基於Omniverse優化CNC加工後的纏屑處理

隨著產線自動化的推進，CNC (Computer numerical control) 加工過程中的纏屑問題易導致刀具損害、生產停機與維護成本上升。為解決此問題，本研究基於 NVIDIA Omniverse 平台建立數位雙生環境，結合強化學習提升 YOLO 模型之辨識效率與準確度，並透過高精度辨識結果即時控制機械手臂執行多階段清除機制，包括噴氣與勾取操作，必要時進行人工輔助。實驗結果顯示，本系統在不同光線與作業條件下辨識準確率超過 90%，有效降低停機頻率、縮短清理時間，提升刀具壽命與加工品質。本研究透過 Omniverse 虛擬訓練與模擬環境，成功實現智慧化纏屑處理，為產線自動化與先進製造提供關鍵技術參考。

With the rapid advancement of production line automation, robotic arms have been widely adopted for loading and unloading tasks. However, swarf generated during CNC machining can entangle cutting tools or workpieces if not promptly removed, leading to production downtime, tool damage, and increased maintenance costs. To address this issue, this study establishes a virtual production line environment using the NVIDIA Omniverse platform and integrates reinforcement learning to enhance the training efficiency and accuracy of the YOLO algorithm. The system employs high-precision swarf detection to activate a robotic arm for a multi-stage removal process, including a custom-designed air-jet mechanism and hook-based extraction, with manual intervention as a last resort. Experimental results show that the system achieves over 90% detection accuracy under varying lighting and operational conditions, significantly reducing downtime and cleaning time while extending tool lifespan and improving machining quality. By leveraging the digital twin environment and reinforcement learning capabilities of Omniverse, this study demonstrates an intelligent, fully automated swarf removal approach, contributing to the advancement of production line automation and smart manufacturing.