In-line geometric fault detection in car parts based on structured light projection and image processing

摘要

An inspection system is developed to replace manual inspection in a production line for car parts. The system, based on projected structured light, combining Gray code and phase shifting and using B/W CCD cameras and multi-media data projectors, provides robust height measurement images with a high resolution. By carefully observing a number of parameters, it is possible to attain this high resolution in a large measurement volume even with low-cost, off-the-shelf components. We have been able to achieve a noise floor in the phase determination of 30 mrad, which is better than the much reported 1 part in 10,000. The use of 4 cameras, 3 projectors and a turning operation allows total coverage of the complex shape part. A model of normal parts is designed using height measurement images of normal parts. This model represents both expected part dimensions in all camera views as well as normal variations. In order to compare measurements of new parts with the model, an alignment of the images is performed. The deviations between the measured part and the model are analyzed. Deviations outside the normal variation are classified as faults. The system is thus able to find geometrical faults as small as 2x2x0.25 mm in a part that measures roughly 400x400 mm and can decide whether or not to remove a part from the production line. Integrating optical metrology, image processing and robotics, we are able to design a complete system for in-line inspection of car parts with total coverage that is able to keep up with the production cycle time.