使用改良式基因演算法與模擬退火進行剪力構架之結構勁度參數修正

Abstract

結構模型修正一直以來都被視為是一個非常重要的研究議題，因為在結構物生命週期的各個階段中，皆會存在一些無可避免之因素，造成結構系統的物理參數受到損害。結構模型的修正可以被認為是一個調整模型的最佳化過程，使得模擬模型之分析結果更能反映出與真實結構物量測取得之資訊的一致性。在過去幾十年中，許多最佳化方法紛紛被提出，其中基因演算法與模擬退火已經各自被驗證，對於結構模型修正問題皆為一套強健的搜尋方法。於本研究中，提出了一套整合了改良式基因演算法與模擬退火的結構模型修正系統架構，假設在各樓層質量已知且不變，也不考慮結構阻尼系統影響的情況下，對勁度進行修正。此外，在整個演算過程中還採用了搜索空間縮減策略，使整個搜尋過程更有效率。為了測試本研究提出之系統架構的可行性，利用兩個數值案例與一個實驗案例分別進行完全量測以及不完全量測來模擬與分析。結果顯示，兩個六層樓剪力構架數值案例之四次模擬分析求得的各樓層勁度，與設計值相比，其平均誤差皆可以達到在1 %以內的準確度；而八層樓剪力構架實驗案例之四次模擬分析，也能準確的識別出各樓層的勁度以及結構的破壞位置。

Structural model updating maybe considered as an optimization process of tuning the model so that it better reflects the observed data from the physical structure being modelled. In the past decades, genetic algorithm(GA) and simulated annealing(SA) have been proved independently that both are a robust search technique for structural model updating problems. In this work, an effective algorithm, combined a modified GA with SA, is proposed to update the structural stiffness parameters of structural models. A strategy of search-space-reduction procedure is adopted, in this algorithm, to improve the effectiveness for whole search process. Two numerical shear-type building cases and one experiment case are employed to verify the performance of the proposed algorithm. Average absolute error of estimated stiffness value are less than 1%, for two 6-DOF numerical cases of know mass system, with 2% noise, assumption does not consider damped system, simulate complete measure and incomplete measure respectively. Finally, the algorithm is applied to assessment the stiffness parameters and detect damage locations of an 8-story shear-type experiment structural model with measured modal data from shaking table test. The simulation results reveal that the proposal algorithm can identify the structural stiffness parameters and locate the damage correctly.