A few-shot electromechanical impedance monitoring method based on a modified prototype network

Bolt loosening monitoring is of great significance to warrant the reliability and safety of bolted structures. The electromechanical impedance (EMI)-based evaluation is effective to perceive bolt loosening. However, EMI signals are highly prone to contamination by temperature fluctuation. Deep learning (DL) based EMI is a promising technique for accurate damage detection in the temperature variation environment. However, DL needs a lot of data to train, which is usually very difficult to collect sufficient structural damage data in real word scenarios. This paper proposed a few-shot EMI monitoring method based on a modified prototype network for bolt looseness detection under temperature varying environment. The approach features a conversion method of the impedance signal to image based on the Hank matrix. A modified prototype network is then developed. An experimental study was carried out on a bolted joint. EMI signals under different bolt loosening conditions were measured in a temperature variation environment. An impedance analyzer and a self-made small lightweight monitoring device were both used to measure the EMI signals to test the cross domain scenario. The proposed method was compared with the transfer learning methods and other typical few-shot learning methods. The experiment results show that the proposed few-shot EMI method can obviously improve the monitoring accuracy of bolt loosening with few samples.

Automated summary

This paper proposes a few-shot EMI monitoring method for bolt looseness detection under temperature variation environment. The approach features a conversion method of the impedance signal to image based on the Hank matrix. Experimental results show that the proposed method can significantly improve the monitoring accuracy of bolt loosening with few samples.