High precision dimensional analysis of manufactured parts: synthetic images and knowledge help image data evaluation and interpretation

摘要

Computer vision based evaluation of industrial workpieces can make efficiently use of knowledge based approaches, in particular for quality control and accurate measurement tasks. One of the possible approaches is to compare real images with conceptual (synthetic) images generated using standard CAD models and taking the application specific conditions into account. Integrated in (industrial) real-life environments, our developed hardware and software have been applied, amongst others, to non-standard surface inspection tasks requiring extremely high processing rates and to robotics. Specifically, inspection relies on the comparison of real and conceptual images, using measured calibration data, CAD models including tolerances and measured acquisition parameters for the generation of the conceptual images. The resulting evaluation methods have been successfully applied to the vision based on-line inspection of manufactured parts including sculptured surfaces (e.g. turbine blades). In particular, the reconstruction of 3D shapes is demonstrated using structured light techniques. As a result, an accuracy of the order of 25pm can be routinely achieved for the reconstruction using either isolated images or spatially registered image sequences. Also, the design, realization and use of a controlled acquisition system (measuring robot) will be shortly presented. This device should facilitate the controlled feature extraction out of the images, in view of a fine modelization of the image contents and application to dimensional measures and 3D reconstruction of rigid or non-rigid objects.