Session: DAC-04-03: Data-Driven Design

Paper Number: 114806

114806 - On-the-Fly Dual Reduction Method on Transient Fluid Topology Optimization 

Transient topology optimization (TO) requires huge compu- tational resources in terms of both time and storage. In this study, we have applied the proper orthogonal decomposition (POD) based on-the-fly dual reduction method on TO problems for time-dependent incompressible fluid channel optimization. Time-dependent Navier-Stokes equation was involved to depict the fluid motion. Total energy dissipation over a given period was considered as the cost function, and total fluid volume was constrained during optimization process. POD based reduced order modeling (ROM) was applied to both the primal equations and the adjoint equations for reduction. We examined the re- duction efficacy though examples including U-bend channel and nozzle. Tolerance and snapshot set size were studied as hyper- parameters to improve the reduction performance. The best cases can reach 99% of skip ratios, 50% reduction in computational time, and 96% saving of memory storage.

 

In this study, we have applied the on-the-fly dual reduction method to the transient incompressible fluid TO problems. The proposed method was implemented with FEM and investigated through 3 examples, U-bend channel, nozzle, and blunt body flow. In the first 2 cases, the ROM method can provide similar final designs with FOM results, with an appropriate final tolerance. The total reduction time can reach above 50%, and the total reduction memory can reach about 95%. This suggests that the on-the-fly dual reduction method is useful for transient fluid TO. However, in the blunt body flow where coherent structures emerge in the flow field, the current scheme fails to provide substantial reduction.

 

Presenting Author: Tianye Wang University of Wisconsin, Madison

Presenting Author Biography: Research Assistant
Ph.D. student

Authors:

Tianye Wang University of Wisconsin, Madison
Xiaoping Qian University of Wisconsin, Madison