Mean-Flow Adjoint Sensitivity Analysis of Unsteady Flow Around Porous Cylinders Using a Homogenized Lattice Boltzmann Method

Abstract

Adjoint-based sensitivity analysis is an indispensable tool for large-scale fluid-dynamic design and distributed control problems, yet its application to unsteady and turbulent flows is frequently hindered by the prohibitive memory footprint of transient checkpointing and the divergence of gradients in chaotic regimes. To address these computational bottlenecks, this paper presents a mean-flow adjoint sensitivity analysis framework for unsteady flows around porous cylinders using the homogenized lattice Boltzmann method (HLBM). Within this framework, solid structures are efficiently modeled as local porous media utilizing a Brinkman penalization approach. We systematically investigate HLBM-based adjoint gradients for drag and energy dissipation objective functionals, transitioning from steady laminar to unsteady, and finally to turbulent flow regimes. For the turbulent case at Re = 3900, a proof-of-concept is conducted where the framework relies on automatic differentiation to automatically generate adjoint kernels containing subgrid-scale (SGS) turbulence models for large eddy simulations (LES), circumventing manual derivation and allowing for a direct comparison against the frozen turbulence assumption (FTA).