Boosting Regression Assistive Predictive Maintenance of the Aircraft Engine with Random-Sampling Based Class Balancing

Abstract

This study presents the development of a data-driven predictive maintenance model in the context of industry 4.0. The solution is based on a novel hybridization of Remaining Useful Life (RUL) generation, Min-Max normalization, random-sampling based class balancing, and XGBoost regressor. The applicability is tested using the NASA's C-MAPSS dataset, which contains aircraft engine simulation data. The objective is to develop an effective and Artificial Intelligence (AI) assistive automated aircraft engine's RUL predictor. It can maximize the benefits of predictive maintenance. The rules based RUL generation provides a ground truth for evaluating the performance of intended regressors. The Min-Max normalization linearly transforms the intended dataset and scales the multi subject's data in a common range. The imbalance presentation among intended classes can lead towards a biasness in findings. This issue is intelligently resolved using the uniformly distributed random sub-sampling. Onward, the performance of robust machine learning and ensemble learning algorithms is compared for predicting the RUL of the considered aircraft engine by processing the balanced dataset. The results have shown that the XGBoost regressor, uses an ensemble of decision trees, outperforms other considered models. The root mean square error (RMSE) and mean absolute error (MAE) indicators will be used to evaluate the prediction performances. The devised method secures the RMSE value of 12.88%. It confirms a similar or better performance compared to the state-of-the-art counterparts.