Reconstruction of vibration noise in plate structures based on Data-Physics driven model and transfer learning

Area of Science:

• Vibroacoustics
• Structural Health Monitoring
• Computational Mechanics

Background:

• Understanding vibroacoustic characteristics of structures is crucial for engineering applications.
• Accurate identification and reconstruction of vibration and sound fields are complex challenges.

Purpose of the Study:

• To develop a high-precision data-physics driven model integrated with transfer learning (DPDT) for vibration and sound field reconstruction of plate structures.
• To enhance the generalization capability of the model across different structures using transfer learning.

Main Methods:

• Combined Kirchhoff-Helmholtz integral equation with convolutional neural networks.
• Leveraged physical information to minimize data requirements.
• Integrated transfer learning to improve cross-structural applicability.

Main Results:

• The DPDT model demonstrated superior prediction accuracy, stability, and faster convergence compared to the DPD model.
• Achieved high R2 and normalized cross-correlation, with low normalized mean squared error.
• Showcased robustness and efficacy in sound field reconstruction with limited data points.

Conclusions:

• The DPDT model offers a robust and efficient approach for vibration and sound field reconstruction.
• The method has significant potential for practical engineering applications, including bridge vibration and noise control.