TinyML Anomaly Detection in Portable Cutting Tools

Through electrical impedance measurement between the bar of the cutting tool and the operator’s body, an active impedance protection system can discriminate between the body proximity while cutting a tree. However, the performance of simple threshold-based systems can be affected by the unpredictability of the working environment in addition to the complexity of cutting tasks. This paper proposes and evaluates a protection system for portable cutting tools based on end-to-end anomaly detection using TinyML paradigm. The model is based on the fully connected neural network and Kmeans clustering algorithm, which is developed on the ESP32- DevKitC-V4. Considering anomaly rank while approaching normal or hazardous situations, the proposed system shows an improvement in detecting different states enabling real-time decision-making through a compact, low-cost, and low-power consumption solution.