Optical Quality Control for Adaptive Polishing Processes

We propose an image-based method to automatically estimate the surface roughness of a polishing process carried out by a numerically controlled machine tool. Given a single photograph of the workpiece, we incorporate techniques from differentiable rendering to infer the object's roughness parameters, resulting in several advantages over existing approaches: since the method fully accounts for global light transport effects, the estimation can occur under general, known lighting conditions and workpiece geometries. This allows deployment of our approach for in-situ measurements by simply equipping the machine tool with a standard digital camera capturing photos of the workpiece. We investigate the feasibility and effectiveness of our novel method in a prototype application considering polished brass plates. Our results demonstrate a promising direction for surface parameter measurement in less restricted polishing process environments.