Structural damage identification using image-based pattern recognition on event-based binary data generated from self-powered sensor networks

A continuing challenge in structural health monitoring is power availability for sensors to collect and communicate data. A way to minimize the communication power demand is to transmit the minimum amount of information, namely, one bit. Event-based binary signals are generated at sensor nodes according to local rules based on physical measurements, but interpretation at the global level requires dealing with discrete binary data, which implies system information with reduced resolution. This study presents an investigation on approaches for the interpretation of event-based binary data provided by a self-powered sensor network using a pulse-communication protocol for use in structural assessment and damage identification. Pattern recognition (PR) methods based on image data analysis techniques were adapted for such purpose. The methods used were classifiers based on anomaly detection, a Bayesian method, and a nearest neighbor method. To improve the performance of the approach, 2-dimensional principal component analysis and 2-dimensional linear discriminant analysis were used as feature extraction techniques along with a nearest neighbor classifier. The PR methods and the performance of the interpretation algorithms were evaluated by using virtual data from finite element simulations and real data from experiments on plates. The ability of the PR methods to identify service demands, load variations, and localized material degradation was examined. Results indicate that image-based PR methods can be effectively used for structural damage identification in plate-like structures using event-based binary data sets in novel wireless self-powered sensor networks.