Flexible Automation and Intelligent Manufacturing, FAIM2005, Bilbao, Spain Feature-Based Pr

Feature-Based Product Model to Support Automatic Control ofGeometrical TolerancesMichele Germani, Ferruccio MandorliDepartment of MechanicsPolytechnic University of MarcheVia Brecce Bianche, Ancona,I 60131, ITALY, m.germani@univpm.itA BSTRACTThe growing use of “augmented” virtual product models, which not only represent the geometric information, butalso collect attributes, parameters, rules and procedures, can facilitate the digital simulation of many realprocesses. The control process of manufactured products can take relevant advantage of such technology. In thiscontext, an approach for the automatic virtual inspection of geometrical tolerances, based on “Full ofInformation” 3D feature-based CAD models and non-contact 3D shape digitizing techniques, is proposed. The“augmented” CAD model encapsulates the designer functional specifications in terms of tolerances and methodsto inspect them. The measured point cloud data report a detailed representation of the real product. Dedicatedprocedures and algorithms can allow to automatically evaluating the designer specifications comparing the twodata model.1. I NTRODUCTIONThe Geometrical Tolerances (GT) inspection process is a set of measurement activities to verify if desired product functional features, defined during the design phase and represented by toleranced geometries, have been respected after the manufacturing process. Generally, these control processes are time consuming, and they require high-skilled operators able to apply the GT inspection rules using Coordinates Measurement Machines (CMMs). Computer Aided Tolerancing and Inspection software packages, based on 3D CAD models data, can be a valid support but they are still expensive and not well integrated into the product development process. In this context, many manufacturing enterprises, especially Small and Medium sized Enterprises (SMEs), do not use properly GT prescriptions causing design errors and generating, consequently, process iterations.The goal of our research is the study of easy-to-use solutions to extend the application of GT. The improvement of the control process automation, for example, by a strong interrelation between the design and the inspection phases can be a suitable approach. In particular, this paper presents the definition of a product model representation able to collect the tolerances information, and a methodology to manage measured data, providing completely automatic product verification.The method is based on the integration of feature-based CAD models and 3D optical digitizing systems. In more details, the product model framework has been defined and extended to implicitly collect, within the model data structure, the designer tolerance specifications as geometry attributes. The product model has been represented by a Full of Information (FoI) CAD model and a knowledge base, which verifies the coherence of tolerances prescriptions. Such a model becomes the reference to compare automatically the measured data. In fact, a fast optical 3D shape acquisition system can provide an accurate virtual representation (point cloud data) of the manufactured object. Point cloud data analysis algorithms have been developed to separate the different zones corresponding to the basic CAD modeling surfaces. A matching method enables to map them with the FoI CAD model features attributes. Finally, specific algorithms have been studied and implemented to apply GT verification rules on these virtual models. In this way it is possible to support a reliable and automatic inspection process based on the ISO and/or ANSI/ASME standards procedures. Furthermore, the designer specifications can pilot directly the product verification stage, involving a minimization of errors due to operators misinterpretations.Currently, the methodology has been tested on orientation tolerances using deep-drawn sheet-metal and plastic injection moulded components, in the paper is reported an experimental case on an automotive application.The preliminary validation of methodology and the related developed prototypal software solutions show promising results in terms of time savings and usability.2. R ELATED WORKIn these last years the ISO/TC 213 (Dimensional and Geometrical Product Specifications and Verification) is carrying out a meaningful rethinking work of procedures and standards related to the tolerancing problem. In particular, they have uncovered a set of rules showing the correspondence between processes of tolerance specification and tolerance verification. This allowed to theorize an harmonizing of set of operations involved in such two activities. The resulting principle has been defined as “duality principle” [1]. In fact, two main operators have been determined: the specification operator and the verification operator. Each operator contains a set of feature operations that are dual between the groups; such one-to-one mapping provides an integrated view of the tolerancing problem.An open issue is the development of methodologies and tools, which allow the efficient application of such principles and operations. They can not disregard the tolerances representation within the design systems, in particular within the three-dimensional CAD software packages, that are used during the design intent definition. Many tolerance representation models have been studied and proposed. They have been finalized to support different product development phases. As reported in literature [2], they can be classified in: documentation oriented, analysis oriented, production oriented and control oriented. From the CAD modeling viewpoint, the use of feature-based approach has been largely investigated. For example an interesting method to link dimensioning and tolerancing schemes in CAD systems, based on graph representation, has been described in [3]. A feature classification (atomic, primitive and compound) has