01挑戰The challenge
刀具崩刃或斷刀不只損失刀具本身,更可能毀損加工到一半的高價工件。依固定時間換刀又會浪費仍可使用的刀具。A chipped or broken tool costs more than the tool: it can ruin a high-value workpiece mid-cut. Changing tools on a fixed schedule, meanwhile, throws away usable tool life.
02我們的做法Our approach
- 以 PHM 方法,比較線性與非線性迴歸演算法,建立刀具剩餘可用壽命(RUL)預測模型Built tool RUL models with the PHM method, comparing linear and non-linear regression algorithms
- 將 online RUL 預測技術內建於工具機,即時評估是否需要換刀Embedded online RUL prediction in the machine tool for real-time tool-change decisions
- 進一步發展自我感知(self-awareness)演算法:依切削狀態判定目前的崩刃形式,並能辨識從未見過的新崩刃型態,據以重新訓練故障診斷模型Developed a self-awareness algorithm that classifies the current chipping mode from cutting conditions, recognizes previously unseen chipping patterns, and retrains the diagnosis model accordingly
03成果Results
- 降低斷刀風險與刀具成本Lower risk of tool breakage and lower tooling costs
- 模型能隨新的失效型態持續進化,不會因產線變化而失準The model keeps learning new failure modes instead of drifting as production changes
許驥、張淵仁*、謝聲偉,〈機器學習應用於切削刀具異常檢測模型研究〉,技術通報,Vol. 285, pp. 30–36,2022 年 1 月。
張淵仁*、陳昱翔、許驥、賴政佑、張平昇、謝聲偉,〈線性與非線性迴歸演算法於 PHM 方法預測刀具剩餘可用壽命之比較〉,機械新刊,第 47 期,pp. 34–44,2020 年 10 月。