基於確定-預測性模型之結構損傷探測

Abstract

本研究旨在發展科技廠房結構健康監測技術，期有助於降低我國科技產業的震害風險。本文結合確定-預測性子空間系統識別技術與狀態空間DLV損傷定位法，由地震監測反應進行結構損傷探測定位。分析時主要考慮了子空間狀態系統識別法(N4SID)與多變量輸出誤差狀態空間法(MOESP)兩種演算法則。經由一系列的數值模擬分析與振動台試驗，證明吾人可透過結構全域之地震反應定位出局部受損位置(樓層)。在交通大學(NCTU)大型結構實驗室之試驗係以移除部分樓層的斜撐模擬破壞條件，國家地震工程研究中心(NCREE)之試驗則以柱端斷面切削模擬破壞條件，兩系列振動台試驗都考慮了單一及多重破壞的情況。其中NCTU振動台試驗採用了台灣大學所發展之無線傳輸系統NTU-WSU進行動態反應監測。系統識別分析考慮在完全觀測與部分觀測條件下進行，結構損傷探測除了以健康結構為比較基準之外，亦探討以既有損傷結構(不良結構)為基礎定位新增損傷或損傷程度加劇的可能性。無論數值模擬分析或振動台試驗結果均顯示，採用N4SID系統識別法得到的損傷探測結果均優於MOESP法，尤其是在多重破壞及部分觀測等更複雜的情況下，其優勢更為顯著。科技廠房之結構損傷探測分析係以一棟典型三疊層式科技廠房為對象，採用N4SID識別法在部分觀測的條件下進行分析，考慮之受損程度約占該樓層勁度的1%，分別落於中間層無塵室的角落與側邊。模擬分析結果顯示，本文所提出之結構損傷探測技術可成功定位出受損樓層，其中以損傷位置在角落者性能表現為佳，初步驗證其應用於科技廠房之發展潛力，未來值得進一步研究。

This study aims at developing a structural health monitoring technique for high-tech industrial fabs, meant to reduce the seismic risk for domestic industries. In this dissertation, a scheme integrated with deterministic–stochastic subspace system identification and the method of damage localization vector (DLV) is proposed for damage detection of structures based on seismic response data. The algorithms considered for system identification are primarily the Numerical algorithms for Subspace State Space System Identification (N4SID) and the Multivariable Output-Error State sPace (MOESP). Via a series of numerical simulations and shaking table tests, it has been shown that local damages (stories) can be identified from the global responses of earthquakes. For the tests conducted in NCTU large-scale structural laboratory, structural damage is simulated by removing part of the diagonal braces, while for those conducted in NCREE, it is simulated by reducing the cross-sectional area of the columns at the bottom ends. Both series of shaking table tests consider single and multiple damage conditions at various locations. The wireless transmission system (NTU-WSU) developed by NTU was adopted to monitor the dynamic responses in the NCTU tests. Both full and partial observation conditions have been considered in the system identification. In addition to the structural damage detection based on comparion with the intact structure, comparsion with the as-damaged (ill - conditioned) structure has also been considered to further identify new or extended damages. It has been shown that the results by using N4SID are consistently better in almost all cases, regardless of numerical simulations or shaking table tests. The advantage of using N4SID is even more pronounced as dealing with more complex conditions such as multi-damage and/or partial observation. The damage localization of fab structures is investigated using a typical triple fab as the object. The N4SID is adopted for system identification under partial observation. Stiffness loss of about 1% in the damaged storey located at either a corner or a side of the structural plane has been considered. Simulation results indicate that the proposed scheme can successfully identify the damaged storey. The performance in terms of damage index is more sensitive as the structure is damaged at the corner. This preliminary study suggests the potential of the proposed scheme for practical use on structural health monitoring of industrial fab structures. It deserves further research.