Structural health monitoring research traditionally focuses on detecting damage in members excluding the possibility of weakened joint conditions. Efficient model-based joint damage detection algorithms demand computationally expensive model that may affect the promptness of detection. Deep learning techniques have recently come up as efficient alternative to this cause. These techniques help in predicting occurrence and location of damage in structures based on some automatically identified features embedded in the measured structural response. This article proposes an output-only approach for joint damage detection in which a 1D-convolutional neural network (CNN) has been introduced to locate weakened joints in semi-rigid frames. CNN architecture merges feature extraction and classification simultaneously within a single learning block to automatically extract abstract features from typically 2D/3D signals. Proposed approach further modifies the usual CNN architecture to enable it to handle 1D response signals. Numerical validation is performed on a 2D-steel frame under different damage locations and severities followed by experimental validation on a steel frame structure. The method is observed to be very precise and prompt in detecting single as well as multiple damage scenarios. False alarm sensitivity of the proposed algorithm is also tested and found to be well within acceptable limits.

传统上，结构健康监测研究的重点是检测构件的损坏，而不包括关节状况减弱的可能性。基于模型的有效关节损伤检测算法需要计算量大的模型，这可能会影响检测的及时性。深度学习技术最近成为解决此问题的有效选择。这些技术基于一些嵌入到测量结构响应中的自动识别特征，有助于预测结构中损坏的发生和位置。本文提出了一种仅用于输出的关节损伤检测方法，其中已引入一维卷积神经网络（CNN）来定位半刚性框架中的弱化关节。CNN架构在单个学习块中同时合并特征提取和分类，以从典型的2D / 3D信号中自动提取抽象特征。提议的方法进一步修改了常规的CNN体​​系结构，使其能够处理1D响应信号。在不同损伤位置和严重程度的二维钢框架上进行数值验证，然后在钢框架结构上进行实验验证。观察到该方法非常精确，可以迅速检测到单个或多个损坏情况。还测试了所提出算法的错误警报敏感性，并且发现其良好地处于可接受的范围内。在不同损伤位置和严重程度的二维钢框架上进行数值验证，然后在钢框架结构上进行实验验证。观察到该方法非常精确，可以迅速检测到单个或多个损坏情况。还测试了所提出算法的错误警报敏感性，并且发现其良好地处于可接受的范围内。在不同损伤位置和严重程度的二维钢框架上进行数值验证，然后在钢框架结构上进行实验验证。观察到该方法非常精确，可以迅速检测到单个或多个损坏情况。还测试了所提出算法的错误警报敏感性，并且发现其良好地处于可接受的范围内。