 3-D White Light Scanner. The tracking system compensates for relatively low absolute positioning accuracy of the robot. The complete system has been successfully verified in a benchmark test for the German automobile manufacturer, AUDI.
In this test, the goal has been to provide a stable running system with outstanding characteristics in accuracy and acquisition time to digitize low volume preproduction parts for documentation and archiving. AUDI engineers want the ability to archive every part, in case they encounter problems in the following production or evaluation phase.
It was essential to the project that the aluminum and steel surfaces of the parts were scanned without any coating or treatment. This requires the system to cope with shiny and specular areas within the parts.
Because the data can be obtained without using any reference points, a complete digitalization of doors and hoods can be done in a few minutes.
An accuracy test has been done according to the guideline VDI/ VDE 2634 Part 3 for optical 3-D measuring systems with a multiple view based on area scanning which is published by the Association of German Engineers (VDI).
By applying this test guideline it is possible to identify three quality parameters: the probing error, the sphere spacing error and the length measurement error. By the use of these three parameters it is possible to compare different optical systems, or to compare optical systems to other devices such as CMM’s in order to secure more transparency for customers.
Background
Before a new model of a car is being built with mass production methods, there are always a varying number of prototypes being made. This functional prototype (also called a working prototype) will, to the greatest extent practical, attempt to simulate the final design, aesthetics, materials and functionality of the intended design. Usually there are between 50 and 150 of these cars produced. These small numbers require very flexible tools and processes involving lower capital costs, which may vary significantly from the final procedures. There is also still a lot of "handwork" involved at this stage. Another widely used approach is using flexible robots for welding, cutting and hemming. The stamping tools are often made from cheaper and reusable material.
All these reasons are responsible for the fact that the produced parts may be beyond the tolerances that are defined for the final production parts. They might even be quite different from part to part.
These larger tolerances can hardly be avoided, but they may lead to problems later on. For example, they may cause larger gap and flush values.
The ideal scenario for the pre-production manager would be to capture these imperfections and differences to the CAD model of each individual part. Not knowing exactly where the area of interest is at this stage, a tactile method would be not a suitable solution. Probing thousands of points, even with a CNC machine, to get the entire picture of the deviations would be by far too time consuming.
Objective
The objective of this project was to build a CNC-like optical measuring system, based on a white light scanner mounted on a robot. The department I/E-V2 at AUDI is responsible for the prototype production of doors, hoods and trunk lids. The digital documentation and archiving of these parts is needed in order to establish a digital archive of the prototypes developed in the pre-production chain. The digitized prototypes and their respective quality reports can then be easily accessed as reference objects in the subsequent optimization and production phase.
With the demanding time requirements in mind, the AUDI engineers chose a white light scanner. White light scanners can collect a data on a large area in less than a second.
The project was carried out at the company MQS Ingolstadt for Audi AG and describes a new approach for the precise automated digitization of large assembly parts in the automotive industry.
 |
|---|
Figure 1 part with a typical surface finish to be scanned in the project |
Requirements and challenges
The scanner selected for the project was a Breuckmann white-light stereoSCAN system. In other industrial projects it has proven its good performance on shiny surfaces while obtaining good resolution and accuracy.
The robot was a standard KUKA robot which is widely used for production purposes in the German car industry.
But it was clear from the beginning that the simple approach with just a scanner mounted on a robot would not be able to fulfill the demanding accuracy requirements. A new approach was chosen to compensate the limited absolute 3-D accuracy of the robot with a large volume optical tracking system. It should verify each scanning position during the scanning setup. These positions served as basis for the subsequent series measurement.
We chose the Metronor Duo system for this purpose. For better part handling and to make front and back side of parts easily accessible a large turntable was integrated into the setup. The following sections will give more details on the above mentioned components:
 |
|---|
Figure 2. Setup with two robots and turntable |
Robot and turntable
A standard KUKA robot, type KR2150 with arm extension, including a VKR-C2 control unit was used. The turntable below was integrated as a 7th axis into this control unit. The communication to and from the scanner PC was established through TCP-IP over Ethernet and a newly developed set of commands. For safety reasons the robot control unit acted as a "Master" providing commands to the scanner PC.
 |
|---|
Figure 3. Large turntable for part handling |
Optical tracking system, naviSCAN
The naviSCAN scanning system is comprised of a standard Breuckmann stereoSCAN white light scanner fitted with a dimensionally stable structure holding LEDs (naviTarget) that are tracked by the Metronor DUO cameras. With LED measurements, the DUO system is able to calculate the scanner’s 6DOF position and orientation in a machine coordinate system or in a local alignment. The NaviTarget is a 3-D framework of aluminum spheres connected with carbon fiber rods. Geometrical stability is provided by the triangular geometry of the structure and the precision components used in the assembly. LEDs are embedded in the spheres, and are distributed such that the NaviTarget can be observed in all orientations and positions within the field of view of the DUO cameras.
