A transfer learning-based framework for acoustic emission source localization of real cracks in reinforced concrete

Abstract

Deep learning models have emerged as a promising solution for acoustic emission (AE) source localization, offering adaptability to composite materials like reinforced concrete. However, existing deep learning models often rely on AE data from simulated sources, such as pencil lead breakage or ball impact, which may not accurately represent real cracks, thereby limiting the performance of deep learning models. This paper proposes a novel deep learning model based on transfer learning for AE source localization of real cracks in reinforced concrete components. The proposed model consists of a source model and a target model, both built with a 1-dimensional convolutional neural network (CNN) and fully connected layers. The source model is trained on a simulated AE dataset, and its 1-dimensional CNN is transferred to the target model. The target model is then fine-tuned using limited AE data from real cracks collected during a four-point bending test. The trained model locates AE sources from real cracks by outputting a grid-based probability map. The performance of the model was compared with and without transfer learning. Additionally, the robustness of the proposed model against noise was investigated through field testing on a real bridge. The generalization performance on unseen reinforced concrete components was also examined. Additionally, t-distributed stochastic neighbor embedding was used to analyze the interpretability. The results indicate the effectiveness of the proposed model in AE source localization of real cracks in reinforced concrete components. Additionally, the model remains robust under noisy conditions, indicates its effectiveness in practical AE source localization applications.