Prestressing wire breakage induced by corrosion is hazardous, especially for concrete structures subjected to severe aging factors, such as bridges. Developing an automated monitoring system for such a damage event is therefore essential for ensuring structural integrity and preventing catastrophic failures. In line with this target, a supervised deep learning–based approach is proposed to detect and classify acoustic emissions released by prestressing wire breakage. The application of advanced signal processing techniques is central to this study to determine optimal model performance and accurately detect patterns of various events. Diverse pretrained convolutional neural network (CNN) architectures are explored and further enhanced by incorporating Bottleneck Attention Mechanisms to refine their performance capabilities. Additionally, a novel hybrid model, AcousticNet, tailored for acoustic event classification in the context of structural health monitoring, is developed. The models are trained and validated using an extensive data set collected from controlled laboratory experiments and in situ bridge monitoring scenarios, ensuring comprehensive adaptability and generalizability. The comprehensive analysis highlights that the Xception model, enhanced with a bottleneck module, and AcousticNet significantly outperform other models in capturing intricate patterns within acoustic signals. Integrating advanced CNN architectures with signal processing methods marks a substantial advancement in the automated monitoring of prestressed concrete bridges.

中文翻译：

---

基于声学事件的混凝土桥梁预应力筋断裂自动监测


腐蚀引起的预应力钢丝断裂是危险的，特别是对于桥梁等遭受严重老化因素的混凝土结构而言。因此，开发针对此类损坏事件的自动监测系统对于确保结构完整性和防止灾难性故障至关重要。根据这一目标，提出了一种基于监督深度学习的方法来检测和分类预应力钢丝断裂释放的声发射。先进信号处理技术的应用是本研究的核心，以确定最佳模型性能并准确检测各种事件的模式。通过结合瓶颈注意力机制来探索和进一步增强各种预训练的卷积神经网络 (CNN) 架构，以改进其性能。此外，还开发了一种新颖的混合模型 AcousticNet，专为结构健康监测背景下的声学事件分类而设计。这些模型使用从受控实验室实验和现场桥梁监测场景收集的广泛数据集进行训练和验证，确保全面的适应性和普遍性。综合分析强调，通过瓶颈模块增强的 Xception 模型和 AcousticNet 在捕获声学信号中的复杂模式方面显着优于其他模型。将先进的 CNN 架构与信号处理方法相结合，标志着预应力混凝土桥梁自动监测的重大进步。