Application of statistical monitoring using latent-variable techniques for detection of faults in sensor networks

The ability of using measurements collected through a sensor network for detecting and locating damage via structural health monitoring algorithms relies on accurate sensor measurements from the deployed sensor network, and therefore, it can be affected by the presence of malfunctioning and/or faulty sensors. In this article, three sensor fault detection and identification techniques based on statistical monitoring, using latent-variable techniques, were implemented, evaluated, and compared with respect to their capability to detect and identify faulty sensors using case studies from an analytical three-dimensional truss and from an actual cable-supported bridge in the metropolitan Los Angeles, California region. It is shown that the leading sensor fault detection algorithms are effective in detecting certain classes of sensor failure mechanisms but are of limited utility when dealing with representative types of sensor faults encountered in typical structural health monitoring of civil infrastructure systems.