A Bayes CNN-LSTM approach for aero-engine remaining life predictions

Remaining life prediction (RUL) plays an important role in prognostic health management. In recent years, deep learning-based methods have been widely used for RUL prediction. However, deep learning-based methods all focus on point prediction of RUL, ignoring the uncertainty associated with inaccuracies in the data and models. Given the prominence of Bayesian methods in uncertainty modelling, this paper uses a Monte Carlo dropout (MC-Dropout)-based Bayesian approximation to capture the epistemic and contingent uncertainty in the prediction process. A Bayes CNN-LSTM-based spatial-temporal feature extraction model is proposed to implement RUL prediction and quantify uncertainty. The proposed method is also validated on the CMAPSS dataset and compared with CNN, LSTM and CNN-LSTM to verify the effectiveness of Bayesian neural networks.