been defined as base to implement a tolerancing module in [4].In our approach the feature-based CAD model, with tolerances, is documentation oriented, since it collects the nominal geometry and the related tolerances. The analysis of data structure allows the identification of the atomic and compound features with tolerances; the resulting sub-model is “augmented” with relations between features and verification procedures. This information can be used to determine the skin model, as defined in [5], and to perform the comparison between the real model and the skin model itself.The other part of the dual model is the virtual representation of the real object, as a set of point cloud data, once it has been digitized. Several technologies allow the acquisition of the 3D object geometry, but the optical systems, especially based on the triangulation principle, have evident advantages in terms of speed and usability. Their adoption for inspection tasks has been widely studied [6], [7]. The measurement accuracy is not comparable with CMMs, but systems are in continual improvement [8]. The connection between CAD models and 3D inspection systems has been approached to determine optimal inspection strategies [9], driving the scanning system for freeform surfaces and related data verification. The point cloud data analysis to perform the geometrical tolerance verification is a consolidated functionality of more common reverse engineering commercial software systems (RapidForm by Inus Tech., Geomagic Qualify by Raindrop, Polyworks by Innovmetric), but they provide only algorithms to facilitate the feature extraction within a dense points cloud. In fact, they are completely disjointed from the tolerance specification process and, furthermore, they require constant user interactive decisions for the verification task and for data segmentation. An advanced and integrated solution proposed in literature [10] is the most effective example of verification automation, but the CAD model is used only as reference to perform the range data segmentation activity. A robust method to partition the point cloud data [11] is one of the main problems to be considered.The proposed method based on optical 3D digitizing systems, generates a dense and noisy point cloud data, hence suitable filtering algorithms have to be applied. The resulting extracted data are partitioned using a methodology based on local differential properties [12], calculated directly on point cloud data. These sub-clouds are used to identify the skin model and to represent the real object geometry to be verified.3. M ETHODOLOGY FOR GEOMETRICAL TOLERANCES VERIFICATIONA simplified block diagram of the inspection process is shown in figure 1. The boxes in the figure indicate the order of processing and illustrate the use of information generated.Figure 1; Block diagram of the inspection process proposed.The designer, during the product modeling phase, can link tolerancing attributes (datums, tolerancing values and tolerancing typologies) to the atomic (such as points, axis, lines) and compound features (such as surfaces, groups of lines, groups of faces, …) on the basis of required product functionalities.These attributes are stored into the CAD data structure. This model can be automatically navigated to extract the explicit needed information for the tolerance verification. In particular, the features extraction operation finds elements interested in tolerance prescription, generating an ordered list of properties for each feature as follows: Tolerance (number), method (group, typology)Feature (number), Typology (i.e. axis), CofG (G x, G y, G z), Datum (letter), Tolerance (typology, value, datum);The first row indicates a number that represents a specific tolerance prescription, highlighted in the following rows; such progressive number univocally identifies each tolerance attached to the CAD model. Then the method to be used, for the specific tolerance verification, is indicated. These methods are collected into a knowledge-base that provides rules and procedures. For example, if an orientation tolerance is prescribed, the first tasks are the datum identification and the determination of configuration (since it could be a combination of features), then the toleranced geometry must be found and, finally, on the basis of tolerance typology (parallelism, perpendicularity, etc.), the list of simulation geometrical operations to be performed, for the virtual control, is reported. In the following rows is identified the specific information related to datums and toleranced geometries.The CAD model data structure is analyzed and when a tolerance specification is located the software system looks for all the related specifications (generally the datums), the final list appears as follows:Tolerance (1), method (group, typology)Feature (number), Typology (i.e. axis), CofG (G x, G y, G z), Datum (letter), Tolerance (typology, value, datum);……Feature (number), Typology (i.e. axis), CofG (G x, G y, G z), Datum (letter), Tolerance (typology, value, datum);Tolerance (2), method (group, typology)Feature (number), Typology (i.e. axis), CofG (G x, G y, G z), Datum (letter), Tolerance (typology, value, datum);……Feature (number), Typology (i.e. axis), CofG (G x, G y, G z), Datum (letter), Tolerance (typology, value, datum);………….Tolerance (n), method (group, typology)Feature (number), Typology (i.e. axis), CofG (G x, G y, G z), Datum (letter), Tolerance (typology, value, datum);The further interesting aspects are linked to the last two tasks of figure 1 (mapping and inspection).The mapping between toleranced features and the corresponding point sub-clouds data is carried out using algorithms to calculate minimum distances. The information of centers of gravity (CofG) coordinates, that is the point itself in the case of a point datum, or the middle point in the case of an axis datum, is compared with the set of the sub-clouds resulting from the segmentation. The cloud at