Prepared for Lift-Off: Hybrid CNN-LSTM Architecture for Aircraft Engine Remaining Useful Life Estimation

Equipment failure is the leading cause of production halts, which can be avoided through Predictive Maintenance indicators, such as the Remaining Useful Life (RUL). This task is particularly noteworthy in domains like aircraft maintenance, where undetected faults can result in loss of human life. To address this, we propose a solution that combines preprocessing techniques with a hybrid architecture consisting of a Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM) network. We rely on a piece-wise linear regression function to pinpoint the start of degradation for modern turbofan engines. Using this method alongside classical preprocessing strategies and the versatile CNN-LSTM hybrid exhibits state-of-the-art predictive capabilities on the chosen dataset. Performance evaluations, in terms of RMSE, PHME, MAE and R-squared are computed across all subsets of the NASA C-MAPSS dataset. Our methodology showcases an RMSE improvement of at least 50% compared to the next-best approach in the current literature. The full implementation code is available at https://github.com/ariandra34/Prepared-for-Lift-off.