Detection, identification, and quantification of sensor fault in a sensor network

In structural health monitoring (SHM) and control, the structure can be instrumented with an array of sensors forming a redundant sensor network, which can be utilized in sensor fault diagnosis. In this study, the objective is to detect, identify, and quantify a sensor fault using the structural response data measured with the sensor network. Seven different sensor fault types are investigated and modelled: bias, gain, drifting, precision degradation, complete failure, noise, and constant with noise. The sensor network is modelled as a Gaussian process and each sensor in the network is estimated in turn using the minimum mean square error (MMSE) estimation The sensor fault is identified and quantified using the multiple hypothesis test utilizing the generalized likelihood ratio (GLR). The proposed approach is experimentally verified with an array of accelerometers assembled on a wooden bridge. Different sensor faults are simulated by modifying a single sensor. The method is able to detect a sensor fault, identify and correct the faulty sensor, as well as identify and quantify the fault type.