Similarity-based anomaly detection method for turning of multi-material workpieces with varying axially constant blank diameter

Geometry and hardness fluctuations of formed blanks are challenging for process monitoring of the subsequent machining process because they lead to deviating process forces during roughing. Ordinary anomaly detection methods require the forces to be similar. In this work, a similarity-based anomaly detection method is proposed that utilizes Dynamic Time Warping to achieve robustness against blank fluctuations during roughing. It extracts the average signal shape from training signals, scales it individually for each novel process, and uses confidence limits for anomaly detection. The method is tested with multi-material shafts whose blank diameter is axially constant but varies between workpieces.