The inspection of milled metallic surfaces has to solve the problem of distinguishing between real defects – such as pores – and other deviations – such as dirt or changes in colour. This is not possible with conventional 2D image acquisition that is only focused on grey values and it lacks the information of depth. Additional sensor modalities provide such complementary information and enable accurate decision about defects.
Machine learning is used to transfer expert knowledge into the system’s software in order to make the correct decision.
TPScan is a technology for the detection of pores, scratches and other defects on the surface. The specific strength is the accurate separation of real defect and other deviations of the surface such as dirt or discoloration that should not be judged as defects. The technology is specifically designed for crankcases and cylinder-heads, but can be adapted to any kind of even surfaces.
Machine learning is used to classify the different types of defects and to learn this classification from a human expert. A specific feature is an “active learning” approach that minimizes the training input that is needed from the expert.