Advanced signal processing for structural identification: experimental studies

The aim of this study is to use observed data from a shaking table test to verify experimentally an SVR-based (support vector regression) structural identification approach. The method has been developed in previous work and showed excellent performance for large-scale structural health monitoring in numerical simulations. SVR is a promising data processing method employing a novel e-insensitive loss function and the 'Max-Margin' idea. Thus the SVR-based approach identifies structural parameters accurately and robustly. In this method, a sub-structure technique is used thus the SVR-based analysis is reduced in dimension. Experimental validation is necessary to verify the method's capability to identify structural status from real data. For this purpose, a five-floor shear-building shaking table test has been conducted and two cases, input excitations to the shaking table of the Kobe (NS 1995) earthquake and a Sine wave with constant frequency and amplitude are investigated.