Multiscale Modeling Of Complex Flows In Bubble Column Reactors Based On Lattice Boltzmann Method

Bubble column reactors are widely applied in process industries. Accurate and fast simulation of gas-liquid two-phase flow in bubble column reactors is important for the design and scale-up of bubble columns. Multiscale problems in gas-liquid two-phase flow, accompanied with the highly turbulent flow and complex geometry of internals, make it difficult to achieve the fast and accurate simulation.In view of these problems, the gas-liquid flows at micro-scale, meso-scale and lab-scale have been investigated by Lattice Boltzmann Method (LBM). A multiscale modeling strategy is put forward:a LBM-based direct numerical simulation (DNS) is developed to simulate the micro and meso-scale problems, and a LBM-based mixture model is used to simulate the flow at the reactor scale. Then LBM is coupled with Immersed Boundary Method (IBM) to settle the problems of complex boundary condition in LBM simulation. Finally, the numerical and computation acceleration issues is analyzed in the coupled LBM-RANS (Reynolds-Averaged Navier-Stokes, RANS) model to deal with the highly turbulent flow.1. Direct Numerical Simulation (DNS) for microscale and mesoscale problems in gas-liquid flow. DNS for gas-liquid flow might be an effective approach to understand the physical background of gas-liquid flow and establish or validate constitutive closure models for the models at upper scales. Currently, there are a number of numerical issues in LBM-based DNS model in simulating gas-liquid flow, such as numerical instability for large kinematic viscosity ratios or high Reynolds numbers. In this work, a LBM-based DNS model for gas-liquid flow is developed. It can simulate the gas-liquid flow with large kinematic viscosity ratio (1:103), high Reynolds number and low Morton number. The developed model has been used to study the bubble dynamics from a systematic and multiscale perspective, that is, progressively probing the behaviors of a single bubble, a bubble pair and a bubble swarm. The Z-type rising process of a single 2mm bubble in water and the wavy rise process of two bubbles in given conditions are captured, and these lateral movements of bubbles are essentially induced by bubble wakes. Coalescence or breakage phenomenon is not captured due to the insufficient turbulence intensity in simulation, which could be further settled by increasing the turbulence intensity and using much more fine grids and GPU-Accelerated simulation.2. A LBM-Mixture model:due to the limitation in computational resources, DNS cannot be used in the simulation of industrial-scale bubble column reactor directly at present, and the continuum-based mixture model solved by LBM and accelerated by multiple GPUs may supply an efficient and fast approach for preliminary concept design of bubble column reactors. The LBM-Mixture model is developed and can overcome the weakness of the original LBM model, in which the no-dimensional relaxation time r should be set as 1 to recover LBM to the macro-scale equations of fluid flow. In LBM, the kinematic viscosity v is a function of r, the time step At and the spatial resolution Ah through the equation v= (ι-0.5) △h5/(3△t). If ι is set as 1, Ah has to be smaller to simulate the low kinematic fluid flow, leading to much higher computational cost. The developed model has been applied in the gas-liquid flow simulation of a central or partial aerated flat rectangle bubble column (air-water system). The lateral oscillations of air phase are captured by simulations. The predicted oscillation frequency and time-averaged liquid velocity are in accordance with the reported experimental data. Then we compared the computational speed of multiple GPU-accelerated LBM-Mixture model with Two-Fluid model solved by Fluent. The computation speed of 4 GPU cards accelerated LBM-Mixture model is about 250 fold faster than Two-Fluid model solved by Fluent using 4 CPU cores, when the numerical setups are the same. LBM-Mixture model accelerated by multiple GPUs has established a solid foundation for Real-time simulation or VPE (Virtual Process Engineering) in gas-liquid system.3. Integrating the LBM and IBM for complex geometry and complex boundary flow problems. The industrial-scale bubble columns or gas-liquid stirred tanks are always accompanied with complex internals or boundaries, such as heat exchange tubes, impellers, baffles and so on. This has introduced many technical and scientific problems in LBM simulations. On one hand, it is difficult for LBM to treat the complex boundaries in the classical cubic discretization framework. On the other hand, the generation of body-fitted grid for the Finite Volume based LBM is also difficult and time-consuming. In this work, we have used IBM to deal with complex boundaries and save the grid generation time. The GPU parallel computation has been used for the acceleration of the coupled model of LBM and IBM. The turbulent flow in a Rushton turbine stirred tank has been simulated for model validation. The computation speed of single GPU card accelerated LBM-IBM simulation is about 50 times faster than the computational speed of single CPU core. The computation speed of 4 GPU cards accelerated LBM-IBM simulation is about 270 times faster than 16 CPU cores accelerated Sliding Mesh simulation based on Fluent software. The integration of LBM and IBM accelerated by multiple GPUs is accurate and fast. This laid some foundation for numerical simulation of gas-liquid flow in bubble columns with complex internals, and it is also expected to be coupled with LBM-DNS, LBM-Mixture and LBM-RANS models developed in this work.4. The convergence or computational efficiency problems in the coupling of LBM and RANS models:Compared with DNS or LES for turbulent flow, LBM-RANS is more economical and practical to deal with the turbulent flow in bubble columns. However, while the explicitly-solved LBM-RANS model is efficient, it is hard to converge. On the other hand, the implicitly-solved LBM-RANS model is more stable, but time-consuming in computation. Then we investigated the numerical convergence problems for the coupling of LBM and the explicitly-solved standard k-c model in 2D cases and found that some numerical treatment combinations could improve or ensure the numerical stability. However, those numerical treatments cannot ensure the convergence for 3D simulation. We then implemented the implicit methods for RANS model to improve the numerical stability, and proposed a new method, i.e., the Spatial-temporal Multi-Scale Asynchronous Method, to accelerate the computation.By resolving the above four issues, this work paved the way for LBM simulation to simulate the gas-liquid flows in industrial processes. We demonstrated that LBM could be an efficient approach to investigate the bubble dynamics at micro-or meso-scales with LBM-DNS and the gas-liquid flow in reactor-scale. The latter covers the preliminary concept design of bubble column reactors with LBM-Mixture model, the treatment of complex internals with LBM-IBM and the simulation of turbulence with LBM-RANS.