为实现对高耸组合结构界面脱空部位的快速自动化识别检测，以张靖皋长江大桥等组合结构桥塔为背景，开展桥塔足尺模型脱空检测试验，通过对声场空间分布数值模拟、时频响应特征分析及卷积神经网络（convolutional neural network, CNN）模型训练，提出一种适用高耸组合结构界面脱空检测的自动化装置及基于声信号的脱空损伤检测深度学习方法。结果表明：基于自动化装置采集的声信号分析方法可作为高耸组合结构脱空损伤自动化检测识别新方法，构建的卷积神经网络模型能够实现对结构损伤、健康及无效数据的高精度分类，分类识别准确率达96.8%，自动化检测装置可实现对高耸组合结构界面脱空的准确与快速检测，该装置及方法可提升超高组合索塔脱空检测的自动化程度降低检测安全风险。

To achieve rapid automatic detection and identification of void damage in high-rise composite structures, a bridge tower full-scale model was tested for damage using Zhangjinggao Yangtze River Bridge's composite structure tower. Through numerical simulation of sound field spatial distribution, time-frequency response characteristics comparison analysis, and convolutional neural network(CNN) model training and visualization. An automatic device for void detection of high-rise composite structures and a deep learning detection method based on acoustic signals were proposed. The results demonstrate that the acoustic signal analysis method based on automatic device acquisition can be used as a new approach for automatic detection and identification of void damage in high-rise composite structures. The constructed CNN model can achieve high-precision classification of structural void state, and the recognition accuracy is 96.8%. The automatic device and intelligent detection method enable automatic real-time detection and classification of high-rise composite structures, improving automation and reducing safety risks.