Natural frequency error versus response residual in dynamic model updating

Two different regularisation criteria for dynamic model updating are compared in this paper. Updating is performed by using a previously developed technique that solves the dynamic equilibrium equation in the least square sense, i.e. by minimising the force residual. However, the problem is ill-conditioned and sensitive to measurement errors: in order to control the error propagation, the use of some regularisation technique is required. The selected technique is based on the Truncated Singular Value Decomposition (TSVD). The two tested criteria use different methods for selecting the number of singular values to be truncated: in one case, the response residual is used; in the other case, the natural frequency error is chosen. The test structure is a steel frame with welded joints whose FRFs are determined using random excitation. Both techniques yield satisfactory results and can be even combined together to improve the robustness of the procedure.