End-to-End Methodology for Predictive Maintenance Based on Fingerprint Routines and Anomaly Detection for Machine Tool Rotary Components

This work introduces an end-to-end methodology, from data gathering to fault notification, for the predictive maintenance of rotary components of machine tools. This is done through fingerprint routines; that is, processes that are executed periodically under the same no-load conditions to obtain a snapshot of the machine condition. High-frequency vibration data gathered during these routines combined with knowledge about the machine structure and its components are used to obtain failure-specific features. These features are then introduced to an anomaly and paradigm shifts detection algorithm. The method is evaluated through three distinct scenarios. First, we use synthetically generated data to test its ability to detect controlled variations and edge cases. Second, we use with publicly available data obtained from bearing run-to-failure tests under normal load conditions on a specially designed test rig. Finally, the methodology is validated using real-world data collected from a spindle bearing installed in a machine tool. The novelty of this work lies in performing anomaly detection using failure-specific features derived from fingerprint routines, ensuring stability over time and enabling precise identification of machine conditions with minimal data requirements.