An explainable approach for prediction of remaining useful life in turbofan condition monitoring

Abstract

The goal of this research is to present an approach that predicts the remaining useful life (RUL) estimation of a turbofan engine system within specific time frames of aircraft operation. In the first phase, due to the high number of features, a group decision-making approach for feature selection is proposed and performed based on various supervised and unsupervised methods. In the second phase, among various machine learning and deep learning approaches, bidirectional long short-term memory (BiLSTM) and multilayer perceptron (MLP) that provided more appropriate results are selected. Unlike most previous studies that focused on determining the status of the equipment (whether it is healthy or faulty), the main objective of this research is to predict RUL. Finally, in the third phase, thorough analysis using explainable AI (XAI) was conducted concerning the importance of features and an investigation into features that led to an increase in errors in some intervals. The results show that the presented approach has been able to predict the RUL well, although it is biased in some time intervals for all turbofans, and the features related to this bias prediction are determined by using XAI. They can be further investigated by the maintenance department. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2024.