An improved dynamical variational autoencoder framework for predicting aero-engine remaining useful life

Accurate remaining useful life (RUL) prediction is vital for ensuring equipment safety, optimizing maintenance schedules, and reducing operational costs. Deep learning networks have emerged as effective means for RUL prediction. However, these methods encounter challenges of multiple sensors and data noise when handling large and complex industrial equipment, which impede the prediction model’s ability to generalize. To address these issues, this paper proposes an improved Dynamical Variational Autoencoders (DVAE) Framework combining principal component analysis (PCA) and long short-term memory (LSTM) networks. The effectiveness of our DVAE RUL prediction framework is validated through experiments conducted on the NASA C-MAPSS dataset. Results show that the approach significantly improves the prediction accuracy.