Guided wave structural health monitoring with large data sets

Structural health monitoring systems collect and process large collections of data taken over many years of a structure's service. Ultrasonic guided wave systems, in particular, must process an abundance of time-domain waveform data from widely distributed sensors. As few as 8 sensors that transmit and receive ultrasonic waves in pitch-catch mode every 10 minutes can accumulate over one terabyte of data in five to ten years. This number quickly rises as systems grow in size and complexity. As a result, computation and storage efficiency is extremely important, and current guided wave damage detection technologies cannot efficiently process such large data sets. This paper presents an approach based on random projection theory, which scales well with large data sets, to address this challenge. With 1447 experimental guided wave measurements, we demonstrate our algorithms to achieve a 2537 times improvement in computational speed with less than a 3% reduction in accuracy.