Improvement of Data-Driven 3-D Surface Quality Inspection by Deformation Simulation

The automated visual inspection of complexly shaped, stamped sheet metal is a challenging task. Model-based methods are currently state of the art for surface quality inspection based on three-dimensional surface measurements. Data-driven models allow the representation of arbitrary three-dimensional surface shapes including possible tolerances and variation due to deformation. Providing a sufficient amount of training data while ensuring adequate data quality is a time-consuming and laborous process. We propose a data driven three-dimensional surface model for quality inspection that allows the integration of deformation simulation in order to increase variance in training data. The effects of simulated data to the model quality are evaluated and results are given for data sets of automobile sheet metal parts.