AI-Enhanced Computational Acoustics for Complex Noise Prediction in Fluid–Structure Interaction Systems

Abstract

Precise noise prediction in fluid–structure interaction (FSI) systems has continued to be a major issue in computational acoustics, because they are strongly nonlinear, containing a range of spectral content, and they are particularly expensive to compute due to the high fidelity of numerical simulations. Traditional physics based models, though strong and physically understandable, can be hard to tune to get an ideal balance between accuracy of predictions and suitability of computational effort when applied to complicated geometries and operating conditions. In order to overcome those drawbacks, this work hypothesizes a computational acoustics model with AI enhancements that will be efficient and precise in predicting noise in coupled FSI systems. The suggested methodology combines a physics based FSI solver with a deep learning neural network based on multiphysics simulation data to learn to nonlinearily know about fluidstructureacoustic interactions. The AI component is used as a proxy and remedying model in the computational acoustics pipeline which allows an evaluation of the acoustic source terms faster at maintaining critical physical properties. It comprises physics-guided constraints in the training process in order to improve generalization of the model and stability under different conditions of flows and structure. The framework is tested on representative benchmark FSI problems of flow induced vibration and vortex induced noise generation. The numerical results indicate that the hybrid method will give better results in terms of prediction accuracy as compared to the traditional computational acoustics approaches, especially in reproducing broadband and high frequency noise contents. Further, significant saving in the computational time is obtained and hence the method is applicable to the parametric studies as well as the repetitive design. The results establish AI-assisted computational acoustics as a highly favorable and scalable solution to complex noise prediction in FSI that has a brighter future in the next generation of acoustic analysis of advanced engineering.