應用無線監測技術之唯輸出系統結構損傷探測

Abstract

本研究結合隨機子空間系統識別法(Stochastic SubspaceI dentification,SSI))與狀態空間DLV損傷定位法，針對唯輸出系統之地震監測反應進行結構損傷定位。經由數值模擬分析及一系列之振動台地震模擬試驗，證明吾人可在未知輸入擾動資訊的條件下，透過結構全域之地震反應歷時定位出平面構架結構之局部受損位置﹙樓層﹚。本研究採用SSI-COV以及結合多變量分析之SSI-PLS、SSI-MLR、SSI-CCA及SSI-ECCA等在隨機子空間系統架構下所發展之系統識別分析方法，其概念為透過輸出向量序列間之協方差關係，建立包含原變量特性之降階系統，減少矩陣維度過高對精確性之影響。數值模擬分析結果顯示，結合多變量分析之SSI識別法，由於引進了卡爾曼過濾器之觀念，因此無論對於損傷程度之敏感性或是噪音干擾之適應性，皆較SSI-COV識別法為佳。振動台試驗採用無線監測技術，以台灣大學所發展之無線傳輸系統NTU-WSU結合加速規進行動態反應監測。試驗結果顯示，包含SSI-COV在內的五種系統識別法結合DLV損傷探測法多能有效定位出受損樓層，其中以SSI-ECCA之結果最佳，在單一樓層受損及複數樓層受損案例中皆能將所有受損樓層定位出，證明其於實例應用上之可行性。惟五種方法都會發生將未受損之ㄧ樓誤判的情況，主要是輸入擾動乃不具穩態特性之隨機過程(白噪音案例亦因振動台性能之限制未能忠實呈現高斯訊號的特性)所致。

In this study, a scheme integrated with the stochastic subspace system identification (SSI) and state-space damage localization (DLV) method has been developed for structural damage detection of output-only systems. A series of numerical simulations and shaking table tests has been conducted. Encouraging results indicate that local (storey) damages of plan frames can be identified from global seismic response data without knowledge of the input disturbance. In this study, the SSI-COV as well as SSI-PLS, SSI-MLR, SSI-CCA, and SSI-ECCA techniques developed for system identification under the framework of stochastic state-space system via the multivariate analysis are considered. The concept of multivariate analysis is to identify system parameters of the dynamically equivalent reduced-order system from the covariance matrix consisting of the output state vector sequences in a way that adverse effects on the accuracy of oversized data set can be eliminated. Simulation results indicate that, with the introduction of Kalman filter, the multivariate-analysis-based techniques are more sensitive to moderate structural damages and adaptive to noise than the SSI-COV approach. The wireless sensing technique has been adopted for dynamic response monitoring in the shaking table tests using the wireless communication device, NTU-WSU, developed by NTU, in connection with accelerometers. Results show that the damaged stories of the output-only systems can in general be effectively located with the DLV damage detection method using the system parameters identified by all the methods, among which the SSI-ECCA proves to be the most effective one. With the SSI-ECCA, all damage stories can be identified regardless of single or multiple-storey damage conditions, indicating great potential of practical applications. Nevertheless, in all cases the first storey has been judged as potentially damaged, whether or not it is really damaged. This might be attributed to the non-Gaussian characteristics of the input excitations The white noise processes adopted in the shaking table tests system does not reflect the desired properties of Gaussian due to performance limitations of the hydraulic actuator.