【摘要】 节点连接区域是结构受力最集中、最重要的部位，一旦发生松动，将导致连接失效，进一步引发结构坍塌。针对空间结构螺栓球节点内部的连接螺栓存在隐蔽性且微小松动难以察觉的问题，提出了一种基于压电阻抗和卷积神经网络的螺栓球节点健康监测方法。制作了一个正放三角锥缩尺网架模型，将压电陶瓷传感器(PZT)粘贴在杆件近球节点一侧的外表面，对套筒施加大小不同的扭矩来模拟螺栓球节点不同程度的健康状态，在固定低电平电流激励下，测量压电陶瓷传感器电导信号(阻抗倒数)的变化以判断节点内部螺栓松动是否发生，进一步采用卷积神经网络方法，以归一化的电导信号矩阵为输入，以损伤程度对应的二进制标签为输出，构建卷积神经网络，利用训练好的卷积神经网络对节点松动程度进行识别。研究表明所提方法可以有效监测节点内部螺栓发生的初期小损伤，卷积神经网络具有强大的鲁棒性，更适合于工程实际，为监测螺栓球节点连接的空间结构安全运营提供了一种新思路。更多还原

【Abstract】 The joint area is the most concentrated and important part of the structure. Once joint loosens, it will lead to the failure of the connection and further cause the collapse of the structure. The connecting bolt is located inside the node and the small connection looseness is difficult to be monitored. A new method of health monitoring of bolted spherical joint based on piezoelectric impedance technology and convolutional neural network is proposed in this paper. A triangular flat truss model is made. Different torques are applied to the sleeve to simulate different loosening states of bolted spherical joints. The piezoelectric ceramic transducer(PZT) is pasted on the outer surface of the bar near the spherical node. Under the excitation of fixed low-level current, the changes of conductance(reciprocal of impedance) signal of the PZT are measured to judge the occurrence of bolt loosening inside the node. Further, the convolution neural network method is adopted, with the normalized conductance signal matrix as the input and the binary label corresponding to the damage degree as the output. The trained convolution neural network is used to identify the degree of node looseness. The results show that the method can effectively monitor the initial small damage of the bolts inside the nodes, and the convolutional neural network has strong robust performance, which is more suitable for engineering practice. This study provides a new idea for monitoring the safe operation of the space structure connected by the bolted spherical joints.更多还原