Global warming is an urgent issue for our planet thereby many countries has declared to achieve zero emission of carbon dioxide in 2050s. In the next generation industry, electricity produced from renewable energy becomes major energy while heat is a major in the current industry. Therefore, electrochemical process should play a critical role. For example, water electrolysis and seawater electrolysis combined with solar energy or wind energy to produce hydrogen must be developed. However, developing non-precious electrocatalysts with high catalytic activity is still a big challenge in material science.Among various electrochemical processes, seawater electrolysis is recently paid much attention as an attractive process for cost effective hydrogen production. In sea water electrolysis, oxygen, chlorine, or hypochlorous acid (HClO) is evolved at anodic side as well as hydrogen in cathodic reaction. In this reaction, HClO is usually regarded as undesired byproduct in mass production. On the other hand, HClO can be high-value-added oxidative material than oxygen, contributing to a cost-effective system, especially in the case of rather small distributed system. Therefore, it is important to selectively generate both oxidative products.Recently, many papers investigated the potential research scheme which is supported by machine learning and/or robots. Here, the authors developed the high-throughput automatic robot for electrochemistry (namely HARfE) as shown in Fig. 1. High-throughput screening of 4-element system was performed for salted water (NaCl solution) electrolysis that is a model reaction for sea water electrolysis.This robot is able to perform fully automatically from preparing samples to measuring electrochemistry. Mainly, this robot is consisted of two parts; sample preparation part, and sample measuring part. HARfE can prepare 88 separated samples in a single run and subsequently measure electrochemistry. One notable advantage of this robot is that the robot can perform almost the same procedure as human researchers do using almost the same sizes of facilities as human researchers use, intending that this robot system can replace human labor force. The size of electrodes (≈0.5 cm2), an electrochemical cell (21 mL of solution), and measurement conditions are very similar to those of human researchers. In addition, a relatively long-term measurement such as stability test which requires a certain amount of solution is also able to be done. Hence, the robot can cover a wide range of electrochemical measurements.The typical procedures are the following. First, the pipetting arm dispenses and mixes up to ten kinds of solution that include metal ion. Second, the robot drops the mixed solution onto the substrates such as F-doped tin oxide (FTO) with a size of 5 mm by 30 mm and dry them on a hotplate. Third, the transfer arm transports the sample holders to an electrical furnace to calcine. Then, the picking-up arm transfers each obtained electrode to a H-type cell to perform electrochemical measurements. Finally, the pipetting arm dispenses a small amount of solution from the cell, mixes with color reagent, subsequently measures absorption to quantitate the products. After each measurement, solution was changed automatically from a tank behind the robot.In this study, the composition effect of Co-Mn-Fe-Ni-Cr electrocatalysts was investigated. Current density at 2.0 V vs. Ag|AgCl reference electrode, Faraday efficiency for hypochlorous acid/oxygen when 1 mA of current flew for 200 s (0.2 C), and the ratio of current density at 2.0 V vs. Ag|AgCl after 1000 s and 100 s were calculated. It was revealed that the obtained color maps clearly showed the composition dependence, and that The optimal composition varied depending on each evaluation criterion. For example, Co-rich composition provides high current densities and Mn-rich electrodes produced little HClO. It is notable that NiO x exhibited poor stability while stability was drastically improved by combining with FeO x .In addition, we demonstrated the composition optimization using Bayesian optimization (BO) algorithm. In this case, we used the dataset of Co-Mn-Fe-Ni four elements system with 286 data points. Composition is explanatory variables and current density at 2.0 V vs. Ag|AgCl is the target variable. Usually, BO-assisted experiment was performed one sample by one sample. Contrary, HARfE can deal with multiple samples in a single run. Therefore, multi-sample Gaussian process regression was performed to effectively search superior compositions. The detailed results will be shown in the presentation. Figure 1
Read full abstract