Evaluation of NARX Network Performance on the Maintenance Application of Rotating Machines

Rotating machines are widely used in many industries even though the maintenance of this equipment is high.Most of the industries employ preventive maintenance to avoid expensive failures and shutdown.From the view of maintenance application, a predictive approach utilizing real time data provide a better understanding of the system degradation.This paper presents the evaluation of ANN technique for a data driven model which is nonlinear autoregressive neural network design with exogenous input (NARX) for a maintenance application for rotating machines.The data used for the research are obtained from NASA Data Depository.Analysis was done to understand the data better.Before model training, a feature selection for the inputs have been made allowing dimensionality reduction narrowing the parameters consisting features that could make an impact on detecting failure.NARX neural network model from the neural net time series is chosen as it can adequately handle a time series data in a forecasting model network.NARX network is first trained in open-loop mode before being trained in closed-loop mode.Results show that the network model is suitable for time series data prediction. © 2023, Institute of Technology PETRONAS Sdn Bhd.