Hydrodynamics characterization of a choanoid fluidized bed bioreactor used in the bioartificial liver system: Fully resolved simulation with a fictitious domain method

Highlights

• Hydrodynamic performance of a bioartificial liver fluidized bed reactor was evaluated.

• Direct-forcing/fictitious domain method was used for fully resolved simulation.

• Small increase of density ratio converted short-circuit flow to asymmetrical cycle.

• Increase of particle number enhanced stagnant zones and particle accumulation.

• Filter screens had a negative effect on particle fluidization velocity.

Abstract

Choanoid fluidized bed bioreactors (CFBBs) are newly developed core devices used in bioartificial liver-support systems to detoxify blood plasma of patients with microencapsulated liver cells. Direct numerical simulations (DNS) with a direct-forcing/fictitious domain (DF/FD) method were conducted to study the hydrodynamic performance of a CFBB. The effects of particle–fluid density ratio, particle number, and filter screens preventing particles flowing out of the reactor were investigated. Depending on density ratio, two flow patterns are evident: the circulation mode in which the suspension rises along one sidewall and descends along the other sidewall, and the non-circulation mode in which the whole suspension roughly flows upward. The circulation mode takes place under non-neutral-buoyancy where the particle sedimentation dominates, whereas the non-circulation mode occurs under pure or near-neutral buoyancy with particle–fluid density ratios of unity or near unity. With particle–fluid density ratio of 1.01, the bioartificial liver reactor performs optimally as the significant particle accumulation existing in the non-circulation mode and the large shear forces on particles in the circulation mode are avoided. At higher particle volume fractions, more particles accumulate at the filter screens and a secondary counter circulation to the primary flow is observed at the top of the bed. Modelled as porous media, the filter screens play a negative role on particle fluidization velocities; without screens, particles are fluidized faster because of the higher fluid velocities in the jet center region. This work extends the DF/FD-based DNS to a fluidized bed and accounts for effects from inclined side walls and porous media, providing some hydrodynamics insight that is important for CFBB design and operation optimization.

Introduction

Although significant development in support therapies has been made in recent years, fulminant hepatic failure still leads to a high mortality worldwide. Currently, the only effective long-term treatment is orthotropic liver transplantation. However, the shortage of organ donations makes this treatment difficult for widespread adoption (Fruhauf et al., 2004; Lee, Squires, Nyberg, Doo, & Hoofnagle, 2008; Yu et al., 2014). In this situation, bioartificial liver systems offer an alternative hope. They can serve as an efficient tool for bridge-to-transplantation use or to treat end-stage liver failure after injury. In these support systems, the bioreactor is the core device in which substances such as albumin and other macromolecules are exchanged between the microencapsulated artificial liver cells and the plasma of a patient. Traditionally, the microcapsules were hosted in a fixed or fluidized bed, where the high shear stress of the perfusion fluid and the existence of channel flow could result in microcapsule damage, stagnant zones, and even invalid perfusion. To improve fluidization quality and mass transfer efficiency between microcapsules and fluid, a novel choanoid fluidized bed bioreactor (CFBB) was proposed by Li et al. (Li, Yu, Chen, Zhang, & Du, 2008; Yu et al., 2014). Several biomedical experiments have been conducted to evaluate the functionality of the CFBB (Li, Li, & Yu, 2013; Yu et al., 2014). Nonetheless, the hydrodynamics of the two-phase flow in the CFBB, which plays an important role in reactor design and operation optimization, is poorly understood.

