Data-driven sensor fault diagnosis for vibration-based structural health monitoring under ambient excitation

In vibration-based structural health monitoring (SHM), the early detection of sensor faults is key to preventing false alarms and misleading conclusions on the condition of monitored structures. Since sensor networks are exposed to hostile environments, they are prone to unexpected errors that might influence the quality of measured data. This paper proposes a novel method for detecting and isolating faulty sensors from vibration response data by establishing an overdetermined system between the measured signals and the actual motion. The method assumes a rigid body motion of the monitored system, describable by a limited number of degrees of freedom (DOFs), to define the overdetermined relation between the sensor outputs and the system’s DOFs. The concept is later extended to systems not governed by rigid body motions by considering their vibration mode shapes. The robustness of the proposed methodology is demonstrated using vibration response data from an experimental monitoring campaign.