Frozen Chronos-2 Backbone Beats Task-Specific Models for RUL Prediction

Using Chronos-2 as a frozen feature extractor with a lightweight regression head outperforms recurrent, convolutional, and transformer-based task-specific models on remaining useful life prediction from industrial sensor data across two device types. The approach eliminates the need for large labeled datasets or extensive feature engineering.