GPU-Accelerated Pore-Scale Transport Resolved Model for Flow-Assisted Battery Design

Redox flow batteries (RFBs) have emerged as grid-scale energy storage technology used in intermittent renewable energy resource applications due to their decoupled power and energy sizing and long life cycles. However, redox flow batteries suffer from low energy and power densities, which may be resolved by improving electrode design. As such, gaining insight into the effect of the electrode microstructure by using computer simulations is paramount in improving RFB performance. Such design optimization would not be possible through experimental procedure due to cost, time, or availability of materials.

Existing volume-averaged models treat the electrode/electrolyte matrix as a homogenous medium of uniform porosity. By contrast, we have developed a pore-scale simulation that distinctly accounts for the separate electrode/electrolyte phases. Our model requires no simplification or assumptions regarding the electrode morphology, and enables us to carry out detailed studies into the effects of the precise electrode microstructure with unprecedented fidelity. A comparison between the two approaches is shown in plots of electrode overpotential in a smooth distribution of a volume averaged model (Fig. 1a) and in a detailed 3D surface distribution of our pore-scale model (Fig. 1b). To simulate our system we use realistic carbon fiber geometry as an input to our simulation. X-ray computed tomography images of carbon fibers (Fig. 2a) are segmented in 2D images and then stacked as a 3D image (Fig. 2b) to create the virtual volume (Fig. 2c) from which we may take a selection of desired porosity. This model may be extended to the treatment of many different flow battery system configurations including all Vanadium, Vanadium/Bromide, and other novel VRB systems.

The primary drawback of a pore-scale approach is the high computational resources required. To overcome this challenge, in lieu of a traditional implementation with Message Passing Interface, which uses many central processing units to divide the computational effort of a simulation, as an alternative we employ the use of Graphics Processing Units (GPUs) which have been shown well suited for solving problems in computational fluid dynamics.

In this work we present an application using GPUs to increase the speedup of our simulations in order to simulate realistic length and time scales not possible using traditional methods (Fig. 3). Computational techniques for efficiently simulating RFB systems are examined along with the effect of domain size and number of processors on scalability and speedup.