鑒於最佳化問題於工程領域中之重要性，近代多種最佳化方法蓬勃發展，其中著名的神經-遺傳法（Neurongenetic Method）人工智慧學習系統，被廣泛應用於各工程領域的問題中。神經-遺傳法由類神經網路及遺傳演算法結合而成，具多種結構組合方式，過往文獻中，往往僅就單一組合進行最佳化分析。本研究之目的，在於探討各組合間之異同及選用效率。訂最佳化問題為臨床上用於治療股骨近端骨折之雙螺絲骨釘（DSN）的遲滯螺絲強度，期能減少臨床上破壞及鬆脫等術後失效的發生。
本研究之方法，首先透過：使用直交表規劃欲探討之遲滯螺絲設計參數及其水準、使用SolidWorks及ANSYS Workbench建立遲滯螺絲模型並模擬其目標值、使用田口品質工程法計算各設計參數對目標值之貢獻度，進行所需資料之準備。接著透過：結合能夠找出目標函數模型的類神經網路、以及能夠找出遲滯螺絲設計參數最佳範圍的遺傳演算法，進行數學計算軟體Matlab對神經-遺傳法運算程式的撰寫與運算。最後透過：使用FEM軟體進行目標值之驗證。而上述步驟，將於各人工智慧學習系統組合中，各進行抗壓強度及抗拉強度的分析；前者訂最大張應力為目標值，後者訂總反作用力為目標值；嘗試之人工智慧學習系統組合依序為：單輸出神經-單目標遺傳法、多輸出神經-單目標遺傳法、單輸出神經-多目標遺傳法及多輸出神經-多目標遺傳法。
本研究之結果，顯示當對最佳化問題個別作單一目標值最佳化時，選擇「單輸出神經-單目標演算法」，可在相近的最佳預測目標值、相近的總系統運算時間下、得到測試平均百分誤差較「多輸出神經-單目標演算法」低約27％的類神經網路。而當對最佳化問題同時作多個目標值最佳化時，選擇「多輸出神經-多目標演算法」，可在相近的最佳預測目標值、相近的原目標值保留度下、較「單輸出神經-多目標演算法」節省26.73％的總系統運算時間。

Due to the importance of optimization method which’s been sprung up these decades in engineering field, Nurogenetic method was used in many disciplines. Literatures only concerned each hybrid type of ANN (artificial neural network) and GA (genetic algorism) per optimization problem. The purpose of this study is comparing the difference between each Nurogenetic type. Thus, these methods were applied to finding optimal design of the orthopedic screw which plays an important role in curing proximal femur fracture and needs to resist the failure of breakage and loosening.
First, the essential data were prepared: arranging the design factor’s level by orthogonal array, building model and finding their layout by Solidworks and ANSYS Workbench, and determining the design factor’s contribution by Taguchi method. Next, ANN and GA were calculated for the optima by Matlab program. In final, data were validated by finite element analysis software. There are four Neurogenetic type were researched in this study: Singleoutput-neuro-singleobjective-genetic method, multioutput-neuro-singleobjective-genetic method, singleoutput-neuro-multiobjective-genetic method and multioutput-neuro-multiobjective-genetic method. Bending strength analysis and pullout strength analysis were taken in each method above.
From the result, when dealing with singleobjective optimization problem, applying singleoutput-neuro-singleobjective-genetic method could get a better ANN model which has a testing mean percentage error lower about 27% under close optimal predict value and close calculating time. When dealing with multiobjective optimization problem, applying multioutput-neuro-multiobjective-genetic method could get a better Neurogenetic model which saves times about 26.73% under close optimal predict value and close original optima preservation.

