跳到主要內容

臺灣博碩士論文加值系統

(44.201.99.222) 您好!臺灣時間:2022/12/05 21:35
字體大小: 字級放大   字級縮小   預設字形  
回查詢結果 :::

詳目顯示

我願授權國圖
: 
twitterline
研究生:林立祥
研究生(外文):Lee-Shan Lin
論文名稱:反置半徑式函數類神經網路模式之發展並應用於電漿電弧銲接參數控制
論文名稱(外文):The Development and Application of Inverse RBF Neural Network Model for Parameters Control in plasma Arc Welding
指導教授:紀勝財紀勝財引用關係
指導教授(外文):Sheng-Chai Chi
學位類別:碩士
校院名稱:義守大學
系所名稱:管理科學研究所
學門:商業及管理學門
學類:企業管理學類
論文種類:學術論文
論文出版年:2000
畢業學年度:88
語文別:中文
論文頁數:105
中文關鍵詞:反置類神經網路演算法半徑式函數類神經網路電漿電弧銲接參數控制銲接參數控制系統
外文關鍵詞:Welding AutomationOptimization of Parameters Design for WeldingPlasma Arc WeldingRadius Basis Function Neural NetworkInverse Neural Network
相關次數:
  • 被引用被引用:3
  • 點閱點閱:498
  • 評分評分:
  • 下載下載:0
  • 收藏至我的研究室書目清單書目收藏:0
在目前物質生活日益充裕的社會裡,年輕人選擇就業的方向通常把勞苦的工作排到最後,因此造成製造工業日漸遭遇到技術工人難求和工資高漲兩大難題。對銲接相關產業而言,為了解決上述這兩項問題,必須實施銲接自動化,因為銲接自動化可以減少產業對於銲工的需求,使得產業在人力的調配上有較大的彈性,另外銲接自動化的實施可以大幅提昇產品品質及生產效率,使得產業結構能夠改頭換面,提昇產業競爭力。
然而,銲接電腦製程自動化的技術在國內尚屬開發階段,並未成熟,一般對於銲接參數設定並沒有公式可循,完全憑藉專家過去的經驗或知識來設定參數,一旦超出專家經驗範圍之外,便無法有效設定最佳參數,因此對於銲接參數及品質都是透過經驗法則在運行的作業,根本無法掌握其中的關聯性,因此無法做好整體的銲接品質控制系統,更談不上銲接自動化的控制系統規劃。
本篇論文的研究重點在發展反置之半徑式函數類神經演算法,分別建構A、B兩個模式的反置半徑式函數類神經網路演算法,其共同點都是屬於間接求解模式,先建構前向半徑式函數類神經網路並透過學習範例訓練網路參數間的影響變化,當前向半徑式函數類神經網路訓練完成後,再透過發展出的反置半徑式函數類神經網路演算法建構出反置半徑式函數類神經網路A、B兩模式。而A、B兩模式的差異在於求解模式時的概念差異,A模式完全利用前向類神經網路所訓練好的網路參數,藉由需求品質跟網路輸出值的差回饋到參數需求組合變數做學習。而模式B則是分為兩階段學習,先透過最小近似解定理找出反置半徑式函數的網路鏈結初始權重,並使網路收斂於前向半徑式函數類神經網路的隱藏層輸出值,接著進行第二階段求解模式B,其誤差回饋轉向為由前向隱藏層激發函數值跟反置半徑式函數隱藏層輸出值的差回饋至參數需求組合變數做學習。
將本研究發展的反置半徑式函數類神經網路模式A、B演算法具體建構出系統,並以傳統反置倒傳遞類神經網路跟其做效能比較,發現A、B兩模式系統效能皆比傳統反置倒傳遞類神經網路還優越,其中以模式A的系統效能最好,因此以模式A之網路模式建立電漿電弧銲接整合參數控制系統。
In the materialized society of Taiwan, the youth always lays the labored job such as welding on the last choice during the job hunting. Such that, the insufficiency of skillful workers and the rising of labor cost have become two tough problems in the manufacturing industries. To overcome these two problems, welding automation has its necessity and it may be the best way in the long run for this kind of industries. Welding automation can not only reduce the needs of skillful and costly welding labors, but also can significantly improve product quality, production efficiency and business competition.
Nevertheless, the technique of welding automation is still in the development stage and is not matured yet in Taiwan. In general, there is no any formula to follow for designing the welding parameters. It used to depend on the experience and domain knowledge of the experts. Once a case out of the experience of the experts, it becomes difficult to effectively optimize the welding parameter design. In past, welding parameter design and quality control were operated by rule of thumb. The relationship between them is hard to be handled. To approach the best performance in the overall welding quality, this research attempts to develop an inverse neural network model for controlling welding parameters in order to reach the required quality and build an automatic welding control system in the future.
Therefore, the primary issue of this research is to create new inverse neural network models for automatic welding parameter control. In this thesis, there have two new models of inverse radius basis function neural network (IRBFN) proposed, called model A and model B. The similarity of these two models is that they are belonged to indirect approaches. The development processes are first to build a feed-forward radius basis function neural network (RBFN) which learns the influence of the input to the output by the training cases and then generate the IRBFN algorithms for model A and model B approaches based on the RBFN. The ideas of these two approaches are quite different. Using the steepest gradient decent-based training algorithm, model A approach