Low-frequency underwater wave absorption governed by dual-objective constrained deep neural network

In response to the performance bottleneck of traditional underwater acoustic absorbing materials in low-frequency broadband absorption and the non-uniqueness issue in reverse design, this paper proposes a rubber-coated wavy-neck Helmholtz resonator (RWHR) and constructs a dual-objective constrained deep neural network (DO-DNN) framework to achieve the reverse design of its geometric and material parameters. The RWHR is constructed with a rigid metal frame as the pressure-bearing core. The neck adopts a wave-shaped contour to enhance thermo-viscous dissipation, and the inner wall is coated with rubber to optimize the acoustic impedance matching. Firstly, a theoretical analytical model of RWHR is established, and its accuracy is verified through COMSOL finite element simulation. Combined with the energy dissipation cloud map and the velocity field distribution, the synergistic sound absorption mechanism of the waveform neck thermal viscous dissipation and the rubber layer viscoelastic dissipation is revealed. Secondly, the DO-DNN is constructed. In addition to the traditional structural parameter prediction loss, acoustic performance physical constraint loss is introduced. The theoretical model describing the physical behavior of RWHR is embedded as a differentiable constraint term into the network training process. A database containing 703126 samples is constructed based on 8 key structural and material parameters. The results of reverse design for 9 target units show that the maximum error between the peak absorption frequency and the target value is only 5 Hz, and the error of the absorption peak is less than 0.007. This verifies the extremely high accuracy of the model. Finally, through the verification of the four-unit composite structure (FUCS) and the underwater pipe experiment, the designed composite structure has an average absorption coefficient of 0.67 within the frequency range of 1500Hz and still maintains good sound absorption performance of 0.57 under a static water pressure of 10 MPa. The research achieves breakthroughs from structural innovation and data-driven methods. This paper proposes RWHR to address low-frequency sound absorption issues, and constructs DO-DNN to overcome the problem of non-uniqueness in reverse design. It provides new ideas for deep-sea acoustic stealth materials.