Time-Series Foundation Model Embeddings for Remaining Useful Life Estimation

The article presents a novel approach for Remaining Useful Life (RUL) estimation using a frozen pretrained time-series foundation model (TSFM), specifically Chronos-2, combined with a lightweight regression head. Experiments demonstrate that features extracted from Chronos-2 outperform traditional models, including recurrent, convolutional, and Transformer-based architectures, as well as gradient-boosting methods, indicating a significant improvement in predictive performance with longer context lengths. This approach offers a data-efficient solution for RUL estimation, reducing the reliance on extensive feature engineering and large labeled datasets, which is crucial for practitioners in industrial predictive maintenance.