應用勁度修正指標與非監督式模糊類神經網路於剪力構架之結構勁度參數修正

Abstract

本研究提出一個可在不完全量測的情況下，有效率的進行剪力構架結構勁度參數修正的方法。此方法以疊代法為基礎，其中分為兩階段。第一階段是由隨機的粒子群搜尋與啟發式的勁度修正指標來組成，一開始在預設的搜尋範圍內隨機產生多個粒子，接著在每次疊代中，計算各個粒子專屬的修正指標，透過指標來引導各個粒子逐步往目標點進行修正。而第一階段將在粒子群與目標點之間的平均差異小於10%時停止，且搜尋機制進入第二階段。第二階段是非監督式模糊類神經網路，在每次疊代中，根據演算法目前所找到的最佳解為中心，在其周圍隨機產生多個粒子，並以粒子群中幾個與目標較為相近的粒子，根據非監督式模糊類神經網路的重心法來求得最佳解。為了測試本研究之方法的可行性與準確性，將以兩個數值案例來進行測試，分別為6層樓與9層樓的剪力構架。測試結果顯示，本研究之方法可以只依靠4個量測資訊(3個量測點加地面已知固定)，就能準確的識別出各個案例的勁度參數，並且求得的解與期望值之間的差異皆小於1%。接著為了測試本研究之方法是否可以進行破壞位置檢測，將以一個數值案例(6層樓剪力構架)與兩個實驗模型(3層樓與8層樓剪力構架)來進行測試。實驗方式是將模型置於震動台上，量測其未破壞與已知破壞點之震動反應(加速度值)。測試結果顯示，不論是單點還是多點破壞，本研究之方法皆可準確的識別出結構的破壞位置，且其精度在工程可接受的範圍。

This work presents an effective approach for updating structural stiffness parameters from measured incomplete modal data via external excitations, such as earthquakes. The proposed method is based on an iterative approach with two significant strategies. The first strategy is a particle-based random search cooperating with a stiffness revise index. The index is a heuristic notion from structural engineering, and assists to guide the search direction in global for each particle during iterations. This strategy will be stopped when the search procedure approaches to a predefined threshold, average error equals to 10%. The second strategy is based on an unsupervised fuzzy neural network mode to generate solutions by local search with similarity among particles in each iterative. Two numerical cases, 6-story and 9-story shear-type structures, are employed to verify the feasibility and correctness of the proposed approach. The analyzing results revealed the approach can accurate identify the corresponding stiffness parameters of the structure using partial mode shapes with only four appropriate measured modal data (including ground) and average error is less than 1%. Following, two shear-type experiments models, 3-story and 8-story steel frames, are excited on shaking tables, and the measured structural response data (accelerations) are then utilized to verify the capability of the proposed approach to locate the damages. The results revealed that the proposed approach can successfully and correctly identify the damage locations for all cases, whether single damage location or multiple damage locations.