本研究結合R4SID系統識別法與狀態空間DLV損傷探測法,由識別參數重建等效柔度矩陣而發展出無參考模型結構損傷探測技術。R4SID係建構於確定-預測性模型,故能提供更完整之系統資訊,且對噪訊有較佳之過濾能力,經由適當的動態反應監測萃取出完整之結構系統參數,得以由結構之全域反應定位出局部破壞。本研究提出由等效柔度矩陣與破壞載重向量之乘積計算結構各柱節點位移,並轉換為正規化層間變位指標。凡層間變位指標數值接近零者,物理上即代表柱剪力為零,故可取代應力指標做為判斷受損桿件之依據,因而能在無參考模型之下篩選出潛在受損柱位。本文並提出由雙主軸方向層間變位指標交互比對的方式,由其交集定位出最可能受損之桿件,將損傷探測提升到桿件定位的層次。經由三維結構之數值模擬分析結果顯示,在雙向地震輸入、且輸入-輸出訊號有高斯白噪訊汙染的情況下,絕大多數案例都能以本文提出之結構損傷探測分析方法找出設定的破壞桿件,具有相當程度之可靠性及實用價值,且其性能表現普遍優於有參考模型的分析結果。由於本技術採遞迴式演算法,使得計算效率大為提升,故能處理較大規模之系統,有助於未來實際之工程應用。
This study integrates the R4SID system identification method with the method of damage locating vector (DLV) under a state-space framework for structural damage detection. The equivalent flexibility matrix is reconstructed from the identified system parameters to facilitate a reference-free damage detecting tool. Based on a Deterministic-Stochastic model that acquires a more complete set of information of the system, the R4SID algorithm is robust in noise-filtering while being able to identify system parameters in a more complete sense, if the dynamic structural responses are properly monitored. And by which, one may identify local damages from global responses of the structure. This thesis proposes to estimate the nodal displacements of the columns as the matrix multiplication of the equivalent flexibility matrix with the DLVs. They in turn are converted into normalized story-drift indices. Columns with nearly zero normalized story-drift indices (NSDI) correspond equivalently to those with nearly zero shear in physical meaning. Therefore, the NSDI can, in place of the stress index, be used as the basis for damage localization, and allows for screening out the potentially damaged structural members without any numerical model as a reference. This study proposes also an overlapping concept of bi-axial NSDI to nail down the most possibly damaged members. As a result, the proposed reference-free damage detection technology has advanced to locating the damaged member(s) rather than just the damaged stories. Numerical simulations of a three-dimensional shear-type building, under biaxial earthquake excitations with the input-output signals contaminated by Gaussian white noises, demonstrate in most cases that the proposed scheme can successfully identify the designated damages. The proposed tool proves reliable to a large extent with great potential in practical application. Performance of the tool in damage identification is in general better than those obtained using a numerical model as the reference. Moreover, since a recursive algorithm is adopted, the efficiency of numerical computation has been greatly improved to allow for handling of large systems favorable for practical applications.