Your Feedback

Energy News Monitoring

AI to improve battery components

John Hill for Argonne National Lab CC BY-SA 2.0

Redox flow batteries are electrochemical energy storage devices which can transform chemical energy into electrical energy through reversible oxidation and reduction of working fluids. Due to the flexibility in its system design, redox flow batteries are promising media in stationary storage of energy from sources such as solar and wind. They can also provide high functionality at low cost, with little harm to the environment. To improve the quality of batteries scientists usually perform reliable molecular simulations using supercomputers. Supercomputers can accelerate the desired materials screening process, while yielding useful information about the possibilities that come with different chemistries. However, even high-throughput simulations run on these supercomputers can only analyse a small part of the possible viable chemistries that exist for certain types of batteries.

Now (2020), scientists at the Argonne National Laboratory have gone one step further in the hunt for the best battery components by using artificial intelligence. The study team researched the mechanisms of redox flow batteries, where chemical energy is stored in dissolved molecules interacting with electrodes. They created the concept of storing and releasing energy with materials called ​redox active polymers, or redoxmers, which were based on larger molecules with tens of charge storing elements. They noticed that during charging and discharging, batteries tend to form an inactive film. To prevent this, the scientists tried to create a redoxmer that could be electrically split at a particular voltage so that it could re-enter the electrolyte solution. To find a redoxmer that would split at the appropriate voltage the supercomputer at the Laboratory Computing Resource Center was used. First, the researchers analysed a set of 1,400 different redoxmers using density functional theory (DFT) calculations, which are highly accurate but expensive. However, these 1,400 redoxmers represented only a tiny spectrum of the total chemical space that the researchers were interested in. To identify the ideal molecules from a larger dataset consisting of more than 100,000 redoxmers without doing DFT calculations, the researchers used a machine learning technique called active learning. This larger dataset comprised redoxmers that were structurally similar to those in the original DFT dataset of 1,400 molecules. To identify 30 molecules with the ideal properties from an initial dataset of 1,400, took them 70 picks. As with random picking only 9 percent of picks would have been successful, the scientists’ approach was a major improvement to existing screening methods.

Scientists have been researching redox flow batteries for many years. In 2011, a new redox flow battery using Fe2+/Fe3+ and V2+/V3+ redox couples in a chloride-supporting electrolyte was designed for potential stationary energy storage applications. The Fe/V redox flow cell using mixed reactant solutions operated within a voltage window of 0.5–1.35 V with a nearly 100% utilization ratio and demonstrated stable cycling with energy efficiencies around 80% at room temperature. Stable performance was achieved at temperatures between 0 °C and 50 °C. The improved stability and electrochemical activity over a broader temperature range (such as Fe/Cr redox chemistry) eliminated the need for external heat management and the use of catalysts.

In 2014, a Mn3+/Mn2+ redox couple with high reaction potential was successfully introduced into a redox flow battery, and the hybrid V/Mn flow cell with multiple redox couples based on Mn(III)/Mn(II) vs. V(III)/V(II) and V(V)/V(IV) vs. V(III)/V(II) was investigated. The results showed that the concentration of sulfuric acid of 4 M was suitable for the viscosity, electrical conductivity, and electrochemical performance. Also, a redox flow battery employing V/Mn and V as positive and negative active species was designed and the charge-discharge performance evaluated. The hybrid V/Mn flow cell showed good capacity retention and energy efficiency could reach up to 80% over 40 cycles. The energy density of 17.85 W h l−1was achieved for V/Mn hybrid cell, 25.3% higher than that of all vanadium cell.

In 2015, a redox flow lithium battery (RFLB) was designed which drastically enhanced the energy density of flow batteries. With LiFePO4 and TiO2 as the cathodic and anodic Li storage materials, respectively, the tank energy density of RFLB could reach ~500 watt-hours per litre (50% porosity), which was 10 times higher than that of a vanadium redox flow battery. The cell showed good electrochemical performance under a prolonged cycling test.

The new approach using artificial intelligence demonstrated not only significant efficiency improvement over the random selection approach but also robust capability in identifying desired candidates in an untested set of 112 000 homobenzylic ether molecules. The findings underline the efficiency of quantum chemistry-informed active learning to accelerate the discovery of materials with desired properties from myriad molecule combinations.

The model guaranteed that by adding this new data point to the training set it would become better, so that afterwards it could be trained again. These new redoxmer flow batteries could have the potential to transform the perception and use of flow batteries for the grid.