Turbofan Engine Remaining Useful Life (RUL) Prediction Based on Bi-Directional Long Short-Term Memory (BLSTM)

Abstract

The aviation industry is rapidly evolving, driven by advancements in technology. Turbofan engines used in commercial aerospace are very complex systems. The majority of turbofan engine components are susceptible to degradation over the life of their operation. Turbofan engine degradation has an impact to engine performance, operability, and reliability. Predicting accurate remaining useful life (RUL) of a commercial turbofan engine based on a variety of complex sensor data is of paramount importance for the safety of the passengers, safety of flight, and for cost effective operations. That is why it is essential for turbofan engines to be monitored, controlled, and maintained. RUL predictions can either come from model-based or data-based approaches. The model-based approach can be very expensive due to the complexity of the mathematical models and the deep expertise that is required in the domain of physical systems. The data-based approach is more frequently used nowadays thanks to the high computational complexity of computers, the advancements in Machine Learning (ML) models, and advancements in sensors. This paper is going to be focused on Bi-Directional Long Short-Term Memory (BLSTM) models but will also provide a benchmark of several RUL prediction databased models. The proposed RUL prediction models are going to be evaluated based on engine failure prediction benchmark dataset Commercial Modular Aero-Propulsion System Simulation (CMAPSS). The CMAPSS dataset is from NASA which contains turbofan engine run to failure events.