| The gradual loss of sensor performance, the so-called soft sensor failure, has been a key issue in sensor networks. We focus on the sensor failure detection problem, which includes the occurrence of multiple soft failures. Two different Soft Sensor Failure Detection (SSFD) algorithms are developed for the model-based system, one with and one without the pre-recorded sensor data. The Residual Based Soft Sensor Failure Detection (RBSSFD) algorithm is developed and implemented numerically when previously obtained failure-free residual data is available. The Voting Scheme Based Sensor Failure Detection (VBSSFD) algorithm is developed on the assumption that no previously recorded data is available. The algorithms performance is evaluated by employing a dynamic system model (Euler-Bernoulli beam) with multiple sensor failures. The developed SSFD algorithms are effective in detecting both hard and soft sensor failures. |
