<p>Structural damage detection is critical for maintaining the integrity and safety of buildings and infrastructure. Traditional approaches are often costly, time-consuming, and susceptible to human error due to their reliance on manual feature extraction. To address these limitations, this study proposes a deep learning-based method for identifying bolt damage in point-supported glass curtain walls directly from raw acceleration signals, thereby eliminating the need for manual preprocessing. The dataset was experimentally obtained from a laboratory-scale glass curtain wall structure and used to train a one-dimensional convolutional neural network (1D CNN) to learn discriminative patterns in the data. The model achieved a testing accuracy of 98.3% for undamaged signals and 97.73% for damaged signals, based on the Probability of Identification (PoI) index. The approach was further extended to a multi-class classification task involving various damage levels. The model achieved classification accuracies of 94.60%, 88.35%, 79.50%, and 99.83% for undamaged, single damage, double damage, and triple damage cases, respectively. These results demonstrate the efficiency and effectiveness of a simple CNN architecture in automatically processing vibration data and accurately classifying structural conditions. The proposed method provides a reliable solution for damage detection and establishes a benchmark model for point-supported glass curtain walls, addressing the current lack of a standard baseline for identifying bolt-related damage in such structures.</p>