Uncertainty-Aware Remaining Useful Life Prediction for Predictive Maintenance


Remaining useful life (RUL) prediction is one of the most crucial tasks in modern prognostic health management (PHM) systems. The ability to reliably predict when a failure might occur allows complex engineered systems such as industrial assets to be maintained proactively and efficiently. In this context, deep learning (DL) emerges as a powerful data-driven method, which is capable of predicting the RUL of an asset given its historical operating data collected by multiple sensors. Like many other prediction tasks, uncertainty is inherent in the task of RUL prediction. However, standard DL tools typically do not take this kind of uncertainty into account. In this project, we aim to develop a DL-based framework that not only predicts the RUL but should also be capable of outputting the associated confidence interval capturing the uncertainty of the RUL prediction. Theses on this topic can be written in English or German.


Basic machine/deep learning knowledge, preferably some hands-on experience with deep learning, good programming skills in python

Ansprechpartner: Quy Le Xuan