A Glass Curtain Wall Bolt Loosening Monitoring using Piezoelectric Impedance Measurement and 1D-CNN-based Transfer Learning

The safety of a glass curtain wall is depending on the tightness of its bolts. Traditional bolt loosening monitoring often uses a number of simple and crude methods to obtain data, such as the hammering method, where the target structure is hammered. Then the prediction is performed using conventional signal processing methods. Such methods tend to be of limited accuracy, time-consuming and laborious, highly dependent on individual experience, and of poor generality. In this research, we introduce the electromechanical impedance method and adopt transfer learning for the structural health monitoring of the glass curtain wall. The impedance of the structures with tight and loose bolts is measured by dual-piezoelectric transducers. The inductance shunt circuit is integrated for enriching the data of the source domain. The features of the structures in the source domain are extracted using a one-dimensional convolutional neural network and then transferred to a target domain. A fine-tuning method is used to improve the accuracy of the monitoring of bolt looseness due to a small sample of the target domain. From what has been discussed above, the proposed method combines EMI and transfer learning, which is convenient to operate, avoids too much human intervention, and has good accuracy and versatility. Experimental analysis proves the effectiveness of the proposed method in the health monitoring of glass curtain walls.