Reducing false alarms in structural health monitoring systems by exploiting time information via Binomial Distribution Classifier

This paper proposes an approach for anomaly/damage detection via structural health monitoring (SHM) systems based on a Binomial Distribution Classifier (BDC). The approach consists of two monitoring levels, labeled as alert and alarm states, respectively, where a damage index (DI) is computed and tracked over time. The alert state is reached when the DI exceeds a given threshold. At this stage, the BDC algorithm starts counting the number of DI values above the threshold within an observation window and computing the related probability of occurrence by using the binomial probability distribution. If the probability falls below a desired limit, the alarm state is triggered. Conversely, the SHM system returns in the non-alert condition. The proposed approach is discussed and evaluated through case studies involving both simulated and experimental data. In the examples, the DI is computed using the Mahalanobis distance of the monitored modal frequencies. The results demonstrate the capability of the BDC to reduce false alarms while preserving the probability of detection.