[1] 沈宗憲等27人，醫學科學叢書12臨床大體解剖學，國立台灣大學醫學院，台北 (2000)
[2] 侯勝茂、林晉，鎖定式骨髓內釘之基礎科學與臨床運用，合記圖書出版社，台北 (2003)
[3] 張季娜，田口是品質工程導論，中華民國品質學會，台北 (2003)
[4] 蘇木春、張孝德，機器學習:類神經網路、模糊系統以及基因演算法則，全華圖書股份有限公司，台北 (2007)
[5] Michael Negnevitsky, Artificial Intelligence: A Guide to Intelligent Systems, 2nd Edition, Pearson Education , Inc. (2010)
[6] Nguyen Ngoc Luong, “Design Optimization Study of Orthopedic Screw by Taguchi Method and Neurogenetic Method,” Master’s thesis, National Taiwan University of Science and Technology, Taipei, Taiwan (2011).
[7] Chao-Ton Su and Tai-Lin Chiang, “Optimizing the IC wire bonding process using a neural networks/genetic algorithms approach,” Journal of Intelligent Manufacturing, Vol. 14, pp. 229-238 (2003)
[8] B. Ozcelik, and T. Erzurumlu, “Comparison of the warpage optimization in the plastic injection molding using ANOVA, neural network model and genetic algorithm,” Journal of Materials Processing Technology, Vol. 171, pp.437-445 (2006)
[9] K. Panneerselvam, S. Aravindan & A. Noorul Haq, “Hybrid of ANN with genetic algorithm for optimization of fictional vibration joining process of plastics,” International Journal of Advanced Manufacturing Technology, Vol. 42, pp.669-677 (2009)
[10] S. J. LIAO and W. H. HSIEH, “Shrinkage and Warpage Prediction of Injection-Molded Thin-Wall Parts Using Artificial Neural Networks,” Polymer Engineering and Science, Vol. 44, No. 11 (2004)
[11] Mirigul Altan, “Reducing shrinkage in injection moldings via the Taguchi, ANOVA and neural network methods,” Materials and Design, Vol. 31, pp.599-604 (2010)
[12] Ching-Chi Hsu, Ching-Kong Chao, Jaw-Lin Wang and Jinn Lin, “Multiobjective Optimization of Tibial Locking Screw Design Using a Genetic Algorithm: Evaluation of Mechanical Performance,” Journal of Orthopaedic Research (2006)
[13] 徐慶琪，「骨螺絲之結構設計與生物力學分析」，博士論文，國立台灣科技大學，台北 (2005)
[14] 張傑富，「腰椎椎弓足螺絲之多目標最佳化設計：有限元素分析與生物力學測試」，碩士論文，國立台灣科技大學，台北 (2010)
[15] 葉駿達，「鎖定式骨髓內釘之遲滯螺絲多目標最佳化設計」，碩士論文，國立台灣科技大學，台北 (2011)
[16] 郭鳳如，「珈瑪骨釘之遲滯螺絲生物力學研究：探討螺絲灣區破斷、螺絲咬和界面鬆脫與近端股骨骨切風險」，碩士論文，國立台灣科技大學，台北 (2011)
[17] Ching-Chi Hsu, Ching-Kong Chao, Jinn Lin and Sandy Tri Putra, “A Neurogenetic Approach to a Multiobjective Design Optimization of Spinal Pedicle Screws,” Journal of Biomechanical Engineering, Vol. 132 (2010)
[18] Jinn-Jong Sheu and Chi-Yuan Chen, “An Integration Method of Artificial Neural Network and Genetic Algorithm for Structure Design of a Scooter,” ICIC 2006, LNCS 4113, pp. 655-662 (2006)
[19] 康鳳梅、許榮添、詹世良，循序學習SolidWorks，全華科技圖書股份有限公司，台北 (2006)
[20] 康淵、陳信吉，ANSYS入門，全華科技圖書股份有限公司，台北 (2009)
[21] 蔡國忠，ANSYS/Workbench有限元素分析與工程應用，知城圖書，台北 (2008)
[22] William J. Palm III, Introduction to Matlab 7 for Engineers, Mc Graw Hill Education Inc.
[23] 洪維恩，Matlab 7 程式設計，旗標出版股份有限公司，台北 (2010)