Multi‐resolution broad learning for model updating using incomplete modal data

Abstract

A novel multi-resolution broad learning (MRBL) approach is proposed for model updating using identified modal data. Due to measurement noise and limited monitoring locations, the identified modal data are incomplete and noise corrupted. Besides, it is inevitable to have modeling errors in finite element models. Therefore, it is nontrivial to establish simple explicit relationship to represent the nonlinear mapping from modal data to structural model parameters. The proposed approach aims to model this implicit mapping using a nonparametric approach. For this purpose, the nonlinear relationship is learnt based on a multi-resolution recursive procedure with expandable broad learning networks. In contrast to conventional deep learning, the proposed approach is computationally very economical. Instead of requiring large volumes of training data, the multi-resolution approach adaptively zooms into the important region for sampling. Hence, satisfactory accuracy in model updating can be achieved by using a feasible amount of training data. Moreover, the broad learning network is expandable to adopt architectural modification, so it can be reconfigured incrementally based on the inherit information from the trained network. To demonstrate the efficacy of the proposed approach, illustrative examples of a shear building and a three-dimensional braced frame with unobserved torsional mode are presented. Finally, an application using real data of Canton Tower is also presented.