Direct Remaining Useful Life Estimation Based on Random Forest Regression
Xin Chen, Ge Jin, Siqi Qiu, Minglei Lu, Danjiong Yu
Abstract
Remaining useful life (RUL) prediction plays a key role in the domain of prognostics and health management (PHM). It could increase the industrial or mechanical systems' reliability and safety while reducing their maintenance cost. Traditional RUL processes rely on the conception of health indicator, the calibration of degradation curve and the estimation of failure threshold. Some recent works apply deep learning-based methods in RUL, despite their high performances, these models are computationally expensive and less intuitive in explanation, causing difficulties to implement in practice. Some other approaches rely on traditional machine learning algorithms like SVR, which are less precise compared to deep learning-based models. In this paper, we propose a direct and high-quality remaining useful life estimation method based on random forest regression. Extensively, a feature selection algorithm based on Lasso regression is embedded in the pipeline. Our experiments on the Turbofan dataset issued by NASA show that the proposed approach is as good as, if not better than, the state-of-art performance of deep learning-based models, while remaining computationally efficient.