Keitaro Sodeyama

Battery research is a rapidly growing and highly interdisciplinary field, making it increasingly difficult to track relevant expertise and explore potential collaborators in similar domains across institutional boundaries. In this work, we present a pipeline for constructing an author-centric knowledge graph of battery research built on OpenAlex, a large-scale open bibliographic catalog. For each author, we derive a weighted research descriptors vector that combines coarse-grained OpenAlex concepts with fine-grained keyphrases extracted from titles and abstracts using KeyBERT with ChatGPT (gpt-3.5-turbo) as the backend model, selected after evaluating multiple alternatives. Vector components are weighted by research descriptor origin, authorship position, and temporal recency. The framework is applied to a corpus of 189,581 battery-related works. The resulting vectors support author-author similarity computation, community detection, and exploratory search through a browser-based interface. The knowledge graph is then serialized in RDF and linked to Wikidata identifiers, making it interoperable with external linked open data sources and extensible beyond the battery domain. Unlike prior author-centric analyses confined to institutional repositories, our approach operates at a cross-institutional scale and grounds similarity in domain semantics rather than citation or co-authorship structure alone.

The search for efficient fuel cell additives is a critical challenge for improving their oxygen reduction reaction activity. Here, we present the first successful application of active learning (AL) to guide the experimental discovery of organic molecules that enhance the mass activity of Pt cathode catalysts in polymer electrolyte fuel cells (PEFCs), a class of fuel cells often referred to in the literature as proton-exchange membrane fuel cells (PEMFCs). Using our in-house SMILES-X molecular characterization tool, we trained neural networks, autonomously tailored for our data, on an initial set of 96 organic compounds, of which 84 passed RDKit validity checks for modelling. Model performance improved over three AL cycles, with the mean absolute error decreasing from 0.202 to 0.165 and the root mean square error decreasing from 0.271 to 0.218. The cycles were guided by predictive ranking, experimental feasibility constraints, and LSTM-based molecule generation. Among the additives identified, melam, a melamine dimer, achieved a 67% improvement in mass activity over the no-additive baseline. These results demonstrate the feasibility of AL in low-data regimes for PEFC additive discovery and highlight the importance of data consistency, model retraining, and close ML–experiment coordination.

Access full-text academic articles: J-STAGE is an online platform for Japanese academic journals.

Introduction An idea of multi-elemental alloying, mainly high-entropy alloying, has gained attention to boost the performance for various electrocatalysis. [1] Indeed, our previous study for single-crystal surfaces of Pt and underlaid equiatomic Cr-Mn-Fe-Co-Ni (known as Cantor alloy [2]) model catalysts demonstrates that enhanced oxygen reduction reaction (ORR) activity at a pristine state and improved electrochemical structural stability even through the potential cycle (PC)-loading. [3] Further improvements of the ORR properties and clarifications of the ORR mechanisms of Pt-“multi-element alloys” surfaces would necessitate the experimental investigation and optimization of a huge number of combinations of synthesis conditions, such as constituent elements, compositions (not only with equiatomic ratios but also extended to compositionally complex alloys (CCAs) with non-equiatomic ratios), and thermal annealing (sintering) temperatures. Therefore, in this study, we used a machine learning approach to mathematically optimize the relationship between the synthesis conditions of the single crystal model catalysts and ORR properties. We explored the alloying constituent elements and compositions as well as the forementioned synthesis conditions of Pt－Cr-Mn-Fe-Co-Ni CCA(111) and/or Pt－Mn-Fe-Co-Ni CCA(111) that outperform the ORR properties of Pt－Cantor alloy(111). Experimental Figure 1 illustrates the procedure of this study. To optimize the synthesis conditions for Pt－CCA(111) surfaces with superior ORR properties, we employed Bayesian optimization (BO), that is a machine learning approach suitable for exploring high-dimensional parameter spaces efficiently. In this study, a custom-defined ORR performance index (OPI) is defined to incorporate both initial ORR activity and durability under applying potential cycle loading and the BO was implemented iteratively proposing synthesis conditions, including the annealing temperatures and elemental compositions of the CCAs, to maximize the OPI. At each iteration, model catalyst surfaces were fabricated with the BO-proposed synthesis conditions and evaluated ORR properties to update the BO model. Eventually, 18 proposed sets of synthesis conditions for up to 6 batches were evaluated. Pt/CCA/Pt(111) were synthesized via arc-plasma deposition (APD) method in an ultra-high vacuum chamber [3]. CCA lattice stacking layers (c.a. 3 nm) with charged Cr-Mn-Fe-Co-Ni or Mn-Fe-Co-Ni compositions were formed by alternately depositing several custom alloy targets on a surface-cleaned Pt(111), followed by Pt surface layers (c.a. 1.2 nm) and precise thermal annealing temperatures. The composition ratios of Cr, Mn, Fe, Co, and Ni are set as integer ratios a : b : c : d : e (the sum of a ~ e is 10), and the annealing temperature T _C is set in the range of 473~873 K in 50 K increments. The APD-synthesized Pt/CCA/Pt(111) are designated as Cr _a Mn _b Fe _c Co _d Ni _e _ T _C K. Electrochemical measurements, including cyclic voltammetry, CO stripping voltammetry, and Linear sweep voltammetry, were performed using a rotating disk electrode setup in N ₂ -purged or O₂-saturated 0.1 M HClO₄. Surface stability, directly related to ORR durability, was evaluated through 0.6 _(3s) -1.0 _(3s) V vs. RHE square-wave potential cycling (up to 5,000 PCs) at room temperature, with activity trend monitored throughout. Results and Discussion Figure 2 presents the initial activities (blue) and the activities evaluated after 5,000 PCs loading (red) of Pt/CCA/Pt(111), and those of Pt/Cantor alloy/Pt(111) and Pt/Co/Pt(111) are also shown as references. [3] Specifically, two synthesis conditions of Pt/CCA/Pt(111), Cr ₁ Mn ₁ Fe ₃ Co ₃ Ni ₂ _573K and Mn ₁ Fe ₂ Co ₅ Ni ₂ _673K annealing, exhibit ORR properties either equivalent to or exceeding those of the benchmark Pt/Cantor alloy/Pt(111) with conducting 18 experimental trials by assisting the BO. Notably, these conditions were successfully found through a search of 2 percent of the total exploration space, including training data. Moreover, it is undoubtedly a crucial insight that, for Pt-“Cr-containing CCA (Cr-Mn-Fe-Co-Ni)”(111), we have deepened the understanding of surface design guidelines based on the cumulative knowledge of high-ORR-performance strategies for Pt-“multi-element alloys”(111) systems. [3,4] Equally noteworthy, however, is our finding that even Pt-“Cr-free CCA (Mn-Fe-Co-Ni)”(111) can achieve high ORR performance through optimal synthesis conditions, which intriguingly broadens the potential of high-ORR-performance Pt-CCA catalysts. Therefore, the results demonstrate that our experimental study platform for a single-crystal Pt-HEA (Pt-CCA) model catalyst becomes more powerful with the incorporation of the machine-learning method, e.g., BO, that enabled the stepwise and efficient exploration of the optimal conditions for practical Pt-CCA ORR catalyst synthesis. At the meeting, we will discuss further insights into the correlation between synthesis conditions and ORR properties, incorporating the results of structural observation. Acknowledgement This study was supported by new energy and industrial technology development organization (NEDO) of Japan, JSPS KAKENHI and JSPS Grant-in-Aid for JSPS Fellows. References [1] Y. Wang et al. , Adv. Mater ., 35 , 2302499 (2023). [2] B. Cantor et al ., Mater. Sci. Eng. A , 375 , 213 (2004). [3] Y. Chida et al ., Nat. Commun. , 14 , 4492(2023). [4] Y. Chida et al ., Electrochem. Commun. , 159 , 107657(2024). Figure 1

