Towards the digital twin of a cutting tool: Evaluation of an anomaly detection based drilling tool condition monitoring system

Uncertainty about the actual condition of cutting tools during drilling leads to conservative tool replacement strategies, resulting in significant unused tool life, production interruptions, and avoidable costs. Existing digital twin-based tool condition monitoring implementations rely predominantly on supervised learning methods that require labeled wear measurements for training. This paper presents a modular five-dimensional digital twin framework for drilling tools that addresses this challenge through anomaly detection: a convolutional autoencoder trained exclusively on motor current signals under normal operating conditions detects tool wear without labeled failure data, and a structured service layer translates reconstruction errors into actionable tool change recommendations through adaptive thresholding and persistence-based decision logic. Evaluated across two workpiece materials and varying cutting parameters without recalibration, the framework achieves 81.2% tool life utilization for gradual wear and detects sudden failure at 95.4% and 98.7% of the respective tool lives with zero false positives across all experiments. The approach demonstrates cross-material generalization from a model trained on a single material and parameter set.