Acoustic emission localization in composite overwrapped pressure vessels via a CNN-based approach

This study proposes a Convolutional Neural Network (CNN)-based approach for Acoustic Emission (AE) source localization in Composite Overwrapped Pressure Vessels (COPVs). The method employs a custom CNN trained on Time-Frequency Analysis (TFA) images generated via Continuous Wavelet Transform (CWT) of AE signals. To enhance localization performance, the architecture incorporates the Difference in Time of Arrival (DToA) extracted from multiple AE sensor channels. A 40.6-liter Type IV COPV, for which a dataset of Hsu-Nielsen signals and related CWT images was generated considering 40 different zones on the vessel, is considered as a case study. To evaluate the effectiveness of the proposed approach, a comparative analysis is performed against the conventional AE localization model for cylinders implemented in state-of-art commercially available systems. Results demonstrate that a minimal sensor setup, comprising two sensors of different types combined with DToA data, achieves an AE source localization accuracy of 87.9%, significantly outperforming the conventional method. The approach demonstrates the feasibility of accurate AE source localization without any knowledge on the wave propagation characteristics and with a limited number of sensors, offering a foundation for data-driven enhancements in AE-based monitoring of composite structures. The dataset of AE signals, documented and available to download, also contributes a valuable benchmark for future research in this domain.