Parameter Optimization in the Drying Process of Catalyst Ink for PEFC Electrode Films with Few Cracks

Abstract

The fabrication of polymer electrolyte fuel cell (PEFC) electrodes is complicated, and consists of four major steps: dispersion, mixing, coating, and drying. Each step is realized by a combination of various phenomena; hence, it is highly challenging to investigate the entire process in detail. The phenomena occurring during drying are particularly complex, and are primarily responsible for the difficulty with control of the PEFC electrode fabrication process [1]. Such complex factors affect the component distribution and structure of the electrode, and consequently, the performance of the fabricated PEFCs. Optimal drying conditions need to be determined for the stable fabrication of high-quality electrodes; however, it is not practical to simulate all complicated phenomena during the process. This leads to reliance on empirical rules put forth by researchers and engineers.Similar problems have long been encountered in the field of materials. In recent years, the application of machine learning, particularly active learning methods such as Bayesian optimization, has achieved some success [2]. In this study, Bayesian optimization was applied to the drying process of PEFC catalyst ink to determine optimum drying conditions in a few trials. A schematic of the workflow for parameter optimization using the Bayesian method is shown in Fig. 1.The ink was fabricated by dispersing carbon particles (Vulcan XC-72R) and Nafion in water and 2-propanol (IPA). Subsequently, 30 μL of the ink was dropped onto a glass slide placed on a rubber heater. The temperature history to be set for the heater was the parameter to be optimized, and drying was performed with the temperature history indicated by the machine learning system. The heater temperature was controlled for 10 min after the drop, and the temperature was specified in eight steps ranging from 30 °C to 100 °C every 2 min. Thus, there were a total of 85 = 32,768 parameter sets to be explored. The captured images of the dried droplets were binarized to detect cracked areas, and the ratio of the cracked areas to the droplet area was defined as the “crack ratio” and used for evaluation.The typical temperature histories and appearances of the samples with high and low crack ratios are shown in Fig. 2. The temperature history that resulted in a low crack ratio was confirmed by active search to be a globally optimal parameter set. Fig. 2 shows that the samples with high crack ratios exhibit distinctive radial cracks and low-thickness-film regions that spread to the periphery, resulting in a high crack ratio. In the samples with a lower crack ratio, there was no region with a low-thickness film. Examining the temperature history revealed that the crack ratio tended to be lower when the temperature was kept low for the first 2 min, then increased, and subsequently decreased again after another 2 min. We proposed the mechanism shown in Fig. 3 to explain why this temperature history resulted in a low crack ratio. Rapid heating immediately after the drop caused non-uniformity in the film thickness because of the flow of solute particles, but refraining from heating during this period may have allowed the concentration to increase gradually and suppress the free flow of particles. If the concentration was sufficiently high, acceleration of heating had little effect on the film thickness, but if heating continued even at the end of drying, the position of the particles may have been rapidly fixed, and cracks caused by stress in the film may have occurred.Thus, using Bayesian optimization, a highly efficient search was achieved for the drying process parameters of the PEFC catalyst ink, which were previously difficult to adjust. Moreover, the process window discovered by the search was limited, and could not have been found by repeating conventional hypothesis-driven experiments. Using the Bayesian method, we obtained the optimal parameters for a certain condition and examined the mechanism of drying using the information obtained in the process window. There can be further investigations of the drying process to acquire more knowledge about the drying phenomenon; the scope of future studies can also be expanded to include other processes in PEFC electrode film formation or a wider range of parameters.This work was supported by JST-Mirai Program Grant Number JP19216242, Japan. References T. Kusano et al. Structural evolution of a catalyst in for fuel cells during the drying process investigated by CV-SANS. Polym. J. 47, 546-555 (2015).A. G. Kusne et al. On-the-fly closed-loop materials discovery via Bayesian active learning. Nat. Commun. 11, 5966 (2020). Figure 1