Machine Learning-Assisted Simultaneous Identification and Localization of Impacts on Metallic Structures Using Fiber Bragg Grating-Based Sensor

Abstract

Structural health monitoring plays a critical role in assessing the condition and performance of high-cost infrastructure. Impact monitoring is one of the crucial components of structural health monitoring. A fiber Bragg grating (FBG) sensor-based impact monitoring system has been demonstrated in this work, in which an FBG sensor bonded on a metallic plate picks up the vibration signals due to impacts caused by different materials. Time-domain and frequency-domain features extracted from the acquired data were fed to various machine learning models, and an accuracy of 88.25% was obtained using a random forest (RF) classifier for impact-type classification. Further, for simultaneous identification and localization of impacts, wavelet decomposition of the impact signals was performed to extract additional better features. Using all such features, impacts on the metallic plate were identified and localized at quadrant-level granularity with the highest accuracy of 92.25% using the soft voting classifier.