Surrogate model-based multi-objective Bayesian optimisation of porous acoustic barriers

Many engineering challenges involve optimising multiple criteria that often represent conflicting targets, posing significant difficulties for standard methods like gradient-based algorithms. This complexity is especially important in the context of acoustic wave propagation, where noise barriers are designed to attenuate sound pressure level (SPL). Achieving optimal performance requires carefully balancing design factors such as shape and material selection with economic constraints, making the optimisation process both technically demanding and computationally intensive. This paper proposes the development of a noise prediction surrogate model for the multi-objective optimisation of acoustic barriers. This model is developed based on data set generated employing a two-dimensional singular boundary method. The optimisation process is conducted using a multi-objective Bayesian optimisation algorithm, which is applied to the problem of acoustic line source diffraction in the presence of a porous noise barrier. Two distinct barrier configurations are considered: a straight-walled barrier and a T-shaped barrier. With a view to reduce the SPL behind the noise barrier, the set of spanned parameters includes the SPL on the side of the barrier opposite to the source, barrier's height, cap length of T-shaped barrier, porosity, tortuosity, and airflow resistivity of the material, integrating both microstructural and macrostructural aspects into the optimisation. Surface impedance boundary condition is used in the model to represent the dissipation at the surface of the noise barrier. The results demonstrate that the proposed optimisation framework enables efficient exploration of trade-offs to achieve an optimal barrier design that balances acoustic performance, material cost, and shape constraints.