Methodology of L-curve based model validation in structural identification
Zhang Li-tao1; Li Zhao-xia2; Zhang Yu-feng1; Fei Qing-guo2
1. Key Laboratory of Large-span Bridge Health Inspection and Diagnosis Technology, Nanjing, 211112, China;2. College of Civil Engineering, Southeast University, Nanjing 210096,China
Abstract：A method of model validation is carried out according to accuracy of the model with identification parameters describing the structure to be identified. Identification parameters are usually not selected appropriately, however, due to the absent knowledge of the structure and to the improvements on calculation efficiency. Therefore, there is always inconsistency between the model and the structure and, when it is rather obvious, the traditional method of model validation often fails. In order to deal with the problem, a L-curve based method of model validation was proposed. The method not only exploited the condition that regularization techniques may be adopted to improve the ill-posedness of solutions of structural identification, but also applied the correlation between objective functions of identification results and norms of as a priori knowledge. As it was carried out according to the rules summarized in numerical calculations, the proposed method was expected to improve the above disadvantage of the traditional one. Validation of the given models was carried out using the L-curve based method with one spring-mass system and BENCHMARK structure as numerical and experimental examples respectively. The results showed that the traditional method could not find the best model among the obtained ones in the experimental example while the proposed method could pick out the best one in both the numerical and experimental example, which proved the validity of the proposed L-curve based method of model validation.
2011, Vol. 30 
