A Scalable Algorithm for Multi-Material Design of Thermal Insulation Components Under Uncertainty

Abstract: This work presents a scalable computational framework for optimal design under uncertainty with application to multi-material insulation components of building envelopes. The forward model consists of a multi-phase thermo-mechanical model of porous materials governed by coupled partial differential equations (PDEs). The design parameter (material porosity) is an uncertain and space-dependent field, resulting in a high-dimensional PDE-constrained optimization under uncertainty problem after finite element discretization. The robust design framework uses a risk-averse formulation consisting of the mean and variance of the design objective to achieve target thermal and mechanical performances and mitigate uncertainty. To ensure the efficiency and scalability of the solution, a second-order Taylor approximation of the mean and variance and the low-rank structure of the preconditioned Hessian of the design objective are leveraged, which uncovers the low effective dimension of the high-dimensional uncertain parameter space. Moreover, a gradient-based optimization method is implemented using the Lagrangian formalism to derive expressions for the gradient and Hessian with respect to the design and uncertain parameters. Finally, approximated $\ell_0$ regularization functions are utilized via a continuation numerical scheme to promote sparsity in the designed porosity. The framework's accuracy, efficiency, and scalability are demonstrated with numerical examples of a building envelope insulation scenario.