Classification of acoustic emissions using modified matching pursuit

We propose methodologies to automatically classify time-varying warning signals from an acoustic monitoring system that indicate the potential catastrophic structural failures of reinforced concrete structures. Since missing even a single warning signal may prove costly, it is imperative to develop a classifier with high probability of correctly classifying the warning signals. Due to the time-varying nature of these signals, various time-frequency classifiers are considered. We propose a new time-frequency decomposition-based classifier using the modified matching pursuit algorithm for an actual acoustic monitoring system. We investigate the superior performance of the classifier and compare it with existing classifiers for various sets of acoustic emissions, including warning signals from real-world faulty structures. Furthermore, we study the performance of the new classifier under different test conditions.