Damage Localization in Plates Using Energy of Acoustic Emission through Gaussian Process Regression

Abstract The modern infrastructures are well equipped with piezoelectric sensors primarily installed for data collection indirectly helps in condition monitoring of the structure. The acoustic source localization (ASL) is one of the domains in which the source of the acoustic emission is identified using guided ultrasonic wave propagation. This paper presents a novel approach for damage localization in the anisotropic plates utilizing the energy of the acoustic response received by sensors. The acoustic signal is filtered and segmented simultaneously through empirical mode decomposition (EMD). For each receiving sensor, the acoustic signal is segmented into at least three intrinsic mode functions (IMF), and the energy of each IMFs is calculated. The nonlinear and complex relation between IMFs energy and the coordinates of impact location is modeled using Gaussian process regression (GPR). For each of the coordinates, there is a separate trained GPR function that relates it to IMFs resulting in two GPR models working in parallel to predict the impact location through the energy response of the signal. The proposed approach is trained as well as validated through the experiment conducted on a carbon fiber-reinforced polymer (CFRP) composite panel instrumented with a sparse array of piezoelectric sensors. The results unveil the performance of the proposed framework that adequately identifies the location of the impact.