Prediction Model Of SCC Crack Growth Rate For Nuclear Power Structural Materials Based On Deep Learning

Stress corrosion cracking(SCC),a serious phenomenon that can occur in nuclear welded structures during service,is one of the most important factors threatening the safe operation of nuclear power plants.The phenomenon is widespread and can lead to many safety and economic problems as a result,so it is necessary to study the rate of stress corrosion cracking growth.Most of the current research on the prediction of SCC crack growth rates is based on traditional models,i.e.quantitative and empirical formulas obtained by calculation,which are based on mechanical or electrochemical theories,while the interaction between different factors is not taken into account.This thesis proposes a research model for the prediction of SCC crack growth rate based on deep learning,constructs a 1D-CNN-LSTM network model based on the fusion of convolutional neural network(CNN)and long short-term memory network(LSTM),and validates it using a single LSTM and MLP neural network in comparison with it.The main research work of the paper is as follows.This thesis introduces the theoretical basis of deep learning,and focuses on sorting out and analyzing the relevant conceptual formulas of CNN and LSTM.Stress corrosion tests were carried out on Alloy 600 using stress corrosion testing machine to simulate the high temperature and high pressure water environment of nuclear power primary circuit,and the effect of stress intensity factor on SCC crack growth rate was verified.The deep learning model for 304SS and Alloy 600 was established,namely,the prediction model of SCC crack growth rate based on LSTM.The influence of different optimization functions on the prediction model was analyzed,and Adam optimizer was used to optimize the training process of LSTM network model.In order to further improve the prediction progress and stability,the 1D-CNN-LSTM fusion network model was established based on the LSTM prediction model and CNN with better information feature extraction was added,and the same optimizer was used to partition the completed data set for training.The predicted results of LSTM were compared with those of 1D-CNN-LSTM.A MLP neural network SCC crack growth rate prediction model was established for 304SS and Alloy 600.The transmission function of the hidden layer was "tansig",the transmission function of the output layer was "purelin",and the training algorithm was"trainlm".The prediction results of MLP neural network model are compared with those of 1D-CNN-LSTM.