Validation of Strain Gauges for Structural Health Monitoring With Bayesian Belief Networks

The application of structural health monitoring (SHM) often employs multiple sensors to monitor the state of health and usage of the structures. The fault of any sensor may lead to an inaccurate or even incorrect inference with the collected sensor data, which will accordingly create a negative impact on higher-level decisions for maintenance and services. Thus, sensor validation becomes a critical process to the performance of the whole SHM system. This paper presents the use of Bayesian belief network to validate the reading of strain gauges on an aluminum plate for loading monitoring. The Bayesian belief network is constructed with the training data. The factors investigated in this paper, which may affect the validation process, include sensor configuration, sensor redundancy, and sensor data range for the discretization. The feasibility of using a Bayesian belief network for SHM sensor validation is demonstrated with the experimental results.