A Neuro-fuzzy Approach to Machine Vision Based Parts Inspection

This paper documents progress on a project whose objective is to improve the performance of a machine vision based parts inspection system through the development and testing of robust neuro-fuzzy based algorithms. An inspection problem faced by a Canadian automotive parts manufacturer is being used as a case study. The problem involves a vision system that is being used to confirm the placement of metal fastening clips on a structural member that supports a truck dash panel. It took the manufacturer over 8 months to tune their commercial machine vision system to detect missing clips. It is hypothesized that a neuro-fuzzy based approach could provide for faster tuning of their vision system. Preliminary results show strong performance of the neuro-fuzzy system and a new algorithm is being developed on this basis to automatically learn the inspection process from a series of training images