結合類神經網路與遺傳演算法於多個分岐管之干涉型消音器最佳化設計

Combining Neural Network and Genetic Algorithms in the Optimum Design of Noise Cancellation Mufflers with Multiply Connected Tubes

謝清中(Ching-Chung Hsieh)

指導教授 ： 張英俊

大同大學/工程學院/機械工程學系所/碩士(2008年)

摘要

本論文結合田口法(Taguchi methods)、類神經網路系統(Neural network system)與遺傳演算法(Genetic algorithm, GA)之最佳化方法，搭配聲學分析軟體SYSNOISE使用邊界元素法(Boundary element method, BEM)於消音器最佳化設計。

論文分二部份：一、干涉型消音器之消音性能：利用將通路分開為二或者更多，再使總路匯集為一來減低噪音。設定尺寸參數有d1(直管直徑)、d2(彎管直徑)、L1(彎管距離直管的距離)以及L2(彎管中心距離)，用d1(直管直徑)為6cm為基礎的消音器，來探討其參數的尺寸變化，對消音性能之影響；二、整合聲學分析軟體SYSNOISE搭配田口法、類神經網路系統與遺傳演算法之最佳化方法，於干涉型消音器之尺寸最佳化設計，針對干涉型消音器單一頻率分為350Hz、500Hz以及650Hz為目標頻率，其輸入組與輸出組數據利用類神經網路建立其網路模式，再結合遺傳演算法搜尋消音器之最佳化尺寸及聲音傳遞損失(Sound Transmission Loss, STL) ，最後將類神經網路與遺傳演算法所設計之最佳化尺寸，用聲學分析軟體SYSNOISE得到相對的STL值。

結果顯示使用類神經網路建立干涉型消音器網路模式，能有效將干涉型消音器針對單一頻率之數學模式簡單化，而之後遺傳演算法所得的最佳化尺寸，對於干涉型消音器之消音性能均有增加，表示類神經網路搭配遺傳演算法在干涉型消音器之尺寸最佳化設計上也有極佳的表現。對於工業界之應用，可有效的減少開發設計時間，加速產品開發與降低錯誤率，更可降低其開發設計時所需的費用。

關鍵字

聲音傳遞損失值 ； 遺傳演算法 ； 類神經網路系統 ； 邊界元素法 ； 干涉型消音器 ； 田口法 ； SYSNOISE ； I-DEAS

並列摘要

The thesis combines the optimum methods of Taguchi methods, Neural Network System and Genetic Algorithm (GA), and uses the Boundary Element Method to the optimum design of muffler by the software of acoustic analysis, “SYSNOISE”.

The thesis is composed of 2 parts: (1) The performance of noise cancellation mufflers: the route is separated into two or more, and converge the entire routes to reducing noise. The setting parameters of dimension included the diameter of the straight pipe (d1), the diameter of curve pipe (d2), the distance from curve pipe to straight pipe (L1), and the distance of center of curve pipe (L2). Using the 6cm of the diameter of the straight pipe (d1) to be a fundamental muffler to investigate the effects of muffler dimensions on the performance of noise cancellation mufflers. (2) Combining the software of acoustic analysis, “SYSNOISE” with the optimum methods of Taguchi methods, Neural Network System and GA, to the optimum design of dimension of the noise cancellation mufflers. Aiming the unit frequency at the noise cancellation mufflers and divide them into three groups including 350Hz, 500Hz and 650Hz as a goal of frequency. The data of input group and output group is created by using Neural Network System, and combining the GA to search the optimum dimension and sound transmission loss (STL) of the muffler. Final, taking the optimum dimension of Neural Network System and GA to get the relative value of STL by the software of acoustics analysis, “SYSNOISE”.

The result shows using the network mode of the noise cancellation mufflers that is created by Neural Network System can efficiently simplify the mathematical mode of aimed the unit frequency at noise cancellation mufflers. For the performance of noise cancellation mufflers, that’s also increasing after gaining the optimum dimension of GA. It shows using Neural Network System collocate the GA also have a good performance on the optimum design of dimension of noise mufflers of interference type. It can also save the time of research and design, accelerate the product research, decrease the rate of error, and especially save the cost of product design for the industrial application.

並列關鍵字

Sound Transmission Loss (STL) ； Genetic Algorithm (GA) ； Neural Network System ； Boundary Element Method ； Noise Cancellation Mufflers ； Taguchi methods ； SYSNOISE ； I-DEAS

參考文獻

[12]Y. C. Chang, L. J. Yeh, M. C. Chiu and G. J. Lai, 2004, Tamkang Journal of Science and Engineering, Vol. 7, No. 3, 171-181, Computer Aided Design on Single Expansion Muffler with Extended Tube under Space Constraints.

[2]C. R. Fuller and D. A. Bies, 1978, Journal of Sound and Vibration 56(1), 45-59, A Reactive Acoustic Attenuator.

[3]C. R. Fuller and D. A. Bies, 1978, Journal of the Acoustical Society of America 63, 681-686, Propagation of Sound in A Curved Bend Containing Curved Axial Partition.

[4]W. Rostafinski, 1974, Journal of the Acoustical Society of America 56, 1005-1007, Transmission of Wave Energy in Curved Ducts.

[5]J. T. Kim and J. G. Ih, 1999, Applied Acoustics 56, 297-309, Transfer Matrix of Curved Duct Bends and Sound Attenuation in Curved Expansion Chambers.