A data-driven matched field processing approach for primary/secondary source localization in plates: proof of concept

Matched Field Processing (MFP) is a generalized beamforming method which matches the received data to a dictionary of replica vectors to localize wave scattering sources (e.g., acoustic sources) in the complex media. The approach has also been used for passive structural monitoring and defect detection. The MFP requires an accurate model of medium, and this is a challenge in some applications. To tackle this issue, data-driven MFP has been recently introduced. Data-driven approaches are considered as model-free methods, which perform with no prior knowledge of the propagation environment to localize a source. This paper introduces a data-driven MFP approach for localizing the primary (i.e., impact) and secondary (i.e., defect) sources in plates. The replica vectors are made using the Fast Fourier Transform of the time history responses of the pristine plate under a controlled external excitation. Then, the MFP is implemented to localize the source. For defect localization, a subtraction approach under Born approximation is employed to remove or weaken the signature of the primary source and extract a set of data which purely contains the acoustic signature of the defect. The performance of the method for primary and secondary source localization is evaluated by studying a small aluminum plate, excited by a controlled broadband noise imposed by an impact hammer. A comparative study is carried out to evaluate the performance of the conventional Bartlett and adaptive White Noise Constraint processors in forming the ambiguity surfaces.