Newsletter 07/21

The basic principle of redox flow batteries (RFBs) is to separate two compounds that would otherwise engage in a reaction, so that they remain passive and stable until we allow an electric current to flow. Then the reaction proceeds and electrons are swapped between the compounds, releasing or storing energy according to demand.

Besides the challenges of running a large and complex technical system such as a large-scale RFB (e.g. pumping, heat management, etc.), the choice of the electroactive material is pivotal. Obviously, it should have a high energy and power density. However, to ensure competitive and sustainable energy storage, the material also needs to be widely available, low cost, environmentally safe, and stable over many cycles of charging and discharging.

So far, the materials in use fail us in at least one of these criteria, so the search for new compounds is underway. Organic chemicals represent a huge search space, and given the existence of demonstrators and/or natural redox compounds, we can be confident that plenty of suitable materials exist.

The project SONAR aims to promote the search for new materials by providing a range of computational tools and techniques needed to investigate every aspect of a novel RFB system virtually, i.e. before it enters the stage of costly lab experiments.

Quantum mechanical calculations give us the theoretical gain in energy for a given redox-reaction, but in practice, we experience losses that reduce the benefit. These are due to additional effects, which we can capture by extending the simulation’s scope step by step, starting from the redox compound and then including the other components of a RFB such as the solvent, processes at the electrode and the membrane. We proceed even further up to the flow field, the organization of the stack and the integration of the RFB within a local micro grid. The final goal is to enable high-volume preselection of potential new organic active materials for redox flow batteries.

With this newsletter we report on the most important results achieved so far in the areas of high-volume preselection, atomistic and meso-scale modelling, and cell-level modelling and simulation.

In this part of the project we are linking the individual tools and techniques developed to create a seamless sequence, starting from the left and working up to the right. At present we have reached roughly the middle. 

Let’s begin with any given supplier database (e.g. PubChem or inhouse data): we have implemented a flexible and customizable filtering sequence to pick compounds based on size, composition, synthetic accessibility and/or a given chemical class, if you prefer to focus on a specific chemistry. This stage (1A) works on topological strings (SMILES) only and can process millions of molecules within a few minutes.

For the resulting selection, we then generate their reduced or oxidized counterparts (1B) as a starting point for calculating formal reduction potentials E°', the primary quantity decisive for energy density.

One group of project partners are developing elaborate QM procedures to compute the most accurate values of E°' (2A), while another is making use of already existing and continuously emerging, computed data: 

We have trained our first surrogate model (ML1.0) on about 45000 published samples to predict in an instant E°' for redox couples that exchange two electrons and two protons in one go; recently a manuscript about this has been submitted for publication in ‘Batteries and Supercaps’ (accepted). In general, the usability of such a surrogate model hinges on the data presented during the training process, so we hope to increase our scope and improve the accuracy along with the ever-growing amount of own or elsewhere published QM data.

Our immediate next step will be to train a complementary, closely related model on one-electron transfer reactions (ML1.1) too, in order to cover this class of redox-active materials as well and complete this stage (2B).

The second factor likely to limit energy density is how much of the least soluble species can be dissolved in the chosen solvent. That is: how much of the electroactive material remains mobile and can fulfil its task of acting as a source of / as a sink for electrons, when passing the electrode’s surface – instead of blocking the tubes or just sitting in the tank.

As before, the achievements in SONAR – elaborate QM procedures and calculation results – will lead (3A) and trained models (3B) will follow to distinguish between the desired soluble redox couples and the useless insoluble ones; the computations are under development.

SONAR partners have also developed theoretical models compatible to the down-stream and up-stream stages, which will be added to the sequence as soon as possible.

To complete the tour through the sequence we will outline the remaining steps:

The results obtained so far – that is the standard reduction potential E°' and the limiting compound’s maximum concentration – will be passed as input to the next level: the investigation of processes at the electrode surface by kinetic Monte Carlo simulations. This step (4) will complete the characterization of half-cells and yield the operation characteristics (voltage in response to current density), which will be input for the next steps: at this stage we will switch gears and assemble full cells by matching up the best performing half cells (5). Thanks to the efficient, yet still physics-based implementation of the 0D-model this final step (6) will be fast enough to evaluate all the resulting full cells with respect to their cell voltage and power density as a function of the applied current and the state of charge of the battery. We will consider the best performing systems worthwhile for further examination.  

The role of the atomistic modelling in the SONAR project is essentially calculating properties of organic redox molecules at the atomistic level from first principles, i.e. using as input only the molecular structure and no experimental data. Quantum chemistry calculations can be performed on known and unknown molecules. The results will be used to train the machine learning models described above, which will be able to predict properties at a much lower computational cost. Another task in this area of work is calculating kinetic parameters that help to describe the speed of the different processes happening in the electrochemical cell of the redox flow battery. In practice, these tasks are mostly performed with widely used electronic structure methods based on density functional theory (DFT).