minimum distance is identified as the right sub-cloud.Once the other tasks have been successfully performed, the virtual inspection process can take place.The appropriated inspection method is selected according to the specific GT category and type to be inspected. Firstly, the method must describe the procedure to precisely identify the datum geometry into the sub-cloud. For each different datum and tolerance type, different strategies can be identified. The definition of the best identification strategy for all cases is still an open issue in our research. At present, we are focused on orientation tolerances with plane datum. Once the virtual datum has been identified, the method describes the procedure for the computation of the tolerance zone. Such procedure allows carrying out the verification process, simulating the operations similar to the traditional verification methodologies. Finally, the points belonging to the sub-cloud corresponding to the geometry to be inspected are checked in order to verify if the tolerance zone bound them.4. P RELIMINARY EXPERIMENTAL RESULTSThe specification and verification prototypal system has been implemented using the development environment of a commercial CAD software system (CATIA v.5.12 by Dassault Systemes). Components have been digitized using a fringe projection 3D scanner (COMET Vario Zoom by Steinbichler Gmbh). The test has been performed on different components typologies, for example sheet metal components for the automotive field. Such products have ideal characteristics for the experimental system application, in fact, they present freeform and regular geometries, they have a medium size, and the tolerance values are sufficiently high to be considered consistent if compared with the measurement accuracy achievable using optical triangulation systems. In figure 2 (upper part) one of the tested components is reported. The point cloud data has been filtered and sub-sampled (figure 2, low part) to highlight the curved zones and the planar zones. The nominal model geometry is illustrated in figure 3.Figure 2; The measured sheet-metal component (upper figure), the point cloud data after filtering and sub-sampling (lowerfigure).Figure 3; The CAD model (nominal geometry) and one of the tolerance specifications (parallelism) prescribed of the testedcomponents.To exemplify the methodology application, only one of the specified tolerances is reported, that is a parallelism between two planar zones. From the design viewpoint it is prescribed to allow an accurate assembly of component. The reference plane is determined by a common zone identified by three small planes (figure 3), while the toleranced geometry is a common zone of two planes (figure 3). Hence, in this case, the meaningful surfaces are two planes, which must be parallel within a range of 0,5 mm. A best-fit method based on the ICP (Iterative Closest Points) algorithm is used to find a preliminary relative positioning of the two models (in figure 4 is reported the color map result).Figure 4; Best-fit method applied to position point cloud data and CAD model.The developed software application provides the point cloud data segmentation using local differential properties (curvature and slope). On the other hand, the data structure analysis of the augmented CAD model allows extracting the design intent in terms of toleranced features and datums. The information is represented in according to the formalism reported in section 3.Tolerance (1), method (orientation, parallelism)Feature (112), Typology (plane), CofG (65.17,-275.9, 603.6), Datum (A), Tolerance ();Feature (111), Typology (plane), CofG (76.7,-8.4, 624.1), Datum (A), Tolerance ();Feature (109), Typology (plane), CofG (65.17,275.9, 603.6), Datum (A), Tolerance ();Feature (139), Typology (plane), CofG (-51.3,602.1, 659.6), Datum (), Tolerance (parallelism, 0.5, A);Feature (387), Typology (plane), CofG (-51.3,-602.1, 659.6), Datum (), Tolerance (parallelism, 0.5, A);The matching with the point sub-clouds data allows identifying the measured zones corresponding to the specified tolerance and datums (as shown in figure 5).Figure 5; Point cloud data segmentation and toleranced zones identification.The tolerance verification process, at this point, can be performed. The knowledge base provides the inspection procedures needed for the parallelism control. Firstly, the simulated datum reference frame has to be identified analyzing the point cloud data corresponding to the surfaces characterized by the presence of datum attributes.Figure 6; Datum plane determination using different point cloud data overlapping (upper left, overlapping 20 %, upper right, overlapping 60 %), toleranced planes determination (lower left), inspection result (lower right).In this case, only the datum A is prescribed, and it is a plane. The simulated datum reference frame generation is an open issue of our research work. In fact, we are evaluating different solutions to simulate virtually the behaviour of the interaction between the measured component and the hypothetical real measuring set-up. Currently, for the planar reference case, we consider as the normal of datum plane, an average of planes normal orientation calculated at different percentages of point cloud overlapping (20%, 30%, 40%, 50%, 60%), as illustrated in figure 6. Once the simulated datum plane orientation is fixed, two planes parallel to it are calculated considering the closest and the most distant points of the point cloud representing the toleranced geometry (figure 6, lower left). To satisfy the specified tolerance such range (distance between these two planes) must be lower than the prescribed tolerance value.In the reported case, the verification process determined an error higher than the desired value; hence the component should be rejected. Currently, the result has been influenced by a set of inaccuracy sources (point cloud data acquisition system, matching errors, datum planes determination, etc.) that