Research On Diagnosis Method Of Workpiece Minor Defects Based On Deep Learning

In the manufacturing process of workpiece,different kinds of defects often appear on the surface of workpiece due to factors such as production environment and manufacturing technology.Surface defects of the workpiece directly affect the quality and performance of the workpiece.The traditional manual detection method was affected by personal subjective factors and mental state,and to some extent,the detection efficiency and quality on the production line couldn't be guaranteed.The automatic defect detection technology based on traditional digital image processing was widely used in the field of defect detection because of its fast detection speed,low labor cost and stable and reliable detection results.However,some products have complex texture features,making it difficult for traditional digital image processing techniques to extract appropriate feature vectors and cannot be effectively detected.Convolutional Neural Networks(CNhN)have been successfully applied in different fields by using their own learning to obtain features.In this dissertation,the metal screws on the chassis are used as research carriers,and the method for detecting micro-defects on the surface is deeply studied.The main work of this dissertation is as follows:Firstly,the related visual hardware such as camera,lens and light source is selected,and a set of defect detection hardware system is designed to obtain high-quality workpiece images.The image graying and image threshold segmentation algorithm are used to separate the screws in the workpiece image from the background,and the screw regions are further separated by image morphology.An extraction algorithm based on area of region contour is proposed to locate the detected target.Secondly,the advantages and disadvantages of the traditional target recognition algorithm are studied.The limitations of the traditional target recognition algorithm for the detection of small defects in screws are verified by experiments.Combined with the structural characteristics of the convolutional neural network model,a method for detecting fine defects of workpiece based on improved deep convolutional neural network is proposed.Third,under laboratory conditions,the defect detection method proposed in this paper is compared with other target detection algorithms.The experimental results show that the detection method of workpiece micro-defects based on region contour area extraction algorithm and improved deep convolutional neural network can effectively solve the problem of detecting micro-defects on the surface of screws.In terms of detection accuracy,this method can reach 97.1 7%on average,which greatly improves the detection rate of micro-defects of workpiece.In terms of detection speed,the average detection speed of this method is 0.3 seconds,which is not as fast as the traditional detection method,but it has met the real-time requirements of industrial detection.In summary,this dissertation uses the metal screws on the chassis as the research carrier,and based on the combination of the area contour area extraction algorithm and the improved deep convolutional neural network workpiece micro-defect detection method,the high-speed and high-accuracy detection results are obtained.It is of great significance for the research of micro defect detection of workpiece.