Open Collections

Zandsalimy, Mohammad

University of British Columbia

Novel stability analysis and improvement tools are designed and developed herein for the improved numerical stability and convergence of computational fluid dynamics simulations. This approach involves the development of a versatile numerical stability enhancement scheme that ensures flexibility, reliability, and scalability, allowing customization to user-specific requirements. Notably, the proposed algorithm is entirely non-invasive and operates independently of the underlying numerical simulation software architecture. Different computational methods are employed for each component of the general stability improvement algorithm and meticulously tested and optimized to identify the most efficient combinations for diverse applications. While the proposed algorithms are capable of modifications to any property of the simulation, this study concentrates on local mesh optimization to enhance the convergence behavior of numerical simulations. It is shown that through the methodologies presented, minor adjustments to a local collection of mesh vertices can yield significant improvements in numerical stability and convergence, often stabilizing fully unstable simulations at a fraction of the computational cost incurred by the flow solver. Innovative methods are provided to address cases where opposing numerical solution modes hinder the effectiveness of the presented algorithm when working with large-scale industrial simulations. Further, novel methods are presented for cell and vertex identification as well as vertex modification vector calculation. The presented algorithms are designed to fully automate the stability improvement and mesh optimization process, minimizing human intervention and decision-making requirements. Furthermore, an innovative approach is presented to remove the unwanted noise in solution modes extracted from the principal component analysis or residual vector analysis. Rigorous testing against various numerical simulations, including our in-house flow solver and the Ansys Fluent industrial solver, demonstrates the effectiveness and efficiency of the presented algorithms. Importantly, implementation with external flow solvers does not necessitate access to the underlying software architecture. The findings highlight the scalability potential of the presented methodologies for large-scale turbulent applications with minimal computational overhead compared to the flow solver.

Text

2024-04-15

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/87838

Mechanical Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/