Research On Location And Quality Inspection Of Enameled Wire Spot Welding Based On Machine Vision

Enameled wire is the most important material for the connection of electronic components.Both the welding operation and quality inspection were completed by manual for the traditional spot welding of enameled wire.The welds is difficult to locate accurately during the resistance welding process because of the tiny size of enameled wire,resulting in various welding quality problems.Manual quality inspection is also very easy to cause fatigue of quality inspectors,which bringing a certain degree of difficulty to the quality inspection of welds.Aiming at solving these problems,this paper applies machine vision technology to automatic spot welding location and quality inspection of enameled wire.From the perspective of the composition and principle of the machine vision system,this article introduces the selection method of the camera,lens and light source,and selects the appropriate camera and lens considering the enameled wire and pad size,inspection accuracy and other factors;considering the color of enameled wire and pad and the advantages and disadvantages of various lighting methods,chooses the color of the light source and the illumination method.The machine vision system was built according to the selected camera,lens and light source.This paper analyzed the possible conditions of poor relative position during the automatic assembly process of enameled wire and the influence of these bad conditions on welding quality.Image processing methods and recognition algorithms are used to eliminate the bad conditions of workpiece.A stable and reliable algorithm is designed to extract the position of best welds after the elimination of workpiece in bad conditions.The assembly condition recognition and locate algorithm can avoid the occurrence of quality problems,and the time required for recognition and extraction is short,which can meet the requirements of automatic spot welding of enameled wire.The welds images of enameled wire were collected on the actual production line,which were divided into dataset.The CNN-SJ model was constructed by using convolutional neural network,which was consists of 4 convolutional layers.In order to suppress the over-fitting phenomenon,the CNN-SJ model is improved by using the Batch Normalization and the L2 regularization method,the improved model was named after CNN-SJ3.The CNN-SJ3 model suppressed over-fitting to a certain extent,and the classification accuracy is increased to 98.2%.DenseNet-SJ,a new enameled wire spot welding quality inspection model,was constructed by the pre-trained DenseNet-121 model using Transfer Learning.DenseNet-SJ model had a better classification performance than CNN-SJ3 model.The results show that the classification accuracy of the trained DenseNet-SJ model is as high as 99.8%,there is no over-fitting phenomenon,the average detection time is 25.3ms,which means that the task of welds quality inspection can be achieved effectively by DenseNet-SJ.