Enhance Chatter Prediction for Increased Productivity

Milling process instability due to regenerative chatter vibrations is a complex phenomenon that has been causing serious damages and imposing productivity limits in the manufacturing industry. Developing a reliable solution that complies with the practical capacities of a shopfloor brings economic and environmental benefits. This paper introduces a physics-supported probabilistic machine learning approach, where chatter monitoring data from one or multiple machines are effectively exploited under a federated learning scheme, also enabling the attained knowledge to be transferred to other machines and processes. The resulting probabilistic prediction offers added reliability by revealing unresolved uncertainties due to limited observation through arbitrary cuts. The approach offers extended applicability for chatter prediction in non-rectangular workpiece-tool interactions, which often occur in industrial milling. The effectiveness is evaluated through experimental case studies and shows a promising solution to the decades-old industry’s struggle with the chatter issue.