This document discusses a data-driven framework for condition monitoring at rolling mills using a residual-based approach for fault detection. It outlines the methodology, including data cleaning, system identification, and model training, and compares its performance with traditional methods like principal component analysis (PCA) and multi-scale PCA (MSPCA). The findings indicate that while PCA can be more effective, the proposed residual-based method shows promise, especially when accounting for non-linearities.