Remaining useful life estimation using a recurrent variational autoencoder


A new framework for the assessment of Engine Health Monitoring (EHM) data in aircraft is proposed. Traditionally, prognostics and health management systems rely on prior knowledge of the degradation of certain components along with professional expert opinion to predict the Remaining Useful Life (RUL). In order to avoid reliance on this process while still providing an accurate diagnosis, a data-driven approach using a novel recurrent version of a VAE is introduced. The latent space learned by this model, trained with the historical data recorded by the sensors embedded in these engines, is used to visually evaluate the deterioration progress of the engines. High prognostic accuracy in estimating the RUL is achieved by building a simple classifier on top of the learned features of the VAE. The superiority of the proposed method is compared with other popular and state-of-the-art approaches using Rolls Royce Turbofan engine data. The results of this study suggest that the proposed data-driven prognostic and explainable framework offers a new and promising approach.

In International Conference on Hybrid Artificial Intelligence Systems
Nahuel Costa
Nahuel Costa
Machine Learning & AI researcher