Recurrent variational autoencoder approach for remaining useful life estimation.

The article focuses on a novel method for estimating the Remaining Useful Life (RUL) of aircraft engines using a recurrent variational autoencoder (VAE) framework that enhances interpretability in prognostics and health management (PHM). Unlike traditional black-box deep learning models, this approach integrates a regressor into the training process to regularize the latent space, enabling a two-dimensional, self-explanatory map that visually clusters engine degradation states and provides accurate numerical RUL predictions. Evaluated on NASA’s C-MAPSS simulation dataset and real-world Turbofan engine data, the model demonstrates competitive or superior accuracy compared to state-of-the-art methods while offering explainable diagnostics through latent space visualization. This framework supports continuous online monitoring and has potential for deployment on hardware with limited computational resources due to its lightweight and recurrent design.