- 學位論文
進化式模糊控制器應用於CNC工具機之熱變形預測
Thermal Deformation Prediction of CNC Machine Tool Using Evolutionary Fuzzy Controller
摘要
近年來的CNC工具機產業邁向高速高精度加工發展，各種精密工件需求量增加，產值也蒸蒸日上，而熱誤差一直是影響精密度的主要因素，隨著目前加工的趨勢，其影響不容忽視，是否能因應這無可避免的物理現象，將是未來工具機技術的重要突破。 由於工具機內部零件複雜，溫度彼此影響，只單純討論少數溫度對位移的影響，已不敷使用，以建立完整的熱預測模型為目標，本研究對CNC立式綜合加工機發展一套資料擷取系統，並在空切削間歇式運轉和持續運轉實驗中，收集溫度和位移的資料，再使用提出的進化式模糊控制器(EFC)來建立熱變形預測模型，透過不同情境的實驗驗證模型預測準確度，預測出的位移值將提供未來補償時所需。 進化式模糊控制器(EFC)與多變數迴歸分析(MRA)在各種不同工作情境下的實驗結果顯示，在建模實驗相同的情況下，表現並無多大差異，但應用在其他實驗時，進化式模糊控制器(EFC)的表現比多變數迴歸分析(MRA)更能適應不同實驗狀況，表現較為良好，可用度較高，平均預估誤差維持在3um以下。
並列摘要
In recent years, the development of CNC machine tool industry towards high-speed and high-precision processing, increased precision machining requirements for workpieces, and the output value is also flourishing. Thermal error has been the main factor affecting the precision; its impact cannot ignored. The breakthrough in machine tool technology is whether it can cope with this unavoidable physical phenomenon. The internal parts of the machine are complicated; the temperature affects each other, we only discuss the impact of a small number of temperatures on the displacement has been inadequate. To establish a complete thermal prediction model as the goal, this study develops a data acquisition system for CNC vertical integrated processing machines. Temperature data and displacement data collection by cutting intermittent operation and continuous operation of the experiment, and then use the proposed evolutionary fuzzy controller (EFC) to establish the thermal deformation prediction model. Through experimental verification prediction accuracy of model in different cases, the predicted displacement value will be required for future compensation. The experimental results of evolutionary fuzzy controller (EFC) and multivariate regression analysis (MRA) in different working situations show that there is no significant difference in the performance of the modeling experiment. However, in other experiments, the evolutionary fuzzy controller (EFC) is more adaptable to different experimental conditions than the multivariate regression analysis (MRA), and the performance is higher, the average estimation error is below 3um.