類神經網路於模態識別之應用

Application of Artificial Neural Network to Structural Modal Analysis

蘇信華(Hsin-Hwa Su)

指導教授 ： 陳振華

中原大學/工學院/土木工程研究所/碩士(2001年)

https://doi.org/10.6840/cycu200100442

摘要

橋梁的抗風設計的依據為臨界風速(Critical Wind Velocity)，而臨界風速則根據以氣動力顫振導數(Flutter Derivatives)之氣動力計算而得。本文應用類神經網路(Artificial Neural Network)進行結構模態識別。由於類神經網路的處理能力及學習精度，可訓練學習橋梁結構之反應特性，藉此得到代表此特性之網路參數，即連結加權值和門檻值，再利用權值建立識別模式，以識別出氣動力顫振導數值。另外，本文首先以數值模擬的方式，加入適當雜訊，模擬實際風力量測情況，其風力包括考慮自激力(Self-excited Force)作用與同時考慮自激力與抖振力(Buffeting Force)同時作用情況下之結構反應，除考慮雜訊之影響外，亦考慮數據型式之影響，以及抖振力對顫振導數之影響，並與其他系統識別方法結果比較，以驗證所建立的類神經網路識別方法(IDNN)之可靠性與正確性。另外，也以兩個斷面模型之風洞試驗(Wind Tunnel Test)資料進行分析，其結果皆顯示本文採用之方法可精確地應用於模態識別。最後，利用現地試驗方法，如衝擊試驗和微動試驗，取得結構物在受到外力下的反應，利用類神經網路識別方法，以求得橋梁之動力特性參數。

關鍵字

抖振 ； 風洞試驗 ； 顫振導數 ； 系統識別 ； 模態識別 ； 現地試驗 ； 類神經網路

並列摘要

Flutter derivatives are important aeroelastic parameters for representing the aeroelastic forces of a cable-supported bridge, and have employed to determine the critical wind velocity for flutter instability. The main purpose of this paper is to establish a procedure of identifying flutter derivatives for bridge deck by using artificial neural network, which has not been reported before. The artificial neural network consists of three layers, namely input layer, hidden layer, and output layer. The aeroelastic responses of the bridge under consideration are used to train the artificial neural network by using BP technique. Then, the weighting matrices in the network are used to determine the structural dynamic characteristics (e.g., frequencies, damping ratios) under wind loads or not. Finally, the flutter derivatives are determined from the identified dynamic characteristics. The accuracy and applicability of the proposed procedure are demonstrated by comparing the present results with those from other techniques in processing numerical simulation data and measured data of section model test. Besides, the proposed method can be applied to process the field test data of bridge (e.g. impact test and ambient test), the results are quite accurate compared with those from ARV identification technique.

並列關鍵字

buffeting ； system identification ； wind tunnel ； field test ； flutter derivatives ； modal analysis ； artificial neural network

參考文獻

1. Baldi, P., and Hornik, K., “Neural Networks and Principal Component Analysis : Learning from Examples Without Local Minima,” Neural Network, 2(1), 53-58, (1989).

2. Barai, S. V., and Pandey, P. C., “Vibration Signature Analysis Using Artificial Neural Networks,” Journal of Computing in Civil Engineering, 9(4), 259-265 (1995).

3. Barnard, E., “Optinization for training neural nets,” Trans.Neural Net, IEEE, 3(2), 232-240(1992).

4. Battiti, R., “First- and Second- Order Methods for Learning: Between Steepest Descent and Newton’s Method,” Neural Computation, 4, 141-166 (1992).

5. Billings, S. A., Jamaluddin, H. B., and Chen, S., “A Comparison of the Back-propagation and Recursive Prediction Error Algorithm for Training Neural Network,” Mechanical Systems and Signal Processing, 5, 233-255 (1991).