在相同加工條件下，全自動虛擬量測（Automatic Virtual Metrology, AVM）具線上且即時之品質全檢特性。然當面對大量客製化需求時，相同加工條件的樣本難以累積，不易提供可用的加工品質估測模型，從而構成應用AVM的挑戰。本論文目標機台以大型臥式搪銑床為例，其為在三軸銑床上增加兩軸旋削面盤，專門加工大型加工件如齒輪箱、閥體、大型管接頭與發電機等，多為客製化的生產型態。
本論文提出通用型自動建模機制(Generic Automated Modeling Scheme, GAMS)，根據機台加工行為從收集不同加工條件的樣本所建立的樣本池中自動建置模型。首先，從樣本池中建立樣本條件樹(Sample Condition Tree, SCT)，為依類別與數值條件和其優先序所先行構成。其次，當欲建立新AVM模型時，則依SCT與建議樣本數從樣本池中擷取可用樣本後，進行自動建模用以估測加工品質。最後，若存在相同加工條件，僅有目標值差異或機台差異，則利用目標差值建模方法(Target Deviation Modeling, TDM)以調整估測模型。如此利用結合不同加工條件之樣本來一起建模的方式，預期將可解決如大型臥式搪銑床欲建構AVM模型時所需相同加工條件樣本數不足的問題。
實驗於搪銑機台的推拔錐度估測結果顯示，當GAMS應用於相同加工條件的樣本數不足時，可根據SCT於相近條件樣本自動補充以提供建模。在公差範圍為±0.1°條件下，在運用2到3筆樣本來更新模型後，模型估測精度即可趨於穩定，其錐度平均絕對誤差與最大誤差分別為0.019°與0.052°左右。而對於在相同機台不同目標值或不同機台相同目標值的條件下，利用現有樣本，依公差範圍在±1 mm之外徑尺寸實驗結果顯示，平均絕對誤差為0.035 mm左右。由此可知，通用型自動建模機制將可緩解樣本不足的困境。

Automatic Virtual Metrology (AVM) has the advantages of online and real-time total quality inspection under the same processing conditions. However, facing the demands of mass customization, it is not easy to provide a feasible predictive model of product quality since collecting modeling samples under the same processing conditions could be difficult, which becomes the challenge of applying AVM. This study takes large boring and milling machines as the target devices equipped with a two-axis rotary facing head based on a three-axis milling machine. These target devices specialize in machining large-sized products, such as gear boxes, valve bodies, large pipe fittings, and turbine housings, whose production types are often needed to be customized.
To solve the above-mentioned problem, a Generic Automated Modeling Scheme (GAMS) is proposed in this study, which can automatically build the AVM models by acquiring modeling samples from the sample pool, which contains many samples of various categorical and numerical conditions as well as their priorities. First, a Sample Condition Tree (SCT) is built from the sample pool. Next, when a new AVM model is to be built, the available modeling samples are gathered from the sample pool according to the SCT, and a model is then automatically built. Finally, if the modeling samples only show difference in either the target value or the machine type under the same processing conditions, the Target Deviation Modeling (TDM) method is then utilized to adjust the estimation model. Therefore, modeling with different conditional samples can be expected to meet the needs of mass customization.
The estimations of taper angle experimented on boring and milling machines show that when modeling samples under the same processing conditions is insufficient for building models, the GAMS can then be applied to automatically supplement samples under similar conditions to complete modeling according to the SCT. With the tolerance range between -0.1 and 0.1 degrees, the model prediction accuracy can be stabilized after refreshing 2 to 3 modeling samples, while the mean absolute error (MAE) and maximum error of taper angle are about 0.019 degrees and 0.052 degrees, respectively. For different target values against the same machine or the same target value against different machines, in the case of using the existing samples, the experimental results of outer diameter show that the MAE is about 0.035 mm that is within the tolerance range of ±1 mm. All in all, GAMS can alleviate the issue of insufficient modeling samples under the same processing conditions.

[1] M.-H. Hung, T.-H. Lin, F.-T. Cheng, R.-C. Lin, “A novel virtual metrology scheme for predicting CVD thickness in semiconductor manufacturing”, IEEE/ASME Transactions on mechatronics, vol. 12, no. 3, pp. 308-316, June 2007.
[2] F.-T. Cheng, C.-A. Kao, C.-F. Chen; W.-H. Tsai, “Tutorial on applying the VM technology for TFT-LCD manufacturing”, IEEE Transactions on Semiconductor Manufacturing, vol. 28, no. 1, pp. 55-69, 2015.
[3] H. Tieng, T. Tsai, C. Chen, H. Yang, J. Huang and F. Cheng, “Automatic Virtual Metrology and Deformation Fusion Scheme for Engine-Test Manufacturing,” IEEE Robotics and Automation Letters, vol. 3, no. 2, pp. 934-941, April 2018.
[4] H. Tieng, C.-F. Chen, F.-T. Cheng and H.-C. Yang, “Automatic Virtual Metrology and Target Value Adjustment for Mass Customization,” IEEE Robotics and Automation Letters, DOI: 10.1109/LRA.2016.2645507, vol. 2, no. 2, pp. 546-553, April 2017.
[5] 王國維。「適用於工具機產業之全自動虛擬量測系統自動建模機制」。碩士論文，國立成功大學製造資訊與系統研究所，2016。
[6] 江律嫺。「AVM自動建模−以五軸工具機為例」。碩士論文，國立成功大學製造資訊與系統研究所，2018。
[7] Xuewei Zhang, Kornel F. Ehmann, Tianbiao Yu, Wanshan Wang, “Cutting forces in micro-end-milling processes,” International Journal of Machine Tools and Manufacture, vol. 107, pp. 21-40, 2016.
[8] Ghan R H, Hashmi A A, Dhobe M M, “A Review on Optimization of Machining Parameters for Different Materials,” International Journal of Advance Research, vol. 3, Issue 2, 2017.
[9] Xinzheng Xu, Tianming Liang, Jiong Zhu, Dong Zheng, Tongfeng Sun, “Review of classical dimensionality reduction and sample selection methods for large-scale data processing,” Neurocomputing, vol. 328, pp. 5-15, 2019.
[10] Xianli Pan, Yitian Xu, “Two effective sample selection methods for support vector machine,” Journal of Intelligent & Fuzzy Systems, vol. 30, no. 2, pp. 659-670, 2016
[11] F.-T. Cheng, H.-C. Huang, and C.-A. Kao, “Developing an Automatic Virtual Metrology System,” IEEE Transactions on Automation Science and Engineering, vol. 9, no. 1, pp. 181-188., January 2012.
[12] Y.-T. Huang and F.-T. Cheng, “Automatic Data Quality Evaluation for the AVM System,” IEEE Transactions on Semiconductor Manufacturing, vol. 24, no. 3, pp.445-454, August 2011.
[13] F.-T. Cheng, Y.-T. Chen, Y.-C. Su, and D.-L. Zeng, “Evaluating Reliance Level of a Virtual Metrology System,” IEEE Transactions on Semiconductor Manufacturing, vol. 21, no. 1, pp. 92-103, February 2008.
[14] F.-T. Cheng, H.-C. Huang, and C.-A. Kao, “Dual-Phase Virtual Metrology Scheme,” IEEE Transactions on Semiconductor Manufacturing, vol. 20, no. 4, pp. 566-571, November 2007.