采用广义回归神经网络的螺栓松动压电阻抗监测

Bolted joints are widely used in engineering structures.In the long-term service, the bolt pre-tightening force is likely to decrease due to the complex loading. This decrease will in turn induces loose connection failure. Piezoelectric impedance method is an effective method to detect the change of local bolt pre-tightening force quantitatively. However, the measured impedance also changes with temperature significantly, which brings great difficulty to the bolt pre-tightening monitoring. Hence, the temperature compensation in the bolt monitoringis necessary. In recent years, the machine learning methods have been received extensive attention. Therefore, the general regression neural network (GRNN) to compensate the temperature effect on the impedanceis proposed. Then the quantitative monitoring of bolt pre-tightening force at different temperatures was achieved. In this method, a small amount of data from different temperature was used to train a GRNN, and then impedance at any temperature can be predicted with GRNN. Furthermore, the predicted impedance is used as the benchmark to calculate the pre-tightening force indicator. Finally, the efficiency and accuracy of the present method are validated by the experimental results and common effective frequency shift method, respectively.