本論文針對工具機進給系統中的雙螺帽滾珠螺桿提出定位誤差的分析與預測方法，藉由財團法人工業技術研究院所研發的複合式預壓/溫度感測模組(Preload/Temperature sensor, P/T sensor)，能即時量測預壓與溫度之數值，可得其值變化對於進給系統定位誤差之影響。 研究目標為主要利用P/T sensor之量測數據，應用倒傳遞類神經網路(Back-Propagation Neural Network, BPNN)建構出定位誤差預測模型(Positioning Error Prediction Model, PEPM)，用以替代高成本之光學尺，預判出系統當前的位置誤差量以維持定位精度。
為了達成研究目標，在實驗方面透過更換滾珠尺寸大小與不同溫度條件下，以仿效滾珠螺桿的長期運作後螺帽磨耗的狀況，並運用BPNN的特性，可有效近似輸入、輸出參數之間非線性關係，建立PEPM。 其中螺帽預壓、溫度與位置作為BPNN的輸入; 輸出則是定位誤差之預測值。 訓練模型過程中是以均方誤差(Mean Square Error)為評估網路參數之性能的依據，藉此挑選出最適用於本研究之模型參數。 經由模擬結果得知，在不同預壓與溫度變化下，預測值皆能符合實際值之趨勢且預測誤差維持在5~10μm以內，故能驗證本論文運用的BPNN之有效性。

In this thesis, the analysis and prediction method of positioning error for double-nut ball screw feed drive systems are proposed. This methodology is based on using the composite preload/temperature sensor developed by Industry Technology Research Institute (ITRI) to measure the preload and temperature, which can be regarded as major factors affecting the positioning error. The research objective is mainly aimed to predict the positioning error of the ball screw systems. To construct Positioning Error Prediction Model (PEPM), measured data from sensors are provided to Back-Propagation Neural Network (BPNN). PEPM is expected to substitute for the costly optical scale. In order to analyze the variation of preload and temperature that affect the positioning accuracy of ball screw systems, the thesis builds up several different experimental conditions to collect the corresponding data. Firstly, different sizes of steel balls are utilized to mimic the worn conditions of the nuts and preload loss. Secondly, the ball screw system is operated under the adjustable temperature environment by a heater unit. With the experimental data, BPNN is capable to fit the nonlinear relationship between the input and output parameters. The preload, temperature, and encoder position by the servo-motor are selected to the input parameters of BPNN. The output parameter of BPNN is the predicted value of positioning error. To evaluate the model performance at the training process, Mean Square Error (MSE) is chosen as the performance index. According to simulation results, predicted values is similar to the trend of actual values, and prediction accuracy is retained within 5~10 micrometers under a certain range of the preload and temperature. Therefore, the results show that BPNN can be usefully applied to ball screw systems.

[1] 上銀科技股份有限公司, “產品手冊”, https://www.hiwin.tw/
[2] Bryan, J., “International Status of Thermal Error Research”, Annals of CIRP, Volume 39, pp. 645-656, 1990.
[3] S. K. Kim and D. W. Cho, “Real time estimation of temperature distribution in a ball screw system”, International Journal of Machine Tools and Manufacture, Volume 37, pp.451-464, 1997.
[4] Kodera T. et al., Real-Time Estimation of Ball-Screw Thermal Elongation Based upon Temperature Distribution of Ball-Screw, JSME International Journal, series C, Volume 47, pp. 1175-1181, 2004.
[5] U. Heisel, G. Koscsák, T. Stehle, “Thermography-Based Investigation into Thermally Induced Positioning Errors of Feed Drives By Example of a Ball Screw, CIRP Annals, Volume 55, pp. 423-426, 2006.
[6] Shyh-Chour Huang, "Analysis of a model to forecast thermal deformation of ball screw feed drive systems", International Journal of Machine Tools and Manufacture, Volume 35, pp. 1099-1104, 1995.
[7] Oyanguren, A., Larrañaga, J. & Ulacia, I. Thermo-mechanical modelling of ball screw preload force variation in different working conditions. Int J Adv Manuf Technol 97, Volume 97, pp. 723–739, 2018.
[8] P.C. Tsai, C.C. Cheng, Y.C. Hwang, “Ball screw preload loss detection using ball pass frequency.” Mechanical Systems and Signal Processing, Volume 48, pp. 77-91, 2014.
[9] Matsubara, A., Kakino, Y., & Watanabe, Y. , “Servo Performance Enhancement of High Speed Feed Drives By Damping Control, ” Proceedings of the Japan USA Symposium on Flexible Automation, Volume 1, pp. 23-26, 2000.
[10] 邱奕範, “命令及摩擦力前饋控制於工具機之研究”, 國立成功大學機械工程學系, 2002.
[11] 葉怡成, “類神經網路模式應用與實作”, 儒林圖書公司, 2009.
[12] 張斐章,張麗秋, “類神經網路導論原理與應用”, 滄海書局, 2010.