Using neural network with convolution layer for automatic quality inspection,

Abstract

At traditional assembly workstations product inspection is performed manually by operators and that takes a lot of time and represents a production bottleneck. Also, due to the appearance of mental fatigue, a drop in concentration in some situations, it is almost impossible for workers to notice the appearance of defects and irregularities. Therefore, in modern assembly systems, the implementation of advanced quality control through quality inspection, visually detect the quality of a product and recognition of irregularities has a crucial role. The main aim of this study is to research the possibility of involving neural networks for defects’ detection of parts in the production process. The research includes using images of two classes (part with defect and well done part) and training the neural network with convolution layer for automatic classification produced part. The success of applied algorithms using this methodology in automatic detection of defects and non-conformity and on this way reducing cost is 96 %. One benefit of the proposed method is the relatively small number of input data set images which enables fast implement method with new production elements.