Eventos Anais de eventos
COBEM 2021
26th International Congress of Mechanical Engineering
The influence of the learning data on the reduced order model of laminar non-premixed flames
Submission Author:
Nicole Lopes M. B. Junqueira , RJ , Brazil
Co-Authors:
Nicole Lopes M. B. Junqueira, Luís Fernando Figueira da Silva, Louise Da Costa Ramos, Igor Braga de Paula
Presenter: Nicole Lopes M. B. Junqueira
doi://10.26678/ABCM.COBEM2021.COB2021-0110
Abstract
Combustion is an exothermic chemical reaction between a fuel and an oxidizer. This process is one of the principal sources of anthropogenic energy conversion, and it is present in several economic sectors, such as transport and industries. Although this process has benefits, it also produces pollutants such as carbon monoxide and soot that are harmful to humans and the environment. Computational fluid dynamics (CFD) has often been applied to the study of combustion, enabling optimize the process and control of the emission of pollutants. This numerical methodology enables the analysis of different flame properties, such the components of velocity, temperature, and mass fractions of chemical species. However, reproducing the behavior observed in engineering problems requires a high computational cost associated with memory and simulation time. The reduced order model (ROM) is a machine learning technique that has been applied to several engineering applications, such as thermal and fluid mechanics, aiming to develop models for complex systems with reduced computational cost. In this way, a high-fidelity simplification of complex systems is created from data provided to learn its behavior and its main characteristics, for example, the flame front and the mass fraction of a species. In this work, the different ROMs are created using CFD simulation results. The CFD model solves the mass, species, energy, and momentum conservation equations for a methane/air laminar diffusion flame stabilized in the Gülder burner. This laminar flame is modeled using a skeletal chemical kinetic mechanism known as DRM19. The static reduced order model uses the singular value decomposition (SVD) algorithm to decompose the CFD data and obtain the system's modes. Then, genetic aggregation response surface interpolation is applied on the higher SVD modes, creating the static ROM. This work aims to analyze the effect of different approaches of learning data on the results obtained by the ROM. This analysis includes the velocity, temperature, species mass fraction field. To that end, reduced order models are created to examine the influence of two different methodological choices applied to the learning data. The first analysis is the impact of increasing the number of learning data points on the SVD error and the flame reconstruction. The second is the effect of creating a ROM for each flame's properties separately or else treating the properties as being coupled.
Keywords
machine learning, Computational Fluid Dynamics, Diffusion Flame, methane/air combustion
