Machine Learning Accelerates the Synthesis of Novel Catalyst for Green Energy
Other Articles
Recent advancements in the application of artificial intelligence (AI) have opened new avenues for the discovery of novel materials for ceramic fuel cells.
Study conducted by Prof. Meng NI and his research team
Ceramic fuel cells, also known as solid oxide fuel cells (SOFCs), convert chemical energy to generate electricity and heat. SOFCs take advantage of high efficiency, low emissions and the flexibility to choose from a variety of fuel sources. These electrochemical devices generate electricity through a fuel oxidation reaction at the anode and an oxygen reduction reaction at the cathode, with an ion-conducting ceramic electrolyte stacked between the anode and the cathode. Oxygen is reduced in the cathode to oxide ions, which are conducted to the anode, liberating electrons to the circuit (Figure 1).
Figure 1. A fuel oxidising reaction in ceramic fuel cells.
The development of high-performance and durable cathode materials is key for efficient and durable ceramic fuel cells. The commonly used cathode materials need to operate at a high temperature (800–1,000°C), which poses difficulties in sealing, accelerates material degradation and results in high operational costs, making them a significant constraint for the commercialisation of ceramic fuel cells. Traditional trial-and-error approach for new material development is time-consuming and expensive, limiting the ability to conduct large-scale material discovery for fuel cell cathodes.
Writing in Nature Energy, a research team led by Meng NI, a Chair Professor in the Department of Building and Real Estate at the Hong Kong Polytechnic University, employed an experimentally validated machine learning approach to accelerate the discovery of four promising perovskite oxides from 6,871 distinct perovskite compositions as oxygen reduction electrodes in SOFCs (Figure 2). The team further delineated ion Lewis acid strength (ISA) as