Prediction Of Automotive Stamping Part Defects Based On Deep Learning

With the rapid development of China’s automobile industry,the demand for automotive stamping is increasing.Automotive stampings are the basis composition of the body and require high surface quality.Whether it can produce high-quality stamping parts is related to the quality of the entire vehicle.Traditional methods cannot predict product defects in real time in the early design period,which can makes the product defects surge in subsequent design and processing process.The deep learning method can realize real-time prediction of product defects in the early design formability analysis of products,so as to truly reduce the probability of product defects.In deep learning,the convolutional neural network(CNN)has a human brain-like visual processing mechanism and does not require complicated pre-processing of images,so it has a great performance advantage in the field of target detection and pattern classification.In order to analyze the forming process in the early stage of the product,this paper proposes a automotive stamping defects real-time prediction method based on Faster R-CNN.At present,there is no public dataset on the prediction of defects in automotive stampings at home and abroad.Therefore,the stamping dataset is firstly constructed in this paper.The data set mainly includes three parts of the stamping thickness nephogram calculated by KMAS/One-step,the Gaussian nephogram of the stamping part and the label file generated by using the thickness nephogram.This paper uses the stamped Gaussian nephogram and label file as training set,validation set and test set for Faster R-CNN(VGGNet-16).The data set was used to train and test the models in the literature.The test results showed that the mean average precision(m AP)of the stamping part defects prediction was only 56.2% for the Faster R-CNN.In order to solve the problem of low accuracy of Faster R-CNN model,this paper selects Res Net-101 as the feature extraction network of Faster R-CNN and proposed a solution to change the size of the anchor box.The final test results show that mean average precision of the automotive stamping defects prediction is 67.19%,among them,the average precision of split reached 76.12%,and the average precision of wrinkling was 61.25%.It is proved that this scheme effectively improves the average precision of the Faster R-CNN model.Finally,using the trained Faster R-CNN(Res Net-101)model to real-time predict defects of the stamping in the video,solve the problem of real-time prediction of stamping defects.