PINN-LEE: Physics-Informed Neural Networks Framework to Solve Linearized Euler Equations for Aeroacoustic Predictions

Traditional numerical frameworks for aeroacoustic predictions often face challenges such as the need for carefully designed numerical schemes, problem-specific mesh generation, and stabilizing techniques like absorbing boundary conditions which may increase implementation complexity and compromise accuracy, especially for farfield noise propagation. To address these limitations, this work proposes a novel, mesh-free, Physics-Informed Neural Network (PINN) framework, based on the Linearized Euler Equations (LEE), referred to as PINN-LEE. The PINN-LEE framework incorporates localized aeroacoustic source information directly into the loss function, alongside physical constraints from the governing equations. To mitigate the challenges in applications of PINNs to different prediction scenarios, techniques such as sinusoidal activations, Fourier feature embeddings, and adaptive loss weighting are integrated into the framework. The overall accuracy and generalizability of this approach, when compared with the traditional solvers and available analytical results, is demonstrated on canonical test cases relevant to aeroacoustics predictions. Overall, this work introduces a robust, physics-informed machine learning framework capable of accurate farfield acoustic predictions, with strong potential for engineering applications where traditional methods may be limited. Future research will focus on reducing the computational cost to improve the practical applicability of the proposed framework.