Reliability Analysis of Multi-Stress Accelerated Life Test Based on BP Neural Network

Abstract

To construct multi-stress accelerated life model, the traditional method utilizes the observed value acquired in the process of accelerated life test to build likelihood equations of accelerated model, however, the excessive parameters of multi-stress accelerated model will lead to the difficulty in solving pluralism likelihood equations. Based on the predictability and convergence of genetic algorithm optimum BP neural network, the multi-stress accelerated life model of genetic neural network is built. Take the level and reliability of accelerated stress in the accelerated life test as neural network input, then draw the scatter diagram by software and use the nonlinear least square to fit raw data to obtain the regression equation. Consequently, generate large quantities of test data, which shall be input into the neural network and optimize the weight and threshold value by genetic algorithm. By this means, conquer the blindness in selecting the original weight and threshold value. Finally, input constant stress and set reliability into well-trained network, thus get the predicting curve of reliability. Simulation results show that the method above is efficient and practical.