Failure and Reliability Predictions by Infinite Impulse Response Locally Recurrent Neural Networks

Abstract

In this paper, Infinite Impulse Response Locally Recurrent Neural Networks (IIR-LRNNs) are employed for forecasting failures and predicting the reliability of engineered components and systems. To the authors’ knowledge, it is the first time that such dynamic modelling technique is used in reliability prediction tasks. The method is compared to the radial basis function (RBF), the traditional multilayer perceptron (MLP) model (i.e., the traditional Artificial Neural Network model) and the Box-Jenkins autoregressive-integrated-moving average (ARIMA). The comparison, made on case studies concerning engine systems, shows the superiority of the IIR-LRNN with respect to both the RBF and the ARIMA models, whereas a similar performance is obtained by the MLP.