Security-enhanced cutting tool wear segmentation with federated learning

Metal cutting tool wear evaluation is a time- and cost-consuming task. Direct measurements (manual microscopic imaging) generate sensitive microscopic tool wear images (MTWI) which, on one hand, can inaugurate this measurement’s automation and, on the other hand, hold confidential information about the cutting process. This article contains an approach for collaboratively automating tool condition monitoring (TCM) while individually preserving the privacy of clients’ images. The method developed consists of a cohort of tool wear image owners and a Federated Learning (FL) pipeline for processing synergistically the owners’ images without sharing proprietary images. Both cohort data setting and training scenario are investigated, evaluated and discussed.