This study introduces a novel Physics-Informed Neural Network (PINN) for 2D acoustic scattering problems, parametrically representing sound source position and wavenumber. A key innovation involves integrating the Burton—Miller-type Boundary Integral Equation (BMBIE) directly into the loss function, mitigating the “fictitious eigenvalue problem” and improving existing BIE-based PINN methods. The model utilizes the Isogeometric Boundary Element Method (IGBEM) for boundary discretization. Numerical experiments showed that the developed surrogate model made an accurate prediction of sound pressure for an unseen source position.
Tanigawa et al. (Wed,) studied this question.