On August 12, 2020, Venkat Viswanathan, Faculty Fellow at the Wilson E. Scott Institute for Energy Innovation and Associate Professor of Mechanical Engineering at Carnegie Mellon University, presented his talk on “Autonomous Discovery of Battery Electrolytes with Robotic Experimentation and Machine Learning” via a live webinar presentation.
Dr. Viswanathan’s talk covered the importance of robotic experimentation, machine learning guided design of experiments, and the frontier of remote experimentation.
View Dr. Viswanathan’s webinar presentation, here.
* Note: You must fill out the registration form to view the webinar *
Following the talk, attendees were given the opportunity to ask Dr. Viswanathan questions in a Q&A session, available below.
Q&A with Dr. Venkat Viswanathan
Q: Isn’t molecular dynamics simulations cheaper and hassle-free than the actual experiment, and what might be the advantages of experiments over MD simulations?
A: Fluids, particularly the fluid electrolytes commonly used in batteries, are difficult to model from first principles. Even qualitatively accurate MD calculations of conductivity in highly concentrated, mixed salt/solvent electrolytes would require much expertise and computational time. A focus on experimental data enables the quantitatively accurate data needed to run closed-loop optimization for such systems.
Q: Hi, are Dragonfly and Otto open to everyone use or they are limited to Carnegie Mellon? How we can get access to it?
A: While access to the test-stand is currently limited, Dragonfly is open-source code available here: https://github.com/dragonfly/dragonfly/
Q: Can this automation technology be used for high viscosity composites, such as polymer/filler composites? Is it possible to apply artificial intelligence or machine learning here to speed up development?
A: Certainly can! I believe polymer synthesis parameters have been optimized via Bayesian methods in previous literature: https://doi.org/10.1038/s41598-017-05723-0
Q: Can you please comment on the importance of initial guess? Is it human-decided? Thank you
A: Initial guesses in each optimization took the form of 5 random guesses. In principle, prior information can be incorporated into the optimization for use in initial sampling.
Q: Could you also search for the best or optimum ELECTRODES for a given aqueous chosen electrolyte?
A: Test geometry for solid materials would look significantly different than the current system. Issues of not altering the electrode interface given repeated testing would need to be dealt with.
Q: You just showed the study on HER and OER for aqueous system and non-aqueous system. Did you consider the gas transport mechanism during your simulation?
A: We did not. Electrodes used were small (.045 cm^2 surface area), and subjected to rapidly flowing electrolyte. Thus, we assumed bubbles generated were sheared rapidly by the fluid, and gas generation and ingress/egress were neglected in reporting our results.
Q: The methodology looks to be more easily applied to liquid electrolytes. Can it be used to solid electrolytes also?
A: Test geometry for solid materials would look significantly different than the current system. Issues of not altering the electrode interface given repeated testing would need to be dealt with. Solid electrolytes also introduce complications of synthesis.
Q: Can this methodology be used to flow battery with polyoxometalates?
A: Yes. I think the test-stand as is today could be used to generate much useful data for flow-battery catholyte/anolyte optimization.
Q: What are the ion transference numbers and ionic conductivities generally recorded in the setup in WISE electrolytes? What about the viscosity issues?
A: Transference numbers and conductivities generally recorded were in the 0.4 to 0.5 range, and 50-200 mS/cm range. See our previous publication for further data: https://doi.org/10.1149/2.0142001JES . Viscosity can be an issue for the highest concentrations of commonly used WiSE electrolytes (like LiTFSI) but was not especially relevant for our results. Our programmable pumps dispense accurate volumes even in high viscosity regimes.
Q: The system considers the bulk conductivity of the liquid electrolyte. However, in Li-ion batteries, there is no presence of the electrolyte as bulk. Indeed, it is always contained within the porous structures of separator and electrode. Have you considered this aspect in the development of your model?
A: Conductivity as measured still plays a key role in battery performance in fast-charging and low-temperature regimes, as well as for alternative geometries like large-format grid storage batteries. As work develops, additional measurements of transport properties in the bulk and at the interface will be included.
Q: Why water don’t decompose at 1.23V in high salt concentration electrolyte?
A: Previous theoretic work explaining this phenomena either focus on a lowered water concentration affecting HER/OER thermodynamics and kinetics, or an insoluble passivation layer depositing on the electrode, enabled by the previous effect.
Q: Which physical property could have caused an increase in the electrochemical stability window with a decrease in salt-concentration? Any thoughts?
A: We have considered the effect that the donicity of the anion may play a role in nonlinearly affecting water activity (outside of water concentration). Such an effect would be more noticeable at high currents, which our data also show.
Q: Is the electrode material of the automated measurements the same as the relevant battery electrode material? If not, what approaches have you identified to help make the automated measurements translatable to a specific battery electrode material?
A: While platinum is an inert electrode, previous work on WiSE have used inert electrodes and electrochemical testing to assess the stability window of a given WiSE. As we move to nonaqueous, new test geometries are required to automate the testing of electrodes on active electrodes.
Sponsor
A special thank you to our sponsor, the Royal Society of Chemistry.
Sponsorship opportunities
ECS is hosting bi-monthly webinars from through December 2020. Each ECS Webinar has four sponsorship opportunities.
Contact edward.jost@ioppublishing.org to reserve your sponsorship space today!
