A robotic surface inspection framework and machine-learning based optimal segmentation for aerospace and precision manufacturing

Quality control is key in the advanced manufacturing of complex parts. Modern precision manufacturing must identify and exclude parts with visual imperfections (e.g., scratches, discolorations, dents, tool marks, etc.) to ensure compliant operation. This inspection process – often manual – is not only time-consuming but also burdensome, subjective, and requires months to years of training, particularly for high-volume production operations. A reliable robotic visual inspection solution, however, has been hindered by the small defect size, intricate part characteristics, and demand for high inspection accuracy. This paper proposes a novel automated inspection path planning framework that addresses these core hurdles through four innovations: camera-parameter-based mesh segmentation, ray-tracing viewpoint placement, robot-agnostic viewpoint planning, and Bayesian optimization for faster segmentation. The effectiveness of the proposed workflow is tested with simulation and experimentation on a robotic inspection of heterogeneous complex geometries.