Integrated environment for 3D automatic inspection of complex parts S. Remy, A. Bernard, G. Ris CRAN, Faculté des Sciences, 54506 – Vandoeuvre Cedex – France Tel : + 33 (0)3 83 91 27 29 ; Fax : + 33 (0)3 83 91 23 90 ; email :
[email protected] ;
[email protected] ;
[email protected] The Research Center on Automatic control located in Nancy (France) has initiated a project to create an automatic 3D digitizing system. This paper describes the progress that has been made on this project and explains the problems we have to yet to overcome. Our research specifically attempts to resolve the problem inherent in the automatic measurement of mechanical parts. We have threrfore proposed a method for automatic free form digitalisation based on a CAD model. This method uses visibility concept. RÉSUMÉ: Le Centre de Recherche en Automatique de Nancy a lancé un projet dont le but est la mise au point d’un système de numérisation automatique d’objets 3D. Cette communication présente l’état d’avancement de ce projet ainsi que les étapes qu’il nous reste à franchir. Nos travaux cherchent, avant tout, à résoudre le problème de contrôle automatique de pièces. ABSTRACT:
Nous proposons, donc, une méthode de génération automatique d’acquisition de formes s’appuyant sur le modèle CAO de la pièce à contrôler et qui utilise la notion de visibilité. KEY WORDS:
3D control, laser scanning, integrated environment
MOTS-CLÉS:
contrôle tridimensionnel, scanning laser, environnement intégré
1. Introduction Other the last few years, many technological improvements have been made at all the different level of conception and machining of a product. 3D digitising of complex free forms is just one of these evolutions. It has become a very important part of the reverse engineering process. However, when one places a digitizing cell in an industrial environment, there are many problems to be solved and many question to be answered The integration of a laser sensor on the CMM belonging to the Atelier Interétablissement de Productique Lorrain (Inter-university research workshop in the Lorraine region of Eastern France), has enable us to highlight all these problems which have been the starting points for several research projects. These projects are parts of a global project managed in the Research Centre for Automatic control located in Nancy (CRAN – France). Stephane DAVILLERD [DAV 99] began by integrating the KLS 51 sensor from KREON Industrie (France) on a CMM produced by DEA. He also developed a simulation and off line programming module in the SILMA Cimstation Robotics environment. Then, he subsequently proposed a state of the art report concerning the different 3D digitising technologies. Benoît SIDOT [SID 99] worked on the problem to determining the digitalization range : what are the different positions required to completely digitize the mechanical part ? In this publication, we will begin by giving a short state of the art review of automatic digitizing, rapidly recalling the results of Stephane DAVILLERD and Benoît SIDOT. Following this, we will clarify where we are up to in our project and the work required to finish it. 2. State of the art During the digitizing operation, the operator must not only take many parameters into account but must also solve many questions concerning the position of the mechanical part. Given the considerable number of constraints, a digitizing operation becomes long and tiresome in cases where the mechanical part is complex. This process must be automated because the industry does not take a favorable view of such time consuming processes. The literature shows certain initiatives which are very often limited to particular applications. In his PhD work, Flavio PRIETO [PRI 99] draws up an assessment of work undertaken on automatic digitizing, using a plane laser sensor. The method used has provided some good results, its originality lying in the definition of a precision
coefficient for the result of the scanning. This coefficient depends on the difference between the optimal position of the sensor and those possible. This approach was tested for parts that present planes and cylinders. This should reveal itself to be valid for more complex parts. Plane laser sensor seems to be the most commonly used. Indeed, a good deal of work is carried out using such equipment. For example, the Autoscan system by Steinbichler uses a ZEISS sensor. This system is known and used in the automobile industry to digitize part of the body. This type of sensor is expensive and its applications for small and medium size parts are not easy. Mr LAMB, BAIRD and GREENSPAN [LAM 99] note that many tasks are repetitive and they propose to automate them. However, they do not provide any exploitable solutions. In 1996, M.J. MILROY, C. BRADLEY and G.W. VICKERS [MIL 96] used an OCS model (Orthogonal Cross Section) to automate the digitizing process. This method is particular and does not take into account the sensor environment and the optical limits of the sensor. Chang SHU and Fengfeng XI use laser plane width to automatically generate trajectories enabling the entire surface to be scanned. This method does present certain some limitations for the optimization of trajectories and for recovery zones due to the fact that it is only geometrically based. Dimitri PAPADOPOULOS ORFANOS [PAP 99] developed a system which allowed him to discover the surface to be digitized step by step. However, this method should be used on a 2 ½ axis system and does not take account of collision problems and optical limits. This work was carried out with laser plane sensors produced by KREON, BIRIS, HYMARC and ZEISS. Of course, other systems and approaches have been studied which are not based on laser technology. For example, RENISHAW [REN 00] has proposed contact sensors and a RETROSCAN environment. This is a scanning system for CNC tooling machines. W. BOSEMANN and F. GASTOU [BOS 99] have used photogrammetry to automate industrial processes. The range of accuracy depends on the size of the parts and the main applications are for design products and objects. This state of the art review shows that the problem of the real positioning of the part in the workspace has not been solved. We would therefore like to outline our approach to this problem and the solutions we have found will be presented later in this paper. First of all, let us introduce our simulation and off line programming module.
