Microstructure Identification Of High Strength Steel Materials Based On Texture Recognition And Deep Convolution Neural Network

With the rapid increase in car ownership,the automotive industry is increasingly demanding energy-saving emission reduction and environmental pollution reduction.The high-strength steel hot stamping technology is increasingly used in the manufacture of structural parts and safety parts for automobiles.High-strength steel can customize parts of certain mechanical properties by obtaining different microstructures through hot stamping.Possible microstructures include all martensite,ferrite/pearlite,martensite/residual austenite,martensite/bainite,and so on.In related research,it is necessary to identify the microstructure of the material to analyze the process,and the process needs to be completed by professionals,so there are disadvantages such as inefficiency,subjective influence of personnel and limited number of professionals.Therefore,it is necessary to study the automatic identification technology of the material microstructureFor the identification and classification of the above four types of organizations,studies are carried out on texture feature extraction,machine learning method classification and convolutional neural network model,and two classification methods are proposed:(1)Extracting microscopic by gray level co-occurrence matrix.The texture features of the image form a texture feature vector and the representation of the feature vector is visualized.Then,thess machine learning methods SVM,kNN,RF are trained on the data set to obtain the classification model,and the classification effect of each model is compared and analyzed.(2)By the idea of transfer learning,the deep convolutional neural network model was constructed based on the VGG-16 network,and the hyperparameters in the network training were explored and determined.Finally,the recognition effect of the deep network model on the microstructure of the material was analyzed,compared with those of the machine learning method.Through experimental verification,the two high-strength steel microstructure identification methods proposed in this paper got good scores in accuracy and have good classification effect.The main conclusions are as follows:(1)Image feature vector based on GLCM can well characterize the texture features of high-strength steel microstructure images,which can basically distinguish the four types of microstructures studied in this paper.(2)In the machine learning classification method based on GLCM feature vector,the classification accuracy of SVM is the highest,reaching 95.50%;and the recognition rate of SVM for all four types of microstructure s is above 90%.(3)In the image recognition model of the improved deep convolutional neural network,the more convolutional layers participating in the training,the lower the classification accuracy;when the number of convolutional layers participating in the training is 0,the classification accuracy is the highest,being 96.25%,exceeding the GLCM+SVM method.