we are going to estimate to establish the systematic error and, hence, to obtain a more robust methodology.5. SUMMARY AND EXTENSIONSWe have presented a methodology for CAD-based geometrical tolerances verification. CAD information reports the design intent in terms of tolerancing attributes linked to the geometrical features, on the other hand, a rapid and sufficiently accurate 3D digitizing system translates the real component in a virtual representation (point cloud data). Such models provide all information needed to perform a virtual inspection of tolerances.The methodology relevance has been analyzed using specific test cases and particular technologies, the preliminary results showed a promising future. In fact, the specification and verification processes can be managed using a single product model limiting the time-consuming problems related to misunderstandings between designers and quality control operators. Besides, the tolerance inspection automation can be well supported because, once the component has been digitized, the process can be completely based on the CAD model information.The main advantages of the proposed methodology concern the integration of design aspects with inspection aspects: once the manufactured product has been digitized, the virtual inspection can be performed in the CAD environment on the basis of procedures that are controlled by the designer. Moreover, the virtual inspection completely overcomes the need to arrange a physical inspection set-up and this will dramatically decrease the inspection costs.However, some aspects related to the proposed methodology need to be deeply investigated. The present limitations for a wider application of the methodology are mainly related to the acquisition phase and the definition of the virtual datums.The acquisition process with optical systems has intrinsic limitations that under certain conditions may prevent the successful application of the methodology. The greatest limitation is the accuracy in the measurement phase; in fact optical acquisition systems ensure +/- 0,06 mm, using the better laser stripe systems, or +/-0,1 mm in the case of fringe projection systems. However, while this accuracy can be acceptable for several applications, the errors introduced during the matching, filtering and sampling of the clouds have to be kept under control.At the present stage of the work we have focused our attention on orientation tolerances but the system will be extended the formalization of rules and methods regarding all geometrical tolerances.We have identified different strategies to define the virtual datums. But the robustness of datums has to be furtherly investigated verifying different algorithms and comparing them with the results of traditional measurement processes. In fact, more experiments are required to identify the best strategy and to extend the set of procedures to define the datum and to compute the tolerance values in case of other GT types. Particular efforts have to be dedicated to cases in which the datum information is contained intrinsically in the point cloud data, for example the axis of a cylindrical element, in this situation the data extraction algorithm must ensure the identification of the right sub-cloud and the modeling algorithms must elaborate the proper simulated representation.Finally, the full integration of tolerance verification software system with rapid three-dimensional scanning systems, using the CAD model as reference, can be a good support to develop the on-line inspection processes. In this direction further research work is focused towards the definition of an optimized scanning strategy based on the geometric information collected within the CAD data model.R EFERENCES[1] V. Srinivasan: "An integrated view of geometrical product specification and verification"; Proceedings of the 7thInternational Seminar on Computer Aided Tolerancing, Kluwer Academic Pub, pp. 1-12, ISBN 1-40201-423-6, 2003[2] T. Kandikjian, J. Shah, J. Davidson: "A mechanism for validating Dimensioning & Tolerancing schemes in CAD systems",Computer-Aided Design, Vol.33, No. 10, pp. 721-737, 2001.[3] J. Shah, Y. Yan, B. C. Zhang: "Dimension and tolerance modeling transformations in feature-based design andmanufacturing", Journal of Intelligent Manufacturing, Vol.9, No. 5, pp. 475-488, 1998.[4] J. Gao, K. Case, N. Gindy: "Geometric elements for tolerance definition in feature-based product model"; Advances inManufacturing Technology VIII, eds. K. Case and S.T. Newmann, Taylor & Francis, pp.264-268; ISBN 0-7484-0254-3,1994.[5] ISO/TS 14750-1: "Geometrical Product Specification (GPS) – General Concepts – Part1: model for geometricalspecification and verification”.[6] Y. Li and P. Gu: "Free-form surface inspection techniques state of art review"; Computer Aided Design, Vol.36, No. 13,pp. 1395-1417, 2004.[7] S. Son, H. Park, K. Lee,: "Automated laser scanning system for reverse engineering and inspection"; International Journalof Machine Tools & Manufacture, Vol.42, pp. 889-897, 2002.[8] F. Prieto, R. Lepage, P. Boulanger, T. Redarce: "A CAD-based 3D data acquisition strategy for inspection”, MachineVision and Applications Journal, Vol.15, No. 2, pp-76-91, 2003.[9] X. Huang and P. Gu: "CAD model based inspection of sculptured surfaces with datums"; International Journal ofProduction Research, Vol. 36, No. 5, pp. 1351-1367, 1998.[10] F. Prieto, R. Lepage, P. Boulanger, T. Redarce: "Inspection of 3D parts using high accuracy range data"; Proceedings ofthe SPIE conference VIII Machine Vision Applications and Industrial Inspection, pp. 82-93; San Josè 2000.[11] P. Benko, T. Varady: "Segmentation methods for smooth point regions of conventional engineering objects"; ComputerAided Design, Vol.36, No. 6, pp. 511-523, 2004.[12] P. Corbo, M. Germani, F. Mandorli: "Aesthetic and functional analysis for product model validation in reverseengineering applications"; Computer Aided Design, Vol.36, No. 1, pp. 65-74, 2004.。