Project Summary

In this proposed research, we plan to develop a framework for machine learning enabled information fusion of heterogeneous sensing for infrastructure health monitoring. Traditionally, off-the-shelf sensors such as accelerometers and strain gages have been used to collect real-time measurements of structural responses to facilitate health monitoring. While certain level of successes have been achieved, they also exhibit limitations such as relatively low detection sensitivity to incipient damage [1]. In recent years, significant progresses have been achieved on exploring new sensors and sensing technologies that achieve structural damage identification through active interrogation using such as piezoelectric transducers [2] and magneto-mechanical transducers [3]. These transducers feature high frequency bandwidth and can produce active interrogation responses in high frequency range with small wavelength. Although possessing high detection sensitivity, they may lead to false alarms due to noise and environment variations. Another challenge is that there always exists discrepancy between physical structure and the numerical model such as finite element model [4]. Therefore any model-based inverse identification may be subject to error. We propose to 1) establish a benchmark testbed that incorporates various sensors to assess different sensing mechanisms; 2) develop a machine learning based fault detection and identification approach that can fuse together heterogeneous sensing information that can take full advantage of different sensors and avoid their respective shortcomings; and 3) investigate scalability strategies that can result in actual implementation of the new framework. Potential applications are large-scale infrastructure such as bridges.