Accelerating the Development of High-Performing Dynamic Electrochemical Processes via Bayesian Optimization

The implementation of electrochemical processes to produce fuels, commodity chemicals, and intermediates can allow for the direct integration of renewable energy sources into in major chemical manufacturing processes, and thus help eliminate more than 20% of industrial greenhouse gas emissions. Electrochemical processes provide unique opportunities to increase energy efficiency and enable previously unexplored chemical transformations, but their practical implementation will require the operation of reactors at high selectivity and throughput. Controlling selectivity in electrochemical reactions can be achieved by balancing the mass transport and reaction rates of reactants and intermediates at the electrode/electrolyte interface. This balance can be realized by imposing dynamic potential pulses with tunable magnitudes and timescales to match the requirements of desired reaction pathways. Predicting optimal pulse sequences is challenging because it requires an accurate knowledge of transport properties, reaction mechanisms, and electrochemical rate laws. Alternatively, empirical searches may be used to discover optimal sequences, but the large number of parameters (i.e., multiple pulse times and potential magnitudes) results in an experimentally inaccessible combinatorial design space.

In this presentation, we will discuss how Bayesian Optimization (BO) approaches can be implemented to efficiently explore the design space of pulsed electrochemical processes and lead to enhancement in performance. Electrooxidation of Cerium(III) to Ce(IV) electrolytes will be discussed as a model electrochemical reaction. This reaction is relevant to redox flow battery and redox-mediated water electrolysis, and due to its high oxidation potential, it competes with parasitic oxygen evolution reaction (OER) which lowers its Faradaic Efficiency (FE). We will show how dynamic potential dosing can help balance Ce(III) transport and electrooxidation rates, allowing finer control over its concentration near the electrode and ultimately enhancing FE. Due to the large number of possible pulse sequences, continuum transport and reaction models were built to identify the appropriate window of operating conditions, and then BO was used to rapidly identify the optimal potential pulses. After 25 experiments, a maximum FE = 0.91 was achieved (active pulse time = 5ms, resting pulse time = 136ms, and active pulse voltage of 2.5 V vs. Ag/AgCl) compared to FE = 0.55 at constant potential operation. Furthermore, we will present a multi-objective BO method to identify the pareto conditions that lead to the optimal trade-off between FE and Ce(III) electrooxidation rate – two relevant performance metrics for practical implementation. Similar BO approaches implemented for dynamic CO2 electroreduction and organic electrosynthesis will also be discussed.