Apply Computer Vision and Neural Network to Glue Dispenser Route Inspection

Abstract

Inspection system design based upon the dispenser route is the focus in this article. The defects of the dispenser route such as deformation, offset, gap, and broken glue may affect the quality of production and efficiency. An automatic dispenser route inspection system in combination the techniques of back-propagation neural (BPN) network with computer vision is developed. The inspection system includes computer vision (image acquisition, binarization, dilation, erosion, Sobel operator, thinning and extraction features), positioning, and inspection. The images are acquired and then preprocessed to extract the features (coordinates of edge) of interest for inspections. Before dispensing, positioning process of the dispenser system is significant. A simple method of positioning can achieve the positioning accuracy in an allowable range is introduced. Thus, the cause-and-effect failure problem due to inaccurate positioning of the dispenser system needs not considered so that it will not influence the investigation of other factors, in particular the needle's condition. Extracting features of the captured image through a series of image processing procedures are evaluated. By checking the number of the searched pixels of the boundary of the dispenser route compared to the edge number of a uniform route, failure can be determined. For further diagnosis, six sets of parameters including the average width and its standard deviation(SD) of the dispenser route, average offset and its SD, and the average of deviation between the neighboring points on the left and right sides are designed as the input units in the input layer of a three-layer neural network. After training with amount of experimental patterns, the recognition rate of the neural network system is able to achieve 96.9 percent.