Estimating the Remaining Useful Lifetime of a Turbofan Engine using Ensemble of Machine Learning Algorithms

Abstract

Predicting the Remaining Useful Lifetime (RUL) of a component or system is important for effective and efficient maintenance. Prognostics approaches, used in predicting the future reliability of a system by assessing the extent of degradation of the product from its expected normal operating conditions, can be classified into physics-based and data driven. The later has received huge attention from researchers as it does not require expertise knowledge of the system at hand. Ensemble technique, associated with aggregation of predictions produced by multiple learning algorithms to improve predictive performance, robustness and accuracy in prognostics is the area of interest for this research that ensembles various selected regression Machine Learning Algorithms (MLA). The effect of ensembling various MLAs and MLAs models built using similar data is presented and a comparative study done with using only a single model. A case study of a NASA turbofan engine degradation simulation data set is presented. Simple averaging approach is proposed in combining the output of different sub models both at the training and predictive stages. Of the selected MLAs, ensemble regression, the best performing, had a Mean Absolute Error (MAE) of 39.63 for a single model compared to 22.91 for an ensemble of various sub models. The numerical results indicate that the ensemble approach outperforms use of individual machine learning models. Keywords— Ensemble technique, Machine Learning, Prognostics, Remaining Useful Lifetime.