Multiple damage prediction in tubular rectangular beam model using frequency response-based mode shape curvature with back-propagation neural network

Abstract

In the civil infrastructure, the structures made up of tubular sections played an important role due to an economic point of view, lower self-weight, and stability against functional loads. The periodical maintenance or inspection of the structure is mandatory nowadays to achieve the potential utilization of the structural system. In this study, an experimental, numerical, and analytical study is carried out to investigate the structural fault and its severity in a tubular rectangular beam made up of structural steel. Modal parameters are extracted with the help of a dynamic data logger (B&K) by exciting an impact hammer on model structure and extracting a data with a set of unidirectional accelerometers. First, two displacement mode shapes are extracted using obtained modal parameters. The modal parameters are expected to contain environmental noise during experimentation, so, de-noising is must to obtain noise-free data. Artificial neural network training is utilized to reduce the noise from experimental modal parameters. Using the modified modal parameters, the mode shape curvature is obtained, and so-called modified mode shape curvature (MMSC) is used to calculate the curvature damage index. The curvature damage index is appropriate to investigate multiple fault locations with different fault levels in tubular rectangular beam structures.