Machine Learning-Based Prediction and Analysis of Electromagnetic Susceptibility in Integrated Underwater Transducer

Underwater transducers play a critical role in underwater sensing and communication, but the complex electromagnetic environment inside submarines necessitates research on the electromagnetic compatibility of these transducers. This study first utilizes bulk current injection (BCI) with single-frequency signals to test the conducted susceptibility of the transducer to electromagnetic interference (EMI). Then the electromagnetic susceptibility thresholds of the transducer under different test conditions are compared. It can be concluded that the transducer is more susceptible at lower frequency bands, with unstable susceptibility thresholds. Finally, multiple machine learning techniques are employed to predict the electromagnetic susceptibility of the transducer. The results show that using a BP neural network yields the smallest prediction error for the electromagnetic susceptibility thresholds and has a narrow confidence interval, indicating good prediction stability.