3. Results S. DAVILLERD and B. SIDOT obtained the following results. Their work was carried out during their last year of engineering studies. The research work has been published several times [BER 99a] [BER 99b]. They successfully finished their engineering studies respectively one and one year and a half ago [DAV 99] [SID 99]. 3.1. The CRAN project This project comes from a study concerning the integration of a 3D digitizing tool in a context of industrial production. The main idea is to propose a system to an enterprise. This system will enable a comparison to be made between the CAD model of a mechanical part and the point cloud from the scanning of this part. Our research work concerns the dimensional control of a mechanical part and not, for the time being, reverse engineering.
CAD model
Visibility processing Position of the mechanical part processing
Sensor paths processing Error map generation Realization of the positions and the paths in the simulation environment Point cloud treatment
Realization of the positions and the paths in the real workspace
Figure 1. Methodology
This project came into existence after the acquisition of a KLS 51 sensor from “KREON Industrie”. We also have a gantry type CMM from DEA. A rotary table on the marble is used as a fourth axis. 3.2. Integration of a laser sensor on a CMM. Simulation and off line programming. This work [DAV 99] is the first step of a global project for the preparation, simulation and automatic generation of the range of digitizing with a laser sensor. The work demonstrates that it is essential to be very familiar with the different parameters of 3D digitizing. With this knowledge, we have been able to integrate our sensor efficiently on our CMM. We have also elaborated a scanning methodology, which has been tested successfully on several industrial cases. A simulation and off line programming environment have also been developed on the CAD system SILMA Cimstation Robotics. The simulation of the points acquisition was performed by analogy with the PAINT function, which carries the deposit of painting on the mechanical part. This function had to be adapted to suit the optical limits of our sensor and the constraints of collision between it and the mechanical part.
Figure 2. Simulation and off line programming environment 3.3. 3D digitizing range processing for a control operation This is the second stage of the project [SID 99]. We needed to define a method and an algorithm to determine the scanning strategy which consists in positioning the sensor and the mechanical part in a known environment.
The method retained to resolve the 3D digitizing range generation is based on the concept of visibility. This has never been used in areas such as milling, inspection, foundry, vision and infography. The phase of research for visibilities is based on spherical geometry and on a method especially developed for the technology of our sensor [BER 00]. When the visibilities are known, we need to determine the different positions of the mechanical part as well as the different stages of scanning i.e. the range of 3D digitising. These positions were processed by matching the directions of the mechanical part visibilities with a reference mark related to the fields of measurement, related itself to the reference mark of the machine (CMM) [SID 99]. 4. Work in progress After this research work, we had to develop two important steps. The first is the physical realization of the mechanical part position, it is the positionning concept (by analogy with machining fixtures) The second is the generation of scanning paths. We also have to validate the entire project, i.e. verify that all the different modules of the project function together. 4.1. Position of the mechanical part in the workspace. As we have seen in Benoît SIDOT's work [SID 99], the position of the mechanical part depends on the sensor visibilities. Our program takes into account a CAD model of the mechanical part and the sensor model. It then suggests some changes in position of the mechanical part in relation to the machine referential. These transformations are easy to realize in a virtual space such as a CAD environment, but their realization in the real workspace of the CMM is harder. As we wish to save time, it is unthinkable to spend a considerable amount of time putting the part in position, thereby wasting the time saved by automatic processing of the trajectories. We also must take care that our "digitizing fixture" (as the machining fixture) does not hide some areas of the mechanical part. For any position of the part, we need to know the areas that will not be scanned in this position (areas that will be scanned in an other position, for example) [SID 99]. When the potential bearing surfaces are known, it is possible to create a "digitizing fixture" in a CAD environment. In this case, Benoît SIDOT's processing algorithm [SID 99] gives us a sufficiently wide range of solutions. Hence, there are not many constraints regarding the making of the "digitising fixture". If we wish to automate all the process, the system must be able to “find” the part in the workspace. We intend to test several ways to do this.
