考慮主軸與切削刀具熱誤差之 CNC 銑削加工即時靜態誤差補償研究

Abstract

在CNC加工過程中，熱誤差約佔整體機台誤差的40%至70%，若能有效補償，將能大幅提升加工精度。然而，現有熱誤差補償研究多將重點聚焦於主軸，主因為可藉由空轉輕易取得主軸溫度與熱變形量，卻忽略了實際加工過程中可能產生的多種變因，以及刀具與其他機構對熱誤差的影響。 本研究提出一種同時考量主軸與刀具熱誤差的補償方法，透過熱電偶刀具無線溫度感測模組解決加工過程中刀具溫度難以監測的問題，並結合熱敏電阻主軸感測系統，以實際面銑加工所獲得之主軸與刀具溫升，以及刀具中心點（TCP）之熱位移作為特徵，建立熱誤差預測模型。 本研究共比較七種常見機器學習模型，包括Linear Regression、Support Vector Regression、Random Forest、Extra Trees、Gradient Boosting、XGBoost及LightGBM，結果以 Linear Regression 表現最佳，預測誤差僅為 3.024 μm。進一步透過內嵌式系統，將預測值以原點飄移方式回授至工具機控制器，實現加工過程中的即時熱誤差補償。驗證結果顯示，補償可將工件誤差由35.38 μm降低至15.57 μm，熱誤差減少達 56%；若排除刀具磨耗影響，誤差更可降至3.23 μm，補償後熱誤差可減少91%，顯示本研究具備高度實用性與應用潛力。

In CNC machining, thermal error accounts for approximately 40% to 70% of the total machine error. Effective compensation is therefore essential for significantly enhancing machining accuracy. However, most existing studies focus mainly on the spindle, as its temperature and thermal deformation can be easily obtained through idle rotation. This overlooks various influential factors during actual cutting, including the contributions of the cutting tool and other mechanical components. This study proposes a compensation method that considers both spindle and tool thermal errors. A wireless thermocouple-based tool temperature sensing module was developed to overcome the challenge of monitoring tool temperature during cutting, and was integrated with a spindle temperature sensing system using thermistors. Features such as the temperature rise of the spindle and tool, as well as the thermal displacement at the tool center point (TCP), were collected from actual face milling operations to construct a thermal error prediction model. Seven commonly used machine learning models were compared: Linear Regression, Support Vector Regression, Random Forest, Extra Trees, Gradient Boosting, XGBoost, and LightGBM. Among them, Linear Regression achieved the best performance, with a prediction error of only 3.024 μm. An embedded system was then used to feed the predicted value back to the CNC controller via reference point drift, enabling real-time thermal error compensation during machining. Experimental validation showed that thermal error was reduced from 35.38 μm to 15.57 μm, a 56% reduction. When tool wear effects were excluded, the error was further reduced to 3.23 μm, corresponding to a 91% reduction in thermal error, demonstrating high practicality and application potential.