Reliability Estimation of Electric Parts by Using Neural Prediction Model

Abstract

Reliability prediction of electric parts by using neural network is described. A mean time to failure (MTTF) in reliability test for carbon film registor is adopted and predicted in this paper. In general, engineer's experienced decision making plays an important role in reliability problems. Therefore, intelligent information processing techniques are required in the field. A neural network is an attractive technique in many fields of reliability problems because the neural network can learn complicated relationships among the factors. This paper describes some experimental results of reliability prediction using a neural network and comparative study on the accuracy between conventional auto-regressive model and the neural network. By using the time series of the mean time to failure in reliability test, we attempt to find an appropriate model for predicting the value of mean time to failure in twenty hours in advance, in order to establish a convinient method for reliability estimation. We first discuss some essential issues to be considered in reliability problems and AI techniques. It is shown that the neural prediction model gives a better prediction results compared with the statistical model. Furthermore, the robustness of each model is indicated by applying to noise confused data of mean time to failure.