We should, for example, use a vision module. In fact, today it is easy for a robotic system to find an object, so it should be easy to find the mechanical part in the workspace. A second solution would be to divide up the CMM workspace. We would then be able to classify the mechanical parts by family of size. Each family would have a reserved area on the table. If the system knows the part and its family, it knows where the part reserved area is and will search for it using an algorithm such as that of PAPADOPOULOS ORFANO [PAP 97]. A third solution would be to use the possibility we have of manually digitizing with the joystick. We simply need to rapidly digitize a characteristic area of the mechanical part. Then, the system will readjust the point cloud and the CAD model to calculate the exact position of the part in the workspace. 4.2. Generating Paths When the mechanical part is in its scanning position, path generation consists in making the sensor's field of measurement successively cover all the areas of the part that will be scanned in this position. This operation has to take into account avoidance phenomena between the mechanical part and the sensor as well as the degradation of the information obtained at the optical limits of the sensor.
Figure 3. KLS 51 and its field of measurment
Many experiments have proved that the sensor is less accurate at the limits of the field of measurement. This observation is also correct for wide angles between the laser plane and the normal of the surface to be digitized. Once processed, these trajectories can be simulated with our simulation and off line programming module. 5. Conclusion Our project aims at providing industry with a solution for the automatic control of mechanical parts via the use of a contactless sensor . Several stages have already been completed : the simulation and off line programming environment functions and the visibilities and position processing module. In a few months, the automatic path generation module and the mechanical part research module will be finished. The stage consisting of the physical realization of the mechanical part position will be assisted and not automated because of the infinity of the parts to be digitized. 6. References [BER 99a] Bernard A., Davillerd S., Sidot B., Analysis and automatic generation of 3D digitizing processes for laser sensor, 8th European Conference on Rapid Prototyping and Manufacturing, Nottingham, UK, Juillet 1999. [BER 99b] Bernard A., Davillerd S., Sidot B., Ris G., Simulation and validation of scanning processes for laser sensor, Third International Conference on Industrial Automation, Montréal, Canada, Juin 1999. [BER 00] Bernard A., Véron M., Visibility theory applied to automatic control of 3D complex parts using plane laser sensors, CIRP Annals Vol. 49/1/2000 pp 113-118, N° ISBN 3-905 277-33-6 [BOS 99] Bosemann W., Gastou F., Système de mesure photogrammétrique pour l’automatisation de processus industriels et le contrôle de procédés, Conférence Numérisation 3D, Paris, Mai 1999. [DAV 99] Davillerd S., Intégration d’un capteur laser sur une MMT. Simulation et programmation hors-ligne, Mémoire d’ingénieur CNAM, Spécialité Mécanique, option Production automatisée, mars 1999. [LAM 99] Lamb D.G., Baird D.L., Greenspan M.A., An automation system for an industrial 3D laser digitizing, Conférence 3DIM’99 p 148, Ottawa, octobre 1999. [MIL 96] Milroy M.J., Bradley C., Vickers G.W., Automated laser scanning based on orthogonal cross sections, Machine Vision and Applications n°9 p 106, 1996. [PAP 97] Papadopoulos-Orfanos, Numérisation géométrique automatique à l’aide d’un capteur 3D de précision à profondeur de champ réduite, Thèse de doctorat de l’ENST, Février 1997. [PRI 99] Prieto F., Métrologie assistée par ordinateur. Apport des capteurs 3D sans contact, Thèse de doctorat de l’INSA de Lyon, 1999.
[REN 00] Documentation RENISHAW, 2000 [SHU 99] Chang Shu, Fengfeng Xi, Model-based scanning path generation for inspection, Conférence 3DIM’99 p 118, Ottawa, Octobre 1999.
[SID 99] Sidot B., Détermination de la gamme de numérisation en vue du contrôle d’une pièce, Mémoire d’ingénieur CNAM, Spécialité Mécanique, option Production automatisée, octobre 1999.
