Research On Image Recognition With Respect To Grinding Burn Of TC4 Titanium Alloy Based On Deep Learning

The high temperature generated during the titanium alloy grinding process can easily cause grinding burn on the surface of the workpiece,reducing the wear resistance,corrosion resistance and fatigue strength of the workpiece,which seriously affects the performance of the workpiece.In order to ensure the quality of workpiece processing and improve the production efficiency of workpieces,an efficient grinding burn identification method is urgently needed in production.The deep learning-based titanium burn image recognition method can reduce image preprocessing,avoid manual feature extraction,improve burn recognition accuracy,and achieve non-destructive rapid detection of grinding burn.Based on deep learning theory,this paper takes TC4 titanium alloy as the research object,conducts in-depth research on the image recognition of grinding burn and lays a foundation for the efficient identification of grinding burn of titanium alloy.The main work and research results of this paper are as follows:1.The titanium alloy grinding tests were used to obtain samples with different degrees of grinding burn.At the same time,the paper completed the design of the image acquisition module,including the choice of light source type,lighting type and industrial camera.Then the collected burn images were identified and pre-processed,and finally the burn data were augmented by using geometric transformations such as rotation,translation,and crop.TGBC-GD and TGBC-ID datasets were developed for model training and test.2.Based on the convolutional neural network,a titanium alloy grinding burn recognition network TGBNet was designed,and the model was optimized by using convolutional layer stacking,global average pooling,and Dropout.Based on the surface characteristics of titanium alloy grinding burn,the effects of model parameter initialization,optimizer,batch normalization,and batch size on the model's convergence speed and stability were studied,and the classification performance of the model on different degrees of burn was analyzed.3.Based on the model migration method,the image recognition of titanium alloy grinding burns was carried out in two ways: constructing support vector machines and constructing new fully connected layers,from the perspective of classifier construction.Design experiments were performed to analyze the classification effect of burn with different degrees.At the same time,the classification results of different models were compared.It was found that the model-based burn identification method had better classification performance in the case of insufficient datasets.Finally,the feature map visualization method was used to analyze the reasons for the difference in recognition performance of different burn recognition algorithms.4.A titanium alloy burn image recognition system was developed,including the design of image acquisition subsystem,model training subsystem and burn identification subsystem.Meanwhile,in order to reasonably allocate computing and storage resources,nine functional modules including a control module and an image acquisition module were designed to realize the image recognition of titanium alloy grinding burn.