ENGINEERING AND CFD

The electro-dialysis process is an efficient desalination technique that uses ion exchange membranes to produce clean water from seawater. This process involves different physical phenomena like fluid dynamics, electrodynamics and diffusive mass-transport along with their interactions. Rigorous assessment of those interactions, especially those near the membranes, are only possible through large-scale coupled simulations. The aim of the APAM compute time project is the detailed flow simulation in this application to understand the effect of different spacer structures between the membranes in this process.

Description of the results:

The electro-dialysis is done in a stack that consists of selective ion exchange membranes. The cation exchange mebranes (CEM) only allow cations to pass, while the anion exchange membranes (AEM) only let anions pass. CEM and AEM are arranged alternatingly resulting in alternating dilute and concentrate flow channels. These membranes are seperated by a complex spacer structure. The spacer design affects the flow properties and ion transport through the membranes.

Ion seperation near the membranes creates an electrical double layer (EDL); a non-electroneutral region in the range of nanometers. This layer can cause concentration polarization, which leads to membrane failure. Thus, to simulate the ion transport in the flow channels, the EDL must be resolved, resulting in a huge number of tiny elements and a need for large computational resources. Therefore, the supercomputer Hazel Hen at HLRS in Stuttgart was used to perform those numerical simulations to investigate the pressure drop and ion transport in the spacer-filled flow channels.

As the simulation of an entire stack of 50 - 100 channels is too expensive, a single unit consisting of 2 channels with 3 membranes and minimal extensions in length and width, depending on the spacer geometry, have been simulated. Two spacer designs, a woven and a non-woven, were investigated, each with varying angles between the filaments. The parameters of the spacer are illustrated in Fig.1.

Since the efficiency of the process depends on the mass inflow rate, the inlet mean velocity was varied from 0.01 m/s to 0.8 m/s. The simulations were performed using the in-house lattice Boltzmann solver Musubi based on an octree data structure. The meshes for the simulation were generated with the octree mesh generator Seeder.

Hydrodynamic simulations revealved that the non-woven spacer with α=60° reached steady state for all velocities. An increase to α=120° turns the flow into an unsteady periodic flow for an inflow velocity of 0.2 m/s and into a turbulent flow for 0.4 m/s resulting in a high pressure drop over the channel. For woven spacers with β = 45°, the flow reached steady state for all velocities, but with β = 90° the flow becomes unsteady and periodic for inflow velocities with at least 0.6 m/s. Thus, increasing α or β increase the pressure drop. Fig.4 shows the pressure drop over the channel length (20 cm) for different spacer configurations over inflow velocities.

For the multi-component flow simulations, the flow channels are coupled with membranes via boundary conditions. The electrical force that drives the ions towards the membrane is included in the form of source terms. Here, three components were considered: H20, Na+ and Cl-. With three interacting species, the computation cost also roughly increased threefold in comparison to the single-component simulations. In Fig.5 it can be seen that the concentration of ions increases near filaments where the convective transport velocity is smaller. Simulations of different spacer designs revealed that the concentration of ions are most evenly distributed for smaller angles, which also reduce scaling and fouling effects in the channel.

Finally, a full unit setting, consisting of 2 channels and 3 membranes, was simulated. Figure 6 shows the concentration of ions in the dilute and concentrate channel, which forms a single unit in the electrodialysis module. Figure 7 shows the concentration profiles of two channels with a woven spacer.

NOTE: This project was made possible by PRACE (Partnership for Advanced Computing in Europe) allocating a computing time grant on GCS HPC system Hazel Hen of HLRS.

Research Team Members: 
M.Sc Kannan Masilamani, Dipl.-Ing. Harald Klimach

Scientific Contact:
Kannan Masilamani
Institut für Simulationstechnik & Wissenschaftliches Rechnen, Universität Siegen 
Hölderlinstr. 3, D-57076 Siegen/Germany
e-mail: kannan.masilamani [at] uni-siegen.de
www.mb.uni-siegen.de/sts/

October 2016

HLRS project ID: PP14112730