Structural health monitoring system using Support Vector Machine

Abstract

This chapter proposes a new structural health monitoring system that can obtain the detailed damage information from the minimum number of sensors by utilizing the support vector machine (SVM). The feature vectors fed to the SVMs are formed based on the physical relationship between damages and modal frequency changes. The SVM uses a hypothesis space of linear functions in a high dimensional feature space. The simplest SVM model is the so-called linear SVM (LSVM) that works only for the case where data are linearly separable in the original feature space. A series of experiments are undertaken to verify the performance of the proposed approach. The damage is introduced by replacing columns by weak columns. By replacing two columns, the storey stiffness is reduced to 60% of the original stiffness. Although acceleration sensors are installed at all storeys, only the top sensor and the bottom sensor are used for obtaining modal frequencies. The modal frequencies are calculated from time histories of the vibration experiment by applying the subspace identification method.