Sensor Fault Detection and Signal Reconstruction using Mutual Information and Kalman Filters

摘要

Long time monitoring of structures which are difficult to access demands sensor networks that are permanently installed. In the ideal case the sensors should have a longer life-time than the structure they have to monitor. In practice, especially under harsh conditions, the sensors have only limited life expectation. Since sensor faults cause false alarms and thus additional costs for inspections of healthy structures, it is necessary to identify the faulty sensors and to reconstruct the affected signals. For this reason a combined approach for sensor fault detection and signal reconstruction is proposed. The detection method is based on the statistical dependency between the sensor signals quantified by the Mutual Information (MI). For the sensor signal reconstruction a model-based approach using Kalman Filters (KF) is proposed. Simulation examples and acceleration measurements were used for the validation of the methods. It is shown that defect sensors can be clearly localized and the signals can be reconstructed.