An output-only damage identification method based on reinforcement-aided evolutionary algorithm and heterogeneous response reconstruction

Abstract

The absence of excitation measurements may pose a huge challenge in the application of many damage identification methods since it is difficult to acquire the external excitations, such as wind load, traffic load. To deal with this issue, a novel output-only structural damage identification approach based on reinforcement-aided evolutionary algorithm and heterogeneous response reconstruction with Bayesian inference regularization is developed. On the one hand, heterogeneous measurements (e.g., displacements, strains, accelerations) are rescaled and reconstructed with the aid of Bayesian inference regularization technique. Structural damages are identified by minimizing the discrepancies between the measured and reconstructed responses. On the other hand, to solve the optimization-based inverse problem, a reinforcement-aided evolutionary algorithm, named Q-learning hybrid evolutionary algorithm (QHEA), integrating Jaya algorithm, differential algorithm, and Q-learning algorithm is proposed as search tool. To validate the feasibility and applicability of the proposed method, numerical studies on a three-span beam structure and laboratory tests on a five-story steel frame structure are carried out. The effects of data rescaling and data fusion on response reconstruction and damage identification are also investigated. The results clearly demonstrate the superiority of QHEA over other heuristic algorithms and heterogeneous data fusion over a single type of measurement. It is shown that both the locations and extents of the damaged elements can be accurately identified by the proposed method without the information of input force.