Solder joint inspection with multi-angle imaging and an artificial neural network

Machine vision has been widely deployed in many industrial applications. However, for solder joint inspection, it has yet to reach the desired maturity level. This paper presents a new inspection methodology using images from both orthogonal and oblique viewing directions to the solder joint. The oblique view was made possible through a mirror pyramid. Image capturing and selection of the soldered region were done manually, but could be automated if the positional coordinates were known. Combined orthogonal and oblique gray-level images at the pixel level were directly input to an artificial neural network (ANN) for processing, eliminating the need to determine heuristic features. Learning vector quantization architecture was used as the classifier. This study was focused on geometry-related joint defects, namely, excess and insufficient. Comparisons show that the oblique view provides more useful information compared to the orthogonal view. The experimental results indicate that the proposed system has an improved recognition rate and good resilience to noise.