A Novel End-to-End Ensemble UNet Network Based on Edge Computing for Tool Wear Detection

Automated tool wear condition monitoring is essential for ensuring product quality. However, achieving both high detection precision and utilizing edge computing are critical for advancing intelligent manufacturing. Consequently, this article develops a tool wear detection system utilizing edge computing techniques. Firstly, a novel end-to-end ensemble UNet-based model is proposed. Through end-to-end feature fusion training, it effectively addresses the robustness and accuracy issues inherent in single models. Secondly, multiple neural computing sticks are involved to reduce the need for cloud transmission and preserve data privacy, thereby enabling scalability and flexibility in deployment. Experiments demonstrate that mean intersection over union (mIoU) and mean dice coefficient (mDice) in tool wear detection can be improved by over 3% and 2%, respectively. Furthermore, the execution time can be accelerated by a factor of 4.