ABSTRACT

Topology optimisation aims to determine the optimal material distribution within a given domain to minimise an objective functional (e.g., structural compliance or pressure drop in fluidic systems) subject to geometric and physical constraints. These objective functionals typically depend on the solution of full-order models, such as linear elasticity or Navier-Stokes equations, which can be computationally demanding. Moreover, the cost can become prohibitive when multiple queries need to be computed for different user-defined parameters. In this talk, we present a hybrid framework combining physics-based and data-driven information to accelerate topology optimisation pipelines by means of artificial intelligence. Parametric surrogate models are trained to predict an educated initial guess for the optimisation algorithm. The proposed approaches integrate autoencoder-based methods, learned mapping techniques, and graph-based architectures. The resulting AI surrogates enable efficient predictions of quasi-optimal layouts for topology optimisation problems depending on user-defined parameters, in both interpolation and extrapolation regimes. Numerical experiments are presented to showcase the improved performance of these AI-augmented topology optimisation strategies in computational mechanics.