Vibration Based Damage Assessment of a Steel Frame Structure Using Support Vector Machine Algorithm

Abstract

The structural damage detection system comes under the vast field of structural health monitoring. This paper deals with the two-stage damage assessment approach, including identification and severity estimation of any damage present in the structure. Free vibrational analysis of the healthy and damaged state of the structure yields two important modal parameters: frequency and mode shape. Eigenvectors, which constitute the mode shape of the structure, are considered for evaluating a damage index by comparing the damaged state with the healthy state. A Normalized Damage Index (NDI)is estimated for the structure subjected to various damage case scenarios. The novel method of estimating NDI provides a unique pattern for each element in the structure. The variation of natural frequency with increasing damage percentage helps estimate damage severity. Support Vector Machine(SVM), with a statistical pattern recognition paradigm, is an efficient supervised Machine Learning (ML) algorithm capable of performing classification and regression analysis. The Kernel-based SVM algorithm effectively identifies damaged elements and estimates each element's severity. A four-storey, three-bay steel frame structure developed in the OpenSees framework is subjected to modal analysis. The results are validated with SAP and finite element-based ABAQUS software. The ability of the proposed model is also verified for a complex 3D structure. The viability of this model is also explored experimentally with a four-storeyed and single-bay steel frame structure. This approach provides an effective way of damage assessment.