Because useful and detailed information about the flow field can be obtained, computational fluid dynamics (CFD) simulation has emerged as a desirable tool for the characterization of fluidization behavior. Generally, the CFD approaches for multiphase flows are classified as a two-fluid model (TFM), a discrete element method (DEM) in combination with a CFD method for fluid phase (DEM–CFD or so-called unresolved discrete particle method) or a direct numerical simulation (DNS or so-called resolved discrete particle method) (Feng & Musong, 2014). Based on a continuum description of the fluid and solid phases, TFM is suitable for simulations of fluidization systems at both laboratory and industrial scales because the expense of computations is less. However, the details of the phase interactions cannot be adequately modelled through TFM (van der Hoef, Annaland, Deen, & Kuipers, 2008). DEM–CFD and DNS both take a Lagrangian viewpoint to track the particle motion. Specifically, DNS employs finer grids to resolve the flow field around an individual particle and to compute the drag and torque by integrating the pressure and viscous forces on the particle surface. Therefore, DNS enables a more accurate and detailed prediction compared with DEM–CFD, and supplies drag models for the CFD strategies at coarser scales. The first application of DNS to the investigation of particle fluidization was conducted by Glowinski, Pan, Hesla, Joseph, and Periaux (2001) in a pseudo-two-dimensional fluidized bed, employing a distributed Lagrange multiplier/fictitious domain (DLM/FD) method. Pan, Joseph, Bai, Glowinski, and Sarin (2002) observed that the relationship between the simulated particle fluidization velocity and solids fraction followed the Richardson–Zaki power law (Richardson & Zaki, 1997). The DLM/FD method was also used to study the self-fluidization phenomenon in particulate flows, where heat transfer caused by an exothermic catalyst reaction occurs on particle surfaces (Wachs, 2011). Based on a similar computation principle embedded in the DLM/FD method, the immersed boundary method (IBM) has been applied to the DNS of fluidized beds. From a comparison between the particle–fluid interaction forces obtained from IBM and DEM, Kriebitzsch, van der Hoef, and Kuipers (2013) suggested a modification of the computing strategy for the gas–solid force. Feng et al. (Feng & Musong, 2014; Feng, Ponton, Michaelides, & Mao, 2014) studied the slip condition at the wall, and the heat transfer characteristics of 225 fluidized particles in a narrow channel using IBM. Another important DNS method used in the fluidization computation is the lattice-Boltzmann method. These approaches have been applied to simulate travelling waves (Derksen & Sundaresan, 2007), flow structures (Wang, Zhou, Wang, Xiong, & Ge, 2010), stability criterion verification (Wei, Wang, & Li, 2013), as well asscalar dispersion (Derksen, 2014) in two- or three-dimensional fluidized beds. Apart from these conventional DNS methods, an improved version of the macro-scale pseudo-particle modelling method was developed to simulate gas–solid systems containing hundreds to thousands of particles with high particle/fluid density ratios (Ma, Ge, Wang, Wang, & Li, 2006; Ma, Ge, Xiong, Wang, & Li, 2009; Tang et al., 2004). In this method, the finite difference technique is combined with weighted averaging techniques to upgrade particle interactions at the fluid element level (Ge & Li, 2003). Using this method, Xiong et al. (2010) also simulated a two-dimensional fluidized bed at moderate and high solid/gas density ratios. With up to 30,000 particles simulated, particle clusters were observed to form.

Based on the extensive computational works described above, DNS has been validated as a promising tool for hydrodynamics characterization of fluidized beds. However, no DNS or even other CFD methods have been used to evaluate the hydrodynamic performance of CFBBs. In our study, we employed the direct-forcing/fictitious domain (DF/FD) method proposed by Yu and Shao (2007) to conduct a fully resolved simulation of a CFBB under different operating conditions. This method is an improved version of a previous DLM/FD method (Yu, Shao, & Wachs, 2006; Yu & Wachs, 2007; Yu, Wachs, & Peysson, 2006), in which the calculation of the particle velocity and body-force was more computationally intensive compared with the DF/IBM. In the DF/FD method, this shortcoming was overcome using a discrete δ-function in the form of a bi-(tri-)linear function to explicitly transfer quantities between the Eulerian and Lagrangian nodes (Yu & Shao, 2007).

This paper is organized as follows: first, the principle and numerical schemes of the DF/FD method are described briefly. Second, the DF/FD method is verified by comparing the drag and particle settling velocity from simulations with data from the literature. The results and discussion on two- and three-dimensional CFBBs are then presented. Finally, conclusions from the numerical simulations are drawn.