Research On Intelligent Prediction Method Of Thermal Barrier Coating Life For Aeroengine Blades

Aeroengine thermal barrier coating(TBC)is a key protective material which works in a bad environment.Once it is peels off and fails,it will cause local overheating and burning of metal parts,which will seriously affect flight safety.To study the failure mechanism of TBC and effectively predict its service life is the key to ensure the safe service of aero-engine blades.Focuses on the inefficiency of existing TBC life research methods that mainly rely on manual experience and formula derivation,introducing deep learning into TBC life in this thesis by studying the TBC interface morphology characteristics,extracting the life information contained therein,and using(Convolutional neural network)CNN to realize the judgment of the life of TBC.the following works are mainly completed:(1)Firstly,Adap-Alex algorithm is proposed to solve the problem that CNN network has too deep structure and too many parameters,which leads to long training time and easy over-fitting when training TBC images with complex morphology and tissue structure.Adap-Alex adjusts the size of receptive field,step size and other parameters to get its structure based on the Alex-Net structure;Secondly,aiming at the problem of low learning speed caused by fixed learning rate,the optimization methods of different learning rate parameters are compared and the optimization methods are obtained;Finally,Finally,an adaptive pooling method is designed to improve the accuracy.(2)The test data set is used to test the improved algorithm.The test results show that:On the MNIST data set,the Adap-Alex structure trains 1009s faster than Alex-Net and 1981s faster than VGG-Net.The test accuracy is 9.81%higher than Alex-Net and 4.15%higher than VGG-Net;On the CIFAR-10 data set,the Adap-Alex structure trains 1002s faster than Alex-Net and 2218s faster than VGG-Net.The test accuracy is 13.77%higher than Alex-Net and 3.33%higher than VGG-Net.(3)The TBC data were obtained through thermal vibration experiments,and the TBC data set was made after data processing.Using the improved algorithm to verify the effectiveness of the improved algorithm on the self-made TBC data set,and the results show that:When the algorithm type is the same,the more the number of thermal vibrations,.the higher the test accuracy.Adap-Alex’ s training time is significantly shorter,125s faster than VGG-Net and 685s faster than Alex-Net;When the number of thermal vibrations does not change,the test accuracy of Adap-Alex is higher than that of Alex-Net and VGG-Net.Adap-Alex is easier to identify the coating,and when the number of thermal vibrations is 300 times the result was the best,with an accuracy rate of 93%.