Used automated image capture of parts from manufacturing lines to gather high-resolution images. Images that contained defects such as, scratches, gouges, burns, tears, and material defects, or other anomolies were then used to train a machine learning algorithm. Once trained, the machine learning system was able to identify and flag parts that had any defects.

Finally, the defective parts were triggered for additional review by an operator, automatic rework, or disposal. This concept can be used for a large variety of parts and assemblies including pipe threads, plastic injection parts, CNC parts, supply materials, and more.

Machine learning provided a great tool for identifying known problems, however, some scenarios were challenging to reproduce. In order to create more learning cases were able to create mocked defects using photo editing software which allowed for improved defect detection with a shorter learning period.