Research On Recognition And Retrieval Technology Of Interface Image Of NiCrAlY Coating/Ni Based Alloy Based On Deep Learning

MCrAlY coatings(M=Co,Ni)are commonly used in turbine blade systems,and the life of the coating is critical to the safe operation of the equipment.In this paper,the image processing technology was applied to the image feature information recognition and retrieval of the micro-morphology,as well as the quantitative relationship model between the image feature information and the dynamic evolution law of coating service life during the service of NiCrAlY coating/Ni-based superalloy.This work emphasized the use of deep learning in the early exploration of materials field,enriched the NiCrAlY coating/Nickel-based alloy database,and realized the identification of material characteristics and the prediction of coating service performance.The specific research results were as follows:In the stage of image feature information recognition and retrieval,NiCrAlY coating/N5 alloy was studied.Based on the acquired cross-section feature image datasets,including 3600 images of 64×64 pixels,a deep learning technique was used to build a convolutional neural network(CNN)for classification and identification of the TCP phase,the interface between the substrate and the coating,and the oxide layer.The convolutional neural networks with two or three convolutional layers were respectively trained to classify,identify and locate these three characteristics.The test set accuracy of the neural network with two or three convolution layers using the RMSProp optimizer was 98%and 90.67%,respectively.The test set accuracy of the convolutional neural network with three convolutional layers using Adam optimizer was 99.17%,and the network performed best when testing 10 images of 1024×943 pixels.The retrieval accuracy of the three types of features reaches 100%on the complete picture.The same neural network was used for the feature recognition of NiCrAlY coating/K38G alloy,and the retrieval accuracy was over 80%on 1160 images of 1280×1024 pixels.In the stage of establishing quantitative relationship model between image feature information and dynamic evolution law of coating service life,NiCrAlY coating/K38G alloy was studied.Short-term constant temperature oxidation experiment had been made for 10,20,50,80,100,150,200 h at 1000,950,900 and 850℃.Based on the 1000 SEM images of 1280×1024 pixels cross-section feature image data,The OpenCV technology is used to build a quantitative relationship model between image feature information and dynamic evolution law of coating service life at different temperatures,including:1.Quantitative relationship model of dynamic evolution of second phase(Cr-rich phase)in service coating with service time.Median filtering in OpenCV library was used to realize denoising preprocessing on the images.The coordinate values of the retrieved interfaces and oxides are returned in matrix form to realize automatic recognition of the coating profile and automatic calculation of the coating area.The average of the area ratio of the second phase in the coating in several pictures were calculated as the final result.The results showed that the dynamic evolution of the Cr-rich phase in the coating as a percentage of the coating area as a function of service time were as follows:(1)950℃:w(Cr)=0.6618t-0.678(2)900℃:w(Cr)=-0.039 ln(t)+0.2248(3)850℃:w(Cr)=-0.0006t+0.1608.The dynamic evolution of depth between interface and Cr-rich phase in coatings with service time were as follows:(1)950℃:d(Cr)=-0.1232t+23.875(2)900 ℃:d(Cr)=-0.1162t+24.247(3)850 ℃:d(Cr)=-0.0771t+28.280.2.Quantitative relationship model for constructing oxide layer thickness with dynamic evolution of service time.The flooding,tiny hole filling,binarization methods in OpenCV library were used to extract the oxide layer profile.The irregular oxide layer area was counted and its shape was approximated into a rectangle.The ratio of the area to the long side of the rectangle was taken as the thickness(μm)of the oxide layer.A number of images were taken to calculate the average as the final result.The change trend of the thickness of the oxide layer is consistent with the change trend of the oxidation kinetic curve.The results show that the relationships between oxide thickness and time at different temperatures were as follows:(1)1000℃:H2=0.0816t+3.2641(2)950 ℃:H2=0.0301t+3.571(3)900℃:H2=0.0555t+0.5906(4)850℃:H=0.0133t+0.3596.The above research results showed that the initial evaluation of coating service life can be achieved by using CNN and OpenCV image processing techniques.