A Support Vector Regression-based Approach Towards Decentralized Fault Diagnosis in Wireless Structural Health Monitoring Systems

Abstract

The reliability of sensors in structural health monitoring (SHM) systems is affected by sensor faults that compromise the quality of monitoring, causing erroneous judgment of structural conditions. To ensure reliable operation of SHM systems, techniques for diagnosing sensor faults have been proposed. In wireless SHM systems, embedding techniques for fault diagnosis (FD) into sensor nodes is of increasing importance as sensor nodes process data on board and communicate analysis results instead of large sets of raw data. As a consequence of on-board processing, raw data is frequently unavailable and sensor faults remain undetected. In this paper, a decentralized fault diagnosis approach based on support vector regression (FD-SVR) is proposed. Due to the high accuracy of the support vector regression (SVR), which can be achieved even with relatively small data sets, the FD-SVR approach enables wireless sensor nodes autonomously selfdiagnose sensor faults, enhancing the reliability of wireless SHM systems without a need for large data sets to be used for fault diagnosis. The ability of the embedded FD-SVR approach to detect and isolate sensor faults, increasing the reliability of sensors in wireless SHM systems, is validated in laboratory experiments.