Contribution à l’inspection automatique des pièces flexibles à l'état libre sans gabarit de conformation

The competitive industrial market demands manufacturing companies to provide the markets with a higher quality of production. The quality control department in industrial sectors verifies geometrical requirements of products with consistent tolerances. These requirements are presented in Geometric Dimensioning and Tolerancing (GD&T) standards. However, conventional measuring and dimensioning methods for manufactured parts are time-consuming and costly. Nowadays manual and tactile measuring methods have been replaced by Computer-Aided Inspection (CAI) methods. The CAI methods apply improvements in computational calculations and 3-D data acquisition devices (scanners) to compare the scan mesh of manufactured parts with the Computer-Aided Design (CAD) model. Metrology standards, such as ASME-Y14.5 and ISO-GPS, require implementing the inspection in free-state, wherein the part is only under its weight. Non-rigid parts are exempted from the free-state inspection rule because of their significant geometrical deviation in a free-state with respect to the tolerances. Despite the developments in CAI methods, inspection of non-rigid parts still remains a serious challenge. Conventional inspection methods apply complex fixtures for non-rigid parts to retrieve the functional shape of these parts on physical fixtures; however, the fabrication and setup of these fixtures are sophisticated and expensive. The cost of fixtures has doubled since the client and manufacturing sectors require repetitive and independent inspection fixtures. To eliminate the need for costly and time-consuming inspection fixtures, fixtureless inspection methods of non-rigid parts based on CAI methods have been developed. These methods aim at distinguishing flexible deformations of parts in a free-state from defects. Fixtureless inspection methods are required to be automatic, reliable, reasonably accurate and repeatable for non-rigid parts with complex shapes. The scan model, which is acquired as point clouds, represent the shape of a part in a free-state. Afterward, the inspection of defects is performed by comparing the scan and CAD models, but these models are presented in different coordinate systems. Indeed, the scan model is presented in the measurement coordinate system whereas the CAD model is introduced in the designed coordinate system. To accomplish the inspection and facilitate an accurate comparison between the models, the registration process is required to align the scan and CAD models in a common coordinate system. The registration includes a virtual compensation for the flexible deformation of the parts in a free-state. Then, the inspection is implemented as a geometrical comparison between the CAD and scan models. This thesis focuses on developing automatic and accurate fixtureless CAI methods for non-rigid parts along with assessing the robustness of the methods. To this end, an automatic fixtureless CAI method for non-rigid parts based on filtering registration points is developed to identify and quantify defects more accurately on the surface of scan models. The flexible deformation of parts in a free-state in our developed automatic fixtureless CAI method is compensated by applying FE non-rigid Registration (FENR) to deform the CAD model towards the scan mesh. The displacement boundary conditions (BCs) for FENR are determined based on the corresponding sample points, which are generated by the Generalized Numerical Inspection Fixture (GNIF) method on the CAD and scan models. These corresponding sample points are evenly distributed on the surface of the models. The comparison between this deformed CAD model and the scan mesh intend to evaluate and quantify the defects on the scan model. However, some sample points can be located close or on defect areas which result in an inaccurate estimation of defects. These sample points are automatically filtered out in our CAI method based on curvature and von Mises stress criteria. Once filtered out, the remaining sample points are used in a new FENR, which allows an accurate evaluation of defects with respect to the tolerances. The performance and robustness of all CAI methods are generally required to be assessed with respect to the actual measurements. This thesis also introduces a new validation metric for Verification and Validation (V&V) of CAI methods based on ASME recommendations. The developed V&V approach uses a nonparametric statistical hypothesis test, namely the Kolmogorov–Smirnov (K-S) test. In addition to validating the defects size, the K-S test allows a deeper evaluation based on distance distribution of defects. The robustness of CAI method with respect to uncertainties such as scanning noise is quantitatively assessed using the developed validation metric. Due to the compliance of non-rigid parts, a geometrically deviated part can still be assembled in the assembly-state. This thesis also presents a fixtureless CAI method for