Research on acoustic reconstruction methods of the hull vibration based on the limited vibration monitor data

Abstract

The prediction of hull vibration response is the basis for the quantitative control of cabin noise and underwater radiated noise, which has an important engineering value to obtain vibration characteristics accurately. In this paper, a new method is proposed to reconstruct hull vibration by some monitoring data in low frequencies. The hull vibration response is reconstructed by the monitoring data and the transfer function, and the ill-posed problem of inversion for the transfer function matrix is solved by the total least-squares regularization. A P-GRU (Peak-based Gated Recurrent Unit) neural network correction model is proposed to improve the prediction accuracy of vibration response. Taking a certain cabin structure as an example, some research such as the hull vibration data monitoring, the vibration transfer function calculation, vibration response reconstruction, and correction based on experimental data were carried out, and the simulation calculations and verification of experiment models were obtained. The multi-parameter comprehensive method is adopted to quantitatively evaluate the results between the simulation and the experiment. The results show that the acoustic reconstruction results are in good agreement with the experimental data. The overall level error is within 3 dB, the main peak value correlation coefficient is above 0.9, and the main peak difference is within 4 dB. The mechanism of the rapid and accurate prediction of the low-frequency vibration for the hull based on limited monitoring data is discussed by using a semi-physical simulation method combining the simulation calculation, vibration monitoring, and machine learning. The problem of phase asynchrony was solved, and strong support was provided for the construction of the ship's acoustic digital twin.