Optimization of the fatigue life of Epoxy Molding Compounds based on BP neural network prediction model

Based on the data from the fatigue test of Epoxy Molding Compound (EMC), firstly with a focus on the application problem of the instability between fitting and prediction error of BP neural network (BPNN), the prediction model of fatigue life for EMC materials is established. In this approach, the network structure is improved with initiative way by reducing input from the perspective of nodes with principal component analysis (PCA). Secondly, in order to deal with the problem of the bottleneck in local flow minimum of BPNN, this study tries to find out the global minimum and improves the convergence performance of the BPNN combining genetic algorithm (GA). The stability and practicality of the GABPNN model is analyzed after training and verifying, and the effect of the input factors on the output factor is studied in turn. Finally, this paper makes use of well-trained GABPNN prediction model to analyze optimum design methods of parameters to predict the fatigue life. The prediction and optimization results show that the well-trained GABPNN model can be used in the forecasting and optimizing design of the fatigue and fracture reliability of the epoxy molding compounds, and is of much practical value.