Raytracing-generated images from surface metrology data for AI training on defect recognition

We present a process workflow to generate realistic images of industrial products via standard raytracing tools. The optical properties of industrial samples, like the BSDF and the surface roughness, are measured in lab and implemented in a raytracing software. The main application of this technique is to provide images of common surface defects of a variety of industrial components, under varying illumination conditions and from different viewpoints. The final goal is a library of computer-generated images to be used for training an AI model in automatic recognition of superficial defects. Generating the training images via optical simulations saves a considerable amount of time and offers much more flexibility than, for instance, taking multiple pictures of real products with an automated defect recognition system (like a camera mounted on a robotic arm).