In our latest preprint, we introduce the final hybrid framework developed to address clogging in nuclear steam generators. This methodology integrates several components, culminating in a Bayesian fusion mechanism that delivers robust probabilistic estimates of the remaining useful life (RUL) for a given asset.

This approach is particularly well suited to settings where data are both scarce over time and heterogeneous in nature—conditions commonly encountered in industrial applications. While the framework relies on a complex simulation code that does not explicitly account for model misspecification, it remains highly practical from an engineering standpoint. In particular, it is scalable across the entire fleet of EDF steam generators and can be integrated into an online digital twin.

Further details can be found in my thesis manuscript. The paper has been submitted to Nuclear Science and Engineering and is available on arXiv and HAL.