Research On The Casting Defeat Recognition Method Based On Deep Learning

Casting is a common product in the industrial field.Due to its wide range of applications and high demand,the quality of the casting product is particularly important.Therefore,the application of X-ray casting defect detection has also received special attention.At present,the evaluation of X-ray images of castings relies heavily on manual labor.This manual inspection method is gradually difficult to meet the production requirements of modern automated casting production lines due to its low efficiency and high rate of missed judgment.Although the existing casting defect detection technology can complete defect identification,most of the technologies are based on artificially designed image features,and it is difficult to realize the versatility of multiple defect features.With the development of deep learning,a fast and intelligent technical approach is provided for the above problems.For this reason,this thesis combines aerospace light alloy casting defect detection images to propose a deep learning-based casting defect recognition method.The main research contents are as follows:(1)Combining the characteristics of image data,this thesis proposes a casting defect detection method based on YOLO idea.Aiming at the characteristics of large image resolution of castings,small amount of defect data,multi-scale defects,and high proportion of small defects,this thesis uses data augmentation based on Overlap cutting to expand small defects,and further enhances the complexity of the image based on simplified Mosaic data augmentation,so that the entire training set is more suitable for small defects;on this basis,multi-scale feature extraction is realized based on the YOLO idea,which builds a casting defect detection network,and the k-means++ algorithm is used to redesign Anchor for castings Image.At the same time,in order to adapt to the model training method based on the overlap cut image,this thesis proposes a test image defect detection method based on bounding box suppression,which completes the defect detection in the test image in a subimage iteration method.(2)In order to further improve the detection efficiency and robustness of the defect detection method,this thesis improves the detection network based on the proposed defect detection method.First,the convolution operation in the network is replaced with a Depthwise separable convolution,and then the inverse residual structure and the SE block with improved activation function are used to realize the lightweight design of the detection model.At the same time,in order to ensure the detection effect of the detection network,this thesis also uses DIo U loss and Focal loss to improve the positioning loss and confidence loss of the network.In view of the above research content,this thesis has carried out research and experiments based on the X-ray inspection images of aerospace light alloy castings,which can realize the rapid identification of various casting defects,which verifies the effectiveness of the method in this thesis.