A large eddy simulation model for buoyant plume fires
Student thesis: Doctoral Thesis
With the ever rapid advancement of computer processing power continuously reducing computational times, large eddy simulation (LES) techniques are becoming more viable for the better understanding of the complex phenomena that is the realm of fire research. Due to the involvement of a very broad range of length and time scales, extremely large amounts of computational resources are necessary for the numerical simulation of fires. Based on the Reynolds averaged Navier Stokes (RANS) approach, Yeoh and co-workers (Yeoh et al., 2002a, 2002b, 2003a, 2003b; Cheung et al., 2004) demonstrated the significant influence of soot in the modelling of heat transfer, a process which is a combination of radiative and natural convection heat loss. The interactions between turbulence, combustion, radiative heat loss and soot formation are indispensable to the list of considerations in fire modelling. Nevertheless, a loss of fidelity was the major drawback of using the Reynolds averaged Navier Stokes (RANS) approach. To overcome this shortcoming, large eddy simulation (LES) technique is employed to reproduce the unsteady behaviour and accommodate all the possible ranges of length and time scales existing within the turbulent flow of a free standing fire. A well-know computer code – Fire Dynamic Simulator (FDS) (McGrattan, 2004) which facilitates the use of large eddy simulation (LES) model in fire problems was used for comparative analysis. Although extensive validation has been carried out for the FDS code, demonstrating the feasibility of LES being applied to fire dynamic simulations, some of the fundamental physics (e.g. pulsation frequency of a fire plume) has still not been modelled authentically. The main objective of the current research is to investigate the fundamental physics of the fire phenomenon using large eddy simulation (LES) numerical models. As the aim is to elucidate the dynamics of fire, it is crucial to consider the interaction between combustion, radiation effects and soot chemistry due to the highly non-linear behaviour of the fire phenomenon. A microscopic LES model is hence incorporated with the microscopic models of combustion, radiation and finite-rate soot chemistry. Moreover, a two-stage predictor-corrector numerical algorithm is engaged in the present LES computer code. To handle the disparity of the broad ranges of flow and chemical scales involved in fire simulations, the current numerical scheme is established from the low-Mach-number variable-density formulation of Knio et al. (1987) and Najm et al. (1998). A numerical study investigating the flickering behaviour of a turbulent buoyant fire is conducted using LES to validate and examine the effects of coupled turbulence, combustion, soot chemistry, and radiation. A computer code was developed based on the three-dimensional, Favre-Filtered, compressible mass, momentum, energy and mixture fraction and its scalar variance conservation equations. A Smagorinsky subgrid-scale (SGS) model is used to close the equations resulting from the non-linear terms appearing in the form of SGS stress tensor and scalar flux vectors, and in reacting flows in the form of filtered chemical source term. The infinitely fast chemistry is adopted as the combustion model. A combination of a presumed Beta filtered density function and a conservation equation for the scalar variance are used to account for the SGS mixture fraction and scalar dissipation fluctuations on the filtered composition and local heat release rate. A soot model of Moss et al. (1988) which incorporates nucleation and surface growth agglomeration is employed in the present study. Radiative heat loss due to contributions of combustion products and soot particles is accounted for; the radiative heat transfer is handled using the discrete ordinates method (DOM). A two stage predictor-corrector scheme using the projection method has been employed to strongly couple the momentum and continuity equations with density. It has been found that the explicit marching scheme that introduced a second-order Adams-Bashforth time integration scheme for the predictor followed by a quasi second order quasi Crank-Nicolson integration for the corrector is stable to a ratio of the burned and unburned temperature up to a value of 10. A second-order central difference scheme is used for the discretization of the all the governing equations. Solution to the Poisson equation for the pressure is achieved through the Krylov method. An acoustic CFL condition of 0.35 was used to achieve time-accurate solution. The dynamic phenomena of puffing and the formation of large-scale vortical structures are well captured in the present LES model. The entire flow oscillates with a predicted puffing frequency closely agreeing with experimentally measured frequencies. Qualitative comparisons are made to the instantaneous, mean and root-mean-square quantities against available numerical results and experimental data. Also, the possible discrepancies and improvements to the model were highlighted. To further demonstrate the versatility of the present numerical models, performances of different subgrid scale (SGS) models are also reviewed. Three SGS models – Smagorinsky model (SMG), Dynamic Smagorinsky model (DSM) and Lagrangian Dynamic model (LGN), are included and validated with lid-driven cavity flow problems. It was found that the advantages of dynamic SGS models cannot be demonstrated in the case of laminar regimes. In contrast, both Dynamic Smagorinsky model (DSM) and Lagrangian Dynamics model (LGN) provide a better prediction for the turbulent lid-driven cavity flows. This gives further encouragement to persist with this research undertaking and expand into the area of simulating compartmental fires. This area would involve the problem of solving the dynamics of the fire in the near wall region, using perhaps the LES modelling incorporated with dynamic SGS models.
- Fires, Computer simulation, Plumes (Fluid dynamics)