以機器學習為主之環境適應性CNC工具機機上溫升熱變形補償方法之研究

Study of on-machine fusion error compensation-based machine learning for Termal induced errors with environment temperature adaptability for CNC machine tools

黃彬晏(Bin-Yan Huang)

指導教授 ： 陳夏宗 ； 王世明

中原大學/工學院/機械工程研究所/碩士(2020年)

https://doi.org/10.6840/cycu202000950

摘要

本研究主要為建立環境適應性溫升熱變形模組，以往預測方式為在不同伺服軸劃分不同的量測點位，並建立數多個迴歸方程式來進行曲線擬合，不同的環境溫度即需要不同的環境溫度實驗模組，將會需要收集許多數據來達成預測，本實驗透過Python程式語言中的函式庫Tensorflow，來搭建類神經網路架構使能進行學習，透過少量的數據訓練，訓練出一個模組可以預測該軸各量測點位之變形量，在不同的環境溫度，訓練資料外都能有良好的預測。補償方法主軸補償利用CNC控制器上之機械座標原點偏移功能進行線上補償，而伺服軸補償則是輸出補償指令以螺距誤差(Pitcherror)補償方式進行。並建立一個友善的人機介面，介面上能夠了解目前各軸的預測誤差量，並使使用者能夠自行填寫希望加工件之精度，並進行自動補償功能，也將會提醒使用者如果當下需要進行補償的話，該進行的操作步驟來進行補償。本系統在不同的環境溫度影響下，在考慮機台升溫與環境溫度的情況下，可改善80%以上之熱變形誤差，達到降低工具機熱變形誤差之目的。

關鍵字

CNC工具機；人工神經網路；機器學習；誤差補償；熱變形

並列摘要

This research is mainly to establish an environmentally adaptive temperature rise thermal deformation module. In the past, the prediction method was to divide different measurement points on different servo axes and establish multiple regression equations for curve fitting. Different ambient temperature experiment modules, Many data will need to be collected to achieve prediction. In this experiment, a function library in the Python programming language Tensorflow is used to build a neural network-like architecture to enable learning. With a small amount of data training, a module can be trained to predict the axis. The deformation amount of each measurement point can have a good prediction outside the training data at different ambient temperatures. Compensation method The main axis compensation uses the mechanical coordinate origin offset function on the CNC controller to perform online compensation, while the servo axis compensation is performed by the output compensation command in the form of pitch error compensation. And establish a friendly human-machine interface, the interface can understand the current forecast error of each axis, and enable the user to fill in the desired precision of the workpiece, and perform automatic compensation function, will also remind users if they need to In the case of compensation, the operation steps should be performed to compensate. Under the influence of different environmental temperatures, the system can improve the thermal deformation error of more than 80% under the conditions of machine temperature and environmental temperature, and reduce the thermal deformation error of the machine tool.

並列關鍵字

CNC machine tools； artificial neural networks； machine learning； error compensation； thermal deformation

參考文獻

Moshe Barash,"Thermal Effects on the Accuracy of NumericallyControlled Machine Tools", Annals of the CIRP,Vol.35/1, 2000, pp. 255-258.

Optiz,H.and R.Noppen,"A Finite Element Program System and its Application forMachine Tool Structural Analysis "MTDR,Vol.13,2001,pp.55-60.

Bryan,J.,"International Status of Thermal Error Research", Annals of CIRP，Vol.392, 2002

J.S.Chen,J.X.Yuan,J.Ni,and S.M.Wu,“Real-Time Compensation for Time-Variant Volumetric Errors On a Machining Center”, Transactions of ASME,Journal of Engineering for Industry,Vol.115, pp. 472-479, 2003.

J.S.Chen,“Fast Calibration and Modeling of Thermally Induced Machine Tool Errors in Real Machining”,Int.J.Mach.Tools Manufact.Vol.37, No 2, pp.159-169, 2001