- 期刊
數位雙生技術應用於切削加工系統之鑑別與預測
Application of Digital Twin Techniques on the Identification and Prediction of Cutting Processing Systems
摘要
隨著智慧化機械的不斷發展，對於加工效率優化與自動化機械有著高度關注，且朝向數位雙生加工的趨勢發展。在本文中以西門子RMVM（Run My Virtual Machine）加工軟體模擬東台CT-350五軸工具機台作為對象的實際操作，透過RMVM提供切削參數包括位置、速度、加速度以及負載等，並使用GCODE進行數據分類，藉由非線性Hammerstein-Wiener模型、小波網路以及高斯過程等技術系統辨識建構切削力模型，可針對各軸向進給系統辨識，預測模擬輸出與實際結果的辨識度，結果說明實機訊號以小波網路模型辨識時，在大多數情境下均表現出較高的辨識度，可達96.7%；若以RMVM訊號分析時，相同模型下的辨識度也可達到辨識效果90%以上，GCODE分類辨識度皆能在95%以上，透過分段項數參數調整後辨識效果可達99%以上，將電流與能耗的輸出進行分析與計算，從而即時調整加工參數以能耗優化，最終可達到能在不接觸實際機台的情況下就能預測能耗。
並列摘要
As the development of intelligent machinery continues, there is a growing focus on optimizing machining efficiency and automation, with a clear trend toward digital twin machining. This article presents a case study using Siemens RMVM (Run My Virtual Machine) software to simulate the Tongtai CT-350 five-axis machine tool. By leveraging RMVM, cutting parameters such as position, velocity, acceleration, and load are provided. GCODE is used for data classification, and cutting force models are built using system identification techniques such as the nonlinear Hammerstein-Wiener model, wavelet networks, and Gaussian processes. These models enable the identification of feed systems along different axes, predicting the recognition accuracy between simulated outputs and real-world results. The findings demonstrate that when identifying signals from the actual machine using the wavelet network model, the recognition accuracy reached 96.7 % in most scenarios. When analyzing RMVM signals under the same model, recognition accuracy exceeded 90 %, and GCODE classification accuracy consistently surpassed 95 %. After adjusting segmented parameters, accuracy improved to over 99 %. The article further analyzed and calculated the outputs of current and energy consumption, enabling real-time adjustment of machining parameters for energy optimization. Ultimately, the model allows for energy consumption prediction without the need to interact with the actual machine.