本篇文章主要探討如何尋找合適的電子元件散熱片參數，以獲得更高的散熱效率。本文考慮的資料因子分為三種類型，共享因子 (Shared Factors)、分支因子 (Branching Factors)、巢狀因子 (Nested Factors)。共享因子可以是定性因子 (Qualitative Factors)或定量因子 (Quantitative Factors)。分支因子皆為定性因子，並且在每個分支因子水準中存在不同的巢狀因子，本文中的巢狀因子皆為定量因子。在本文中將散熱器參數調整問題轉化為一個最佳化問題，並採用 替代曲面協助法 尋找較佳的散熱器參數。替代曲面協助法可分為三個部分組成。第一個是初始點設計，這裡使用拉丁超立方設計 (Latin Hypercube Design)生成初始點。然後基於初始點建構替代 輔助模型， 本篇文章使用高斯過程模型 (Gaussian Process Model)作為替代輔助模型 。最後需要一個填充準則選取下一個 實驗點。本篇文章針對複雜的因子結構，提出了一個新的高斯過程模型。根據數值實驗的結果，提出的替代輔助模型有較好的表現結果。

This thesis focus on finding the proper parameters for electronic component heat dissipation fin to get higher cooling efficiency. The factors considered in this thesis are divided into three types, shared factors, branching factors and nested factors. Shared factors can be qualitative factors or quantitative factors. Branching factors are treated as qualitative factors and there are different nested factors in each levels of branching factors. Here nested factors are only quantitative factors. In this thesis, first we transfer this tuning problem as an optimization problem and a surrogate-assistant approach is adopted to find the solution.
The surrogate-assistant approach contains three key components. The first one is the initial design. Here a Latin hypercube design is used to generate the design points. Then the surrogate model is constructed based on the current explored points and usually the Gaussian process type model is used for surrogate construction. Finally we need to have an infill criterion for us to identify the next explored points. Due to complex factor structure, a new Gaussian process model is proposed. Based on our numerical results, the proposed surrogate-assistant approach seems performing well.

[1] Jerome Sacks, William J. Welch, Toby J. Mitchell, Henry P. Wynn (1989). “Design and Analysis of Computer Experiments.” Statistical Science, 4, 409-423.

[2] Alexander I. J. Forrester, Andras Sobester, Andy J. Keane (2008). “Engineering Design via Surrogate Modelling: APractical Guide” Wiley.

[3] Ying Hung, V. Roshan Joseph, Shreyes N. Melkote (2009). “Design and Analysis of Computer Experiments With Branching and Nested Factors.” Technometrics, 51(4), 354-365.

[4] Qiang Zhou, Peter Z. G. Qian, shiyu Zhou (2011). “A simple approach to emulation for computer models with qualitative and quantitative factors.” Technometrics, 53(3), 266- 273.

[5] Shan Ba, Willian R. Myers, William A. Brenneman. (2015). “Optimal sliced Latin hypercube designs.” Technometrics, 57(4), 479-487.

[6] Haitao Liu, Jianfei Cai, Yew-Soon Ong (2018). “Remarks on multi-output Gaussian process regression.” Knowledge-Based Systems, Volume 144, 102-121.

[7] Peter Z. G Qian, Huaiqing Wu, C.F. Jeff Wu (2018). “Gaussian Process Models for Computer Experiments With Qualitative and Quantitative Factors.” Technometrics, 50(3), 383-396.