Temporal Conditional Normalizing Flows for Data Augmentation in Remaining Useful Life Prediction under Data Scarcity

Guillaume Prevost, Esteban Cabanillas, Jérôme Boutet

Keywords: Normalizing flows, data augmentation, remaining useful life

TL;DR: Data augmentation strategy for bearing degradation analysis using a generative normalizing flow model

Abstract: Predicting the remaining useful life (RUL) of bearings suffers from a lack of industrial data, limiting the generalization of supervised models. We propose a temporal conditional normalizing flow model (TCNF), a generative model conditioned on RUL and a temporal context encoded by GRU, capable of generating complete and consistent degradation trajectories. An affine normalization layer is integrated directly into the flow, ensuring bijectivity and exact log-determinant computation. On the XJTU-SY dataset, TCNF improves the RMSE compared to the baseline without augmentation and outperforms the comparative GAN and VAE approaches.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 5