The first achievement in this area so far is the establishment of a standard procedure to calculate the redox potentials of molecules in water as a function of the pH.  The reduction reaction is the addition of one or more electrons (coming from the electrode), which may be coupled to the addition of one or more protons (coming from the water). The redox potential depends on the free energy change of the reduction reaction, i.e. the difference between the energies of the oxidized and reduced forms of the molecule. In practice, these energies are obtained by optimizing the geometry of each redox form. The effect of the aqueous solvent is modeled with an implicit solvation model, where the molecule is surrounded by a continuous dielectric medium (without explicitly modeling water molecules). At room temperature, the molecules’ vibrations also contribute to their free energy. This thermal contribution is estimated by performing a separate calculation of the vibrational frequencies in gas phase.

This procedure has been validated against a set of experimental redox potentials of molecules belonging to the quinone family. As shown in Figure 1, the calculated potentials at pH = 0 are in good agreement with the experimental ones. We have also developed and refined a procedure to transform the pH = 0 potentials to higher pH values.

However, there are some outliers for which the prediction is far from the experimental reference. The cause of this error has been narrowed down to a failure of the implicit solvation model. To verify this, we are in the process of recalculating the potential of one of the outliers with a mixed explicit-implicit solvation model. As shown in Figure 2, the molecule is surrounded by up to 18 water molecules modelled at the same level of theory, and this solute-solvent cluster is embedded in the dielectric continuum. This method improves the agreement with the experimental data, but a better sampling of the solvent configurations is needed to understand if explicit solvation can recover the experimental redox potential.  

CNRS, France, has developed a novel type of kinetic Monte Carlo (kMC) model targeting the organic redox flow battery’s electrode interface simulation.

Considerable obstacles exist to screening the candidate redox couples in the organic redox flow battery field. It’s challenging to interpret and predict the batteries’ performances due to the variety of organic compounds, unknown interactions, and side reactions.

Different types of physical events could occur at the electrode interface. Taking the methyl viologen (MV) anode as an example, the dominant events are Brownian motion and electrochemistry events. When the active materials interact with the electrode surface, adsorption and desorption could take place. Furthermore, the degradation of the active material should not be ignored (in the case of methyl viologen, dimerization). The frequency and the driving force for these actions are different, and the process is more or less stochastic. Ideally, if we could trace all the events, the general performance of methyl viologen as anode material can be obtained.

Following the same idea, the research team of CNRS (France), led by Prof. Alejandro A. Franco, developed a kinetic Monte Carlo (kMC) model. At each simulation step, the kMC algorithm selects and executes an event based on the input event rate.

The kMC modeling scale is several nanometers, where the interface between the electrolyte and the electrode is under non-negligible electrical field impacts. The kMC model is therefore combined with an Electrical Double Layers (EDL) approach to consider the impact of the electrical field.

The model has been tested under the galvanostatic discharging process with different input current densities and electrolyte concentrations. Along the simulation time, the model reproduces the system’s electrochemistry response, where the potential evolution and the formation of EDL have been observed via the model. The EDL structure reveals that the positively charged ions oscillate with negatively charged ions inside the Debye length.

In general, the model is able to explicitly show how the species formulate the double layer at the vicinity of the electrode surface. The electrode potential can be obtained directly from the model which holds more credibility than empirical Butler-Volmer-like equations which involve a lot of assumptions. The parameters acquired by the model can be further used in other larger scale models.

Jia Yu - a PhD student from CNRS - presented this model in the 239th ECS meeting, which was held online from 29th May to 3rd June this year.  

Electrochemical cells are one of the fundamental building blocks of flow batteries. Thus, to understand, study and optimize the interconnected physicochemical processes within flow battery systems, accurate and efficient models for the simulation of electrochemical cells are required. The research group Electrochemical Cells and Microstructures at the Institute for Computational Physics at ZHAW in Switzerland focuses on the development of macroscopic models at the cell scale, which allow for the prediction of important metrics, such as cell voltage, power density or crucial degradation mechanisms, such as capacity fade.

The 0-D U-I-SoC model expresses the total cell voltage and power density explicitly in terms of the battery state of charge (SoC) and the electric current density, which enables efficient predictions of the cell performance of organic redox flow batteries. The model takes into account important processes, such as the overpotentials resulting from the electron transfer at the electrode and concentration overpotentials, due to differences in species concentrations between the electrode surface and the bulk of the electrolyte. Additionally, the model considers the electro-osmotic drag through the membrane, which leads to changes in the electrolyte volumes in the half-cells as a function of the state of charge. Thanks to the reduced model dimensionality, the model can be evaluated in a fraction of a second, allowing for real-time applications and fast parameter studies.

Fig. 6 shows preliminary model predictions of the cell voltage and power density in terms of the state of charge and current density.

In collaboration with JenaBatteries GmbH, simulation results of the 0-D U-I-SoC model have been compared against experimental charge/discharge and polarization experiments performed with a small test cell using the TEMPTMA/MV electrolyte system. Preliminary comparisons show good agreement between the model and the experimental results within the range of the model validity. A first version of the 0-D U-I-SoC model, which is under active development, has been published as open-source software and can be obtained from the GitHub repository.  

Several of the FLORES projects presented their results at the recent online conference " Next Generation - Flow Battery Conference and Networking Event", which was hosted by the FlowCamp project on 11th March 2021, attended by 250 participants. A joint workshop is also planned at the upcoming International Flow Battery Forum (IFBF) conference in July 2021.

More information about the group’s activities can be found on the LinkedIn page.