We investigated oxygen reduction reaction (ORR) properties of Pt-containing compositionally complex alloy (Pt-CCA) single-crystal model catalyst surfaces to optimize dry-process synthesis conditions, that is, CCA compositions of less-noble alloying elements and their synthesis (annealing) temperatures. Using a machine-learning approach, we effectively navigated the large space of possible synthesis conditions to minimize the experimental workload. The ORR activity and durability of the Pt/CCA/Pt(111) model catalyst surfaces (synthesized through vacuum deposition on a Pt(111) substrate of nonequiatomic Cr-Mn-Fe-Co-Ni or Mn-Fe-Co-Ni alloy (111) lattice stacking layers, followed by a surface Pt(111) layer) depend upon the alloy composition and synthesis temperature: the model catalyst surfaces synthesized with specific combinations of these two parameters outperformed benchmark surfaces such as Pt/equiatomic Cr-Mn-Fe-Co-Ni/Pt(111) in terms of the ORR durability during potential-cycle loading. The outstanding ORR properties are attributed to the use of machine learning to predict synthesis conditions that are closely linked to the atomic-level surface microstructures that favor enhanced ORR properties. These microstructures enable the formation of a so-called "pseudo-core-shell-like structure", i.e., surface Pt(111) underlaid with CCA(111) lattice stacking layers with atomically distributed active elements (Co and/or Ni) close to the surface that are beneficial for ORR property enhancements. This study demonstrates that not only the "high-entropy" effect of charged less-noble CCA elements but also the precise control of elemental distributions in the near-surface vicinity in the pristine state, resulting from optimized CCA compositions and synthesis temperatures, are the key factors to improve Pt-CCA catalyst material systems.

In materials informatics, feature selection and model selection are utilized in the knowledge extraction process, and the confidence evaluation of the selection result is crucial for ensuring the reliability of the extracted knowledge. In this study, we propose a novel method to quantitatively evaluate the significance of low-dimensional models obtained from an expanded feature space using the density of states (DoS) for the evaluation metrics. This method allows us to compare the performance of the selected model to those of other models and evaluate its significance. We further propose an evaluation method for feature importance based on marginal posterior probabilities using the Bayesian model averaging (BMA) framework, which considers all models. We demonstrate the effectiveness of our proposed methods through their application to a crystal structure dataset. Our results show that the DoS analysis reveals the presence of a large number of models with a comparable performance with the best model reported in a previous research work. This suggests that knowledge extracted from a single selected model can be unreliable and that considering other models is crucial. We also show that the BMA-based feature importance evaluation provides valuable insights into the importance of features, highlighting both primary features and functional forms. In addition, we demonstrate that the LASSO + L0 method commonly used in existing research deteriorates the search space for model selection. Our findings suggest that our proposed methods provide valuable tools for assessing the significance of low-dimensional models and extracting knowledge from large-scale data in materials informatics.