Open Access

ABSTRACT

Unsupervised Support Vector Machine Based Principal Component Analysis for Structural Health Monitoring

by Chang Kook Oh¹, Hoon Sohn¹

1 Korea Advanced Institute of Science and Technology, Daejeon, Republic of Korea

The International Conference on Computational & Experimental Engineering and Sciences 2008, 8(3), 91-100. https://doi.org/10.3970/icces.2008.008.091

Abstract

Structural Health Monitoring (SHM) is concerned with identifying damage based on measurements obtained from structures being monitored. For the civil structures exposed to time-varying environmental and operational conditions, it is inevitable that environmental and operational variability produces an adverse effect on the dynamic behaviors of the structures. Since the signals are measured under the influence of these varying conditions, normalizing the data to distinguish the effects of damage from those caused by the environmental and operational variations is important in order to achieve successful structural health monitoring goals. In this paper, kernel principal component analysis (kernel PCA) using unsupervised support vector machine is developed and incorporated with a time prediction model for data normalization by characterizing the relationship between the extracted features and an unmeasured environmental parameter. This method performs a nonlinear principal component analysis by using kernel functions in high-dimensional feature spaces without involving computationally expensive nonlinear optimization. The advantages of the proposed method are demonstrated using a numerical example with comparison results obtained by applying an autoassociative neural network.

---

Keywords

---