 |
|---|
Figure 4. The system in principle: two cameras in the background precisely locate the scanner’s position |
These positions and orientations are saved and used to correct the less accurate positions read from the robot control unit.
 |
|---|
Figure 5. In the setup at MQS the two tracking cameras were wall mounted (marked with red circle) and covered the whole working volume of the turntable |
White light scanner
The stereoSCAN system is a topometric system that makes use of the principle of optical triangulation in combination with structured illumination. The 3-D sensor is based on the Miniature Projection Technique (MPT), and utilizes a combination of Gray code and phase shift technique, which guarantees an unambiguous determination of recorded 3-D data. Patterns of well defined periodic fringes are projected onto the object, and recorded by two high-resolution camera positioned at a defined angle. Subsequently, the acquired images are further processed within a powerful image processing system, running in a 64 Bit Windows XP operating system.
 |
|---|
Figure 6. This graphic illustrates the principle of optical triangulation with a white light scanner |
It is important to notice that due to its rigid mechanical setup which incorporates two extremely rigid carbon fiber bars, the scanner can operate in any orientation and produce stable results.
 |
|---|
Figure 7. Within one second, the stereoSCAN system projects a series of 12 patterns onto the surface |
Technical specs of the scanning system:
Because the combined system consists of several independent sub-systems that all utilize their own coordinate systems (scanner, robot, turntable, tracking system) it is essential to perform robust calibration and alignment strategies. Otherwise errors from all subsequent steps will add up and spoil the total result. Therefore the goal was to be extremely precise with each calibration procedure and use redundant and over determined methods for the alignment process. These methods are able to provide feedback about the current accuracy level.
The Metronor Duo cameras are calibrated by presenting a bar with three LEDs in several different orientations and positions. This procedure requires the operator to move the calibration artifact to cover the desired volume. It can be completed in about 10 minutes and only needs to be redone if the cameras have been moved.
 |
|---|
Figure 8. Second step: Calibrating the Metronor system with LED artifact. |
The stereoSCAN system is calibrated in a similar way, but a calibration board is used instead of a bar. Usually the board is moved manually in front of the scanner.
 |
|---|
Figure 9. First step: automated calibration process of a stereoSCAN system |
For this project a robot program was written that performed the necessary movement, leaving the calibration board in a fixed position. Again it is required to move the board through the entire field of view for the scanner. This process takes about five minutes and will be executed on a daily basis.
A good definition of the rotation axis is important to precisely fit the back and front side of parts together. This was accomplished by using 3 tooling balls and measuring the center points with the Lightpen device provided with the Metronor Duo device. This device is equipped with LEDs that are tracked by the Metronor cameras. It can also be used as a handheld CMM probe. The center coordinates of the spheres are calculated in the cameras’ coordinate system and after three rotations of the table the rotation axis will be obtained. This process takes about 15 minutes.
 |
|---|
Figure 10. Third step: measuring the rotation axis of the turn-tilt table |
One general note at the beginning:
There are lot of experimental results on many different parts like doors, hoods and trunk lids but due to the confidential nature of this project, that deals with parts in the very early stage of new car models, it is not allowed to share all of them.
Most of the project goals could be reached, especially a very good repeatability (same part was measured 10 times, but left on fixture) and reproducibility (same part was measured 10 times with removing and reloading it) of parts could be proven.
See below the results compared to the list of requirements.
 |
|---|
Figure 9. An example: comparison of the inner side of a door to its CAD model, using Polyworks software. The white lines show the edge lines, the deviation of surfaces is color coded |
Results - shiny surfaces
All aluminum parts (hoods) and galvanized steel parts could be scanned without prior treatment or spraying. In some cases problems were encountered when the hall lights were mirrored on a surface. This resulted in missing point data in this area.
 |
|---|
Figure 1010. The door handle area on a shiny galvanized door is almost completely captured, without preparation of the surface. |
Results - surface and edge data
Surface data could be acquired in good quality on all parts, as well as edge data on all sharp edges. Some more work has to be done to improve the quality on hemmed edges.
 |
|---|
Figure 11. Scan data of surface and edge (white lines) in the area of the door handle |
Results - fully automated
The system is fully automatic, including robot control, turntable and automatic report generation on a second evaluation PC. The data is shared via network coupling.
Results - time limit
The time required to scan an entire part strongly depends on the number of scans needed. It was decided to use a relatively large field of view with about 725 mm diagonal (29 inches) to reduce the number of scans necessary.
