Intelligent Defect Detection and Adaptive Compensation Technology for Automation Devices based on Deep Learning

In order to improve the production efficiency of automation devices and ensure product quality, the SURF feature matching algorithm is utilized to crop out the automation device localization big picture and reduce the amount of computation. Combining deep convolutional generative adversarial network and Gaussian mixture model to generate defect images, YOLO-v3 network is utilized to realize intelligent detection of defects in automation devices. The adaptive compensation algorithm of sliding window and multi-frame compensation is used to utilize the information redundancy between multi-frame images, and the blind elements in the images are adaptively compensated to provide more accurate inputs for defect detection. The results show that the maximum recognition error values are 0.57 mm in depth, 0.8 mm in length, and 0.55° in angle, and all the error values are relatively small. In the defect category identification and detection effect is higher than other methods, the proposed method has the highest sensitivity of the scratch pit defect feature, which reaches 91%, and only 0.88 deformation variable is needed in the X image dispersion compensation performance, and the deformation is reduced by about 41 %. Remarkable results have been achieved in practical applications, providing strong support for the development of industrial automation field.