ABSTRACT
Structural Health Monitoring (SHM) can be applied to evaluate the health conditions of in-service civil structures and then help decision-makers confirm a proper maintainence plan for safe and sustainable operation. However, inaccurate evaluation results may be reflected by false alarms or missed detections if the monitoring data are distorted by various sensor faults. It is therefore necessary to detect potential sensor faults occurred in SHM systems before evaluating the health conditions of monitored structures (Huang et al. 2015; Huang et al. 2016). Due to the excellent performance in handling non-Gaussianity, Independent Component Analysis (ICA)-based statistical monitoring technique has a great potential for establishing sensor fault diagnosis method for SHM. Unfortunately, there are two major disadvantages to traditional ICA, i.e., the ignorance of dynamic property and the problem of dimensionality reduction. This paper presents a Dynamic Independent Component Analysis (DICA) methodology that successfully addresses these two shortcomings. On the basis of DICA, a sensor fault detection method is established for SHM. The remainder of this paper is divided into four primary sections. First, the theoretical background of traditional ICA-based statistical monitoring is briefly reviewed, and the research motivation of this paper is described. Second, the dynamic property hidden in structural-
matrix,W ∈ℜ ×m m , such that the sources estimated by
s Wx( ) ( )t (2) are as independent as possible. These sources are therefore called independent components. As a direct extrapolation, W estimates the inverse of the mixing matrix, i.e., W A−1.