With this FOV it was possible to meet the requirement of 20 minutes maximum scan time even on the largest part, which was a hood that took 45 scans.
Limitations and further ideas for improvements
As stated before, there is still room for optimization at several points:
The accuracy tests are part of a diploma thesis by coauthor Sebastian Zinck, and are too comprehensive for the scope of this article. Therefore only the essence of this thesis should be described here.
The work is based on the German guidelines VDI/VDE 2634 part 3 and VDI/VDE 2617 part 6.2. Both cover image processing systems that produce dense three dimensional point-cloud data from more than one viewing direction (multiple view system based on area scanning). The guideline 2634 part 3 covers such optical systems as a standalone system, whereas guideline 2617 deals with optical system that are used as a probe on coordinate measuring machines (CMMs).
By the use of such standardized methods it can be assured that systems that utilize different measuring technologies can be compared according to accuracy and repeatability. The guidelines also define methods for conducting re-verification on a regular basis and how to trace back accuracy numbers to national standards.
Last but not least, these guidelines are a good tool for users to compare systems from different suppliers to get a more reliable information about performance in terms of accuracy.
 |
|---|
Figure 12. The accuracy tests of the total system were performed according VDI procedures that are based on ball bar artifacts |
Principles according to VDI
Both above mentioned guidelines define three characteristics that are derived from measurements on calibrated tooling balls and ball-bar artifacts.
According to the needs of optical scanners, all balls should have a matte surface.
In the following only the results on "probing error" and "sphere spacing error" and repeatability will be explained.
Probing error
To determine the probing error a 30 mm diameter ceramic sphere was used and placed at 4 different positions in the scanning volume. Each of these positions was scanned from six different directions. (Please note that although the graphic below shows 6 different positions, only 4 are used.)
 |
|---|
Figure 13. A sphere with a diameter of 30 mm was placed at four locations and each scanned from six different directions |
| Position | Probing Error (Shape) in mm |
|---|---|
| 1 |
0.131
|
| 2 |
0.126
|
| 3 |
0.079
|
| 4 |
0.114
|
Table 1. Probing error
Sphere spacing error
The ball-bar-artifact shown in Fig. 12 consists of eleven ceramic balls on a 1800 mm long support. Six ball distances were selected and the artifact positioned in seven different directions within the working volume. The graphic below shows the results of comparing the measured distances to the calibrated ones. This was first done using only the robot coordinate system, which shows deviation up to 2.5 mm. The second graphic shows the improvement through using a tracking system: deviations are below 0.2 mm.
 |
|---|
| Figure 15. The ball-bar-artifact was positioned in these seven directions |
 |
|---|
Figure 16. Sphere spacing error (in mm) with seven positions and six different sphere spacings using only the robot coordinate system |
 |
|---|
Figure 17. Sphere spacing error (in mm) with seven positions and six different sphere spacings of the total system including the optical tracker |
Repeatability
To specify the repeatability of the described system, the measurements on the ball bar artifact were repeated three times.
For the complete system, including the optical tracker, the sphere spacing error showed an standard deviation (one sigma) of 0.015 mm.
But even when only the robot coordinate system accuracy has been used, the sphere spacing error varied by only 0.080 mm (standard deviation).
This shows that even with a relatively poor volumetric accuracy a robot can be still very repeatable. This offers the opportunity to use the optical tracker only for programming a new part. When the subsequent parts are being measured, the system can rely on the good repeatability of the positions which have previously been programmed and verified. So it can perform good measurements even without utilizing an accurate reference system for each part. This method has been verified during the course of the project.
By using a new combination of robotic automation with the 3-D positioning data of the naviSCAN, it was possible to capture large quantities of high resolution scan data without the use of reference points and without treatment of the shiny parts. This made it possible to completely digitize doors, hoods and trunk lids within a few minutes. At an average cycle time of approx. 15 minutes, this combination of systems easily met the time allowance and accuracy requirements for the scanning process and part changing, required for this project.
We would like to thank Audi AG Germany and MQS Measuring and Quality Solutions GmbH for their cooperation and support in this project, as well as for their kind approval of the resulting application report. Especially Norbert Scheitler of Audi and Christian Strauss and Andreas Kleinsteuber of MQS for their hard work and good support at all stages of the project.
We also like to thank William J. Mongon, president and CEO of Accurex Dimensional Measurement for his support in preparing the article.
"Real-World Application of Scanning and Discreet Probing of Large Components Using an Integrated System of a Structured Light Scanner and a Portable CMM", Coordinate Measurement Systems Conference 2008, Charlotte, NC, by Joel Adams, Erik Klaas, Nils Thune.
