生成式視覺模組導引機器人智慧研磨技術

中大型鑄件如鐵路連結器之澆口去除、研磨與除鏽長期仰賴人工，品質受疲勞與經驗影響，潛藏 安全風險。本文整合 3D 視覺、AI 瑕疵辨識、自動換刀主軸與浮動研磨頭，提出適用於少量多樣鑄件之 智慧系統。系統採「感測 → 決策 → 執行 → 回授」架構，即便無 CAD 資訊亦能由點雲自動辨識待修區 並規劃路徑，加工後再經 AI 視覺檢出剩餘瑕疵並補修。實證顯示，1 m × 1 m 區域於 40 秒內完成掃描； 瑕疵辨識率達 91%；澆口由 5–20 mm 精修至 0.5 mm；導入後產能由每日 2 件提升至 3 件、人力減半，並 建立品質追溯機制。

The removal of gates, grinding, and derusting for medium-to-large castings, such as railway couplers, has long relied on manual visual inspection and handheld tools. However, quality is often compromised by operator fatigue, workpiece variations, and experience-based inconsistencies, posing risks to fatigue strength and safety. This article integrates 3D vision, AI defect recognition, automatic tool-changing spindles, and floating grinding heads to propose an intelligent system for low-volume, high-mix casting production. Utilizing a “Sensing- Decision-Execution-Feedback” architecture, the system automatically identifies repair areas and plans robotic paths from actual point clouds, even without reliable CAD data. After processing, AI vision detects remaining defects for precision re-repair. Empirical results show that a 1m x1m area can be scanned within 40 seconds, with a 91% defect recognition accuracy. Gate height can be reduced from 20 mm to 0.5 mm. Post implementation, daily productivity increased from 2 to 3 units while halving labor requirements. The establishment of quality traceability mechanisms further confirms the system’s feasibility for intelligent and internationalized post-casting applications.