運用機器學習演算法開發熱變形分析預測系統

工具機的發展在國內外已邁向高精度加工，隨著主軸轉速不斷提高與相關操作程序日趨複雜，溫升熱效應已經是目前加工誤差的主要原因之一。在機器長時間的運作並配合不同的操作參數，機台之間的溫升熱效應也會有所差異。因此，本研究希望能發展出一套穩健的機台加工之自動修正平台。首先，將線性迴歸、類神經網路、支援向量迴歸等來建構平台，再彙整操作參數與感測器資料於所開發之「機台數據倉儲」，以建立分析資料庫。加入變數縮減的功能，希望能列出變數的權重值來讓使用者可以進行變數的選擇，以減少溫度感測器的數量。透過反覆的比較分析，從各預測模型中依據製程記錄找出自動替各機台發展專屬的分析模型與參數。期許能透過這個機制加強模型精度之穩定性。

關鍵字

熱變形；誤差補償；資料倉儲；資料探勘

並列摘要

The development of machine tools has been towards high-precision processing. Thermal expansion error has become the main factor of the machining accuracy with the increasing motor speed and complicated steps. This effects would be various among these machine tools after working for a long period. Therefore, this study aims to develop a stable platform that can be capable of tuning each individual machine automatically. First, we try to develop the platform by adopting several conventional algorithms, which includes linear regression, neural network, support vector machine and support vector regression. Then, we build up a machine data warehouse by collecting operational parameters and data from all sensors. In addition, the proposed system aims to reduce the number of input variables. Corresponding weights can be generated for user reference. Consequently, this platform could automatically select the proper prediction model for an individual machine by evaluating the accuracy among all models. We would try to improve the efficiency by using feature selection technique and enhance the stability by choosing a proper model constantly.

並列關鍵字

thermal deformation ； error compensation ； data warehouse ； data mining

國際替代計量

運用機器學習演算法開發熱變形分析預測系統

全文下載

主題瀏覽