Health-Index Based Prognostics for a Turbofan Engine using Ensemble of Machine Learning Algorithms

Abstract

A turbofan engine is a critical component of the aircraft, and monitoring its performance is important to avoid catastrophic failures and expensive downtime. Technologies in condition monitoring have made this possible by using sensors to collect data regarding fault propagation in systems. Machine Learning Algorithms (MLA) are useful tools for data analytics modeling. They use features from datasets to detect patterns and build predictive models. The predictive models are then used with new data, to determine the future reliability of a system by assessing the extent of degradation from its expected normal operating conditions. This in turn facilitating determination of the system's Remaining Useful Life (RUL). Several prognostics approaches have been proposed to predict RUL for complex systems. There is a need to further increase their accuracy and robustness, with the aim of increasing reliability. This can be achieved by use of ensemble techniques. Ensemble of predicting models developed using different MLAs or models developed using similar datasets are some of the ensemble techniques used in RUL modeling. Their results have demonstrated to achieve better performance compared to single modeling. This work aims at further increasing the prediction accuracy and robustness by combining these two ensemble techniques. A case study based on the National Aeronautics and Space Administration (NASA) turbofan engine degradation simulation dataset FD001 is presented. Evaluation results demonstrate that the developed ensemble model had better performance having a score value of 115. This is in comparison to the best approach in literature using similar dataset, where modeling was done using a single MLA and a score value of 231 was achieved. This illustrates the superiority of the developed prognostics approach having a diverse strategy in developing the RUL predicting model.