A Comparative Study of Machine Vision Based Methods for Fault Detection in an Automated Assembly Machine

Fault detection and classification in an automated assembly machine using machine vision (MV) based inspection methods is the subject of this paper. A high speed automated assembly machine is used as the test apparatus. The machine is designed to assemble circular O-rings, from a bulk supply, onto continuously moving carriers at a rate of over 100 assemblies per minute. Video data was collected for both normal and abnormal machine conditions, and in particular transfer track jams. Three MV classification methods were adapted to this application and subsequently tested: 1) Gaussian Mixture Models (GMMs) with blob analysis, 2) optical flow and 3) running average. The methods are compared with a previously developed fault detection method based on spatiotemporal volumes (STVs). It is observed that the new methods require less training and processing time and are able to detect faults faster than the STV method. Amongst the three new methods, the running average method is shown experimentally to be the best in terms of having the lowest processing time per frame and the fastest response time. The work continues in order to see how the methods perform as different machine faults are introduced.