geometrically deviated (presenting defects) non-rigid parts to evaluate the feasibility of mounting these parts in the functional assembly-state. Our developed Virtual Mounting Assembly-State Inspection (VMASI) method performs a non-rigid registration to virtually mount the scan mesh in assembly-state. To this end, the point clouds of scan model representing the part in a free-state is deformed to meet the assembly constraints such as fixation position (e.g. mounting holes). In some cases, the functional shape of a deviated part can be retrieved by applying assembly loads, which are limited to permissible loads, on the surface of the part. The required assembly loads are estimated through our developed Restraining Pressures Optimization (RPO) aiming at displacing the deviated scan model to achieve the tolerance for mounting holes. Therefore, the deviated scan model can be assembled if the mounting holes on the predicted functional shape of scan model attain the tolerance range. Different industrial parts are used to evaluate the performance of our developed methods in this thesis. The automatic inspection for identifying different types of small (local) and big (global) defects on the parts results in an accurate evaluation of defects. The robustness of this inspection method is also validated with respect to different levels of scanning noise, which shows promising results. Meanwhile, the VMASI method is performed on various parts with different types of defects, which concludes that in some cases the functional shape of deviated parts can be retrieved by mounting them on a virtual fixture in assembly-state under restraining loads. \n \nLe marché industriel compétitif exige une production de haute qualité de la part des compagnies de fabrication. Le département de contrôle qualité dans les secteurs industriels vérifie les exigences géométriques des produits en se référant aux tolérances. Ces exigences sont présentées dans les normes de Dimensionnement Géométrique Et Tolérances (DG&T). Toutefois, les méthodes conventionnelles de mesure et de dimensionnement sont couteuses et longues. De nos jours, les méthodes de mesure manuelles sont remplacées par les méthodes automatisées dites Inspection Assistée par Ordinateur (IAO). Les méthodes IAO appliquent les améliorations dans le calcul informatique et les dispositifs d’acquisition de données 3-D afin de comparer le maillage du modèle scanné de la pièce fabriquée avec le modèle conçu par ordinateur utilisant la Conception Assistée par Ordinateur (CAO). Les normes de métrologie, telles que ASME-Y14.5 et ISO-GPS, exigent la mise en oeuvre de l'inspection à l'état libre dans lequel la pièce est soumise uniquement à la gravité. Les pièces souples (non rigide) sont exemptées de la règle d'inspection à l'état libre en raison de l'écart géométrique significatif de ces dernières au dit-état tenant compte des tolérances. En dépit du développement des méthodes IAO, l’inspection des pièces souples demeure un sérieux défi. Les méthodes d'inspection conventionnelles appliquent des gabarits complexes pour récupérer la forme fonctionnelle des pièces souples. Cependant, la fabrication et la configuration de ces gabarits de conformité sont compliquées et chères. Depuis que les clients et les industriels exigent des gabarits d’inspection répétitifs et indépendants, le prix de ces derniers a doublé. Les méthodes d'inspection sans gabarit des pièces souples basées sur les méthodes IAO ont été développées afin d'éliminer l’utilisation couteuse des gabarits de conformité. Ces procédés visent à distinguer les déformations flexibles des pièces à l'état libre des défauts. Les méthodes d'inspection sans gabarits doivent être automatiques, fiables, précises et reproductibles pour les pièces souples aux formes sophistiquées. Le modèle scanné, qui est obtenu sous forme de nuages de points, représente la forme d'une pièce à l'état libre. Ensuite, l'inspection des défauts est réalisée en comparant les modèles scannés et CAO, mais ces modèles sont présentés dans des systèmes de coordonnées indépendants. En effet, le modèle scanné est présenté dans le système de coordonnées du système de mesure (système de numérisation) tandis que le modèle de CAO est dans le système de coordonnées de conception. Pour effectuer l'inspection et faciliter une comparaison précise entre les modèles, le processus de recalage est nécessaire afin d’aligner et ramener plus près les modèles scanné et CAO dans un système commun de coordonnées. Le recalage inclut une compensation virtuelle de la déformation flexible des pièces à l'état libre. Après, l’inspection est assurée par une comparaison géométrique entre les modèles CAO et les pièces souples scannées. La présente thèse porte sur l’élaboration de méthodes automatiques d'inspection assistée par ordinateur sans gabarit. Ceci constitue une amélioration de la méthode d'inspection numérique généralisée (Generalized Numerical Inspection Fixture (GNIF)). Cette thèse présente également la

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.