adjusts the input parameters based on the feedback error between the output of the RBFN and the desired output. In model B approach, the training process is divided into two stages. In the first stage, the initial connection weights between the input layer and the output layer of the IRBFN are derived by the theorem of Least Square Approximation in Linear Algebra and the weights are then trained and converged to the output of the hidden layer of the RBFN. In the second stage, the input parameters of RBFN are trained and reach the solution according to the feedback error between the output of the RBFN on the hidden layer and the output of the IRBFN on the hidden layer.
The proposed two IRBFN approaches are compared with the conventional inverse multi-layer perceptron neural network (IMLPN). From the results, these two IRBFN approaches have better performance in training time and accuracy than the conventional IMLPN. Especially, the IRBFN model A approach has the best performance. Therefore, this research finally will select the IRBFN model A approach as the base to build the parameters control system for plasma arc welding.
目 錄
謝誌 …………………………………………….……………………...Ⅰ
中文摘要 …………………………………….…..…………………….Ⅱ
英文摘要 ……………………………………….……………………...Ⅲ
目錄 …………………………………………….……………………...Ⅳ
表目錄 …………………………………………………………………Ⅵ
圖目錄 …………………………………………………………………Ⅶ
第一章 緒論 …………………………………………………………..1
第一節 研究動機 ……………………………………………………………..1
第二節 研究目的 ……………………………………………………………..2
第三節 研究限制與範圍 ……………………………………………………..2
第四節 研究方法 ……………………………………………………………..3
第五節 研究流程 ……………………………………………………………..4
第二章 文獻探討 ……………………………………………………..5
第一節 類神經網路 …………………………………………………………..5
第二節 倒傳遞類神經網路(Back-propagation Neural Network) …………….7
第三節 半徑式函數類神經網路 ……………………………………………13
第四節 傳統參數設計方法 …………………………………………………20
第五節 類神經網路於反置問題的應用 ……………………………………23
第三章 研究方法 ……………………………………………………30
第一節 參數選擇與品質特性的定義 ………………………………………31
第二節 前向RBF類神經網路的建立 ……………………………………...32
第三節 反置RBF類神經網路之演算法 …………………………………...36
第四章 實例驗證 ……………………………………………………48
第一節 傳統倒傳遞類神經網路模式之訓練結果 …………………………48
第二節 半徑式(RBF)類神經網路之訓練結果 ……………………………..53
第三節 傳統倒傳遞網路跟半徑式函數網路模式之比較 …………………60
第四節 電漿電弧銲接整合參數控制系統 …………………………………63
第五章 結論與建議 …………………………………………………65
第一節 結論 …………………………………………………………………65
第二節 研究貢獻 ……………………………………………………………66
第三節 後續研究與建議 ……………………………………………………67
參考文獻 ………………………………………………………………68
附錄A ………………………………………………………….……….71
附錄B ………………………………………………………….……….76
附錄C …………………………………………………………..………86
參考文獻
【1】 及澤正孝,「田口式品質工程講座(1)開發設計階段的品質工程」,中國生產力中心,1997
【2】 李屬濤,「半自動電弧銲接技術」,大中國圖書公司,民國62年6月。
【3】 徐立章,「電漿電弧銲參數最佳化之研究─田口方法與類神經網路之應用」,義守大學管理科學研究所碩士論文,1998。
【4】 姚景星,「實驗設計」,華泰出版社,1989
【5】 張子浩,「整合線性代數」,文笙書局,1997。
【6】 高順興,「智慧型便利商店電只決策資源系統」,義守大學管理科學研究所碩士論文,1997。
【7】 陳永璋,陳冠宏,「電漿電弧銲原理與電漿銲接之應用(1)」,銲接與切割,第7卷第6期pp.57-62,民國86年11月。
【8】 陳俊男,「類神經網路演算架構在逆向問題之研究」,雲林科技大學碩士論文,1998
【9】 陳永璋,陳冠宏,「電漿電弧銲原理與電漿銲接之應用(2)」,銲接與切割,第7卷第3期pp.44 -54,民國86年05月。
【10】陳弘彬,「整合灰色理論與類神經網路於預測模型之建立-以SIMEX台灣股價指數期貨為例」,義守大學管理科學研究所碩士論文,民國87年6月。
【11】葉怡成,「類神經網路模式應用與實作」,儒林圖書有限公司,民國86年。
【12】葉怡成,「應用類神經網路」,儒林書局,1997
【13】鄭燕琴,「田口品質工程技術理論與實務」,中華民國品質管制學會,1998
【14】廖敏智,「應用模糊神經網路逆向問題演算法於參數設計問題之研究」,雲林科技大學碩士論文,1999
【15】Anderson, J.A.,”The BSB model:A Simple Nonlinear Auto-Associative Neural Network”, In Associative Neural Memories,1993
【16】Craddock, R. J. and K. Warwick, ”Multi-Layer Radial Basis Function Networks: An Extension to the Radial Basis Function”, IEEE International Conference on Neural Networks, pp.700-705, 1996.
【17】Daniel, T. D. and J. N. Hwang,” Solving Inverse Problems By Bayesian Neural Network Iterative Inversion with Ground Truth Incorporation”, IEEE Transactions on Signal Processing , vol 45 11, Page(s): 2749 -2757 , 1997
【18】Colina-Morles, E. and N.Mort, “Inverse Model Neural Network-Based Control of Dynamic Systems”, IEEE International Conference on Control , vol 2, Page(s): 955 —960,1992
【19】Eberhart, R. C., “The Role of Genetic Algorithms in Neural Network Query-Based Learning and Explanation Facilities”, IEEE International Workshop on Combinations of Genetic Algorithms and Neural Networks , Page(s): 169 -183 , 1992
【20】Hatem, M., Deif and J. M. Zurada, “Inverse Mapping of Continuous functions using feed-forward neural network”, IEEE International Conference on Neural Networks, Vol 2 , Page(s): 744 -748,1997
【21】Haykin, S., ”Neural Networks”, A Viacom Company, 1994.
【22】Haykin S., “Neural networks a Comprehensive Foundation”. New York: Macmillan.,1993
【23】Hwang, Y. S. and S. Y. Bang,”Recognition of Unconstrained Handwritten Numerals by a Radial Basis Function Neural Network Classifier”, Pattern Recognition Letters, v8, n7, pp.657-664, 1997.
【24】Kuroe, Y. et al., “A New Neural Network Learning of Inverse Kinematics of Robot Manipulator,” IEEE International Conference on Neural Networks, Vol.5, No.8, P2819-24, July. 1994
【25】Langari, R. and L.Wang, ”A Modified RBF network with Application to System Identification ”, IEEE Conference on Control Application, pp.649-654, 1995.
【26】Laxmidhar B., M. Gopal and C. Santanu, “Self-Organizing Neural Network For Learning Inverse Dynamics of Robot Manipulator”, IEEE International Conference on Industrial Automation and Control, Page(s): 457 —460, 1995
【27】Malinowski, A. and J. M. Zurada and J. H. Lilly,” Inverse Control of Nonlinear Systems Using Neural Network Observer and Inverse Mapping Approach”, IEEE International Conference on Neural Networks Proceedings, Vol. 5 , Page(s): 2513 -2518 ,1995
【28】Narendra, K. G., ”Application of a Radial Basis Neural Network for Diagnosis in a HVDC System”, IEEE Transaction on Power System, v13, n1, pp.177-183, 1998.
【29】Rao, A. V., ”Mixture of Experts Regression Modeling by Deterministic Annealing ”, IEEE Transactions on Signal Processing, v45, n11, pp.2811-2819, 1997.
【30】Ramdane-Cherif, A., et al., “Acceleration of Back-Propagation for Learning the Forward and Inverse Kinematic Equations,” IEEE Second International Symposiumon Neuroinformatics and Neurocomputers, p261-5, September. 1995
【31】Renals, S., “Radial Basis Function Network for Speech Pattern Classification”, Electronics Letters, v 25, p437-439, 1989.
【32】Ogawa, T. and Kosugi, Y. and Kanada, H., “Neural Network Based Solution To Inverse Problems”, IEEE World Congress on Computational Intelligence, The 1998 IEEE International Joint Conference on v13 , p2471 -2476 , 1998
【33】Oguztoreli, M.N., and T.M. Caelli, “An Inverse Problem in Neural Processing.”, Biological Cybernetics 53, p239-245, 1986.
【34】Powell, M.J.D., “Radial Basis Function for Multivariable Interpolation:A Review.”, In IMA Conference on Algorithms for the Approximation of Functions and Data., p143-167, 1985.
【35】Powell, M.J.D., “Radial Basis Function Approximations to Polynomials.”, Numerical Analysis Proceedings, p223-241, 1988.
【36】Suykens J. A. and H. Verrelst, ”On-Line Learning Fokker-Planck Machine”, Neural Processing Letters, v 7, p81-89, 1998.
【37】Satoshi Y. and Hidekiyo I. and Yoshikazu N, “An Inverse Model Learning Algorithm Using The Hierarchical Mixtures of Experts”, IEEE International Conference on Neural Networks, v.5 ,p2738 —2742 ,1995
【38】Tikhonov, A.N., “On Solving Incorrectly Posed Problems and Method of Reregularization .” Doklady Akademii Nauk USSR 153, p49-52, 1963.
【39】Tikhonov, A.N. and V.Y. Arsenin, ”Solution of ILL-Posed Problems”, J.Wiley and Sons, New York,1977.
【40】Zhang, J., “Dynamics and Formulation of Self-Organizing Maps.”, Neural Computation, vol 3,p54-66, 1991.
【41】Zurada, J. M., ”Introduction to Artificial Neural System“, West Publishing Company, 1992.
QRCODE
 
 
 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                               
第一頁 上一頁 下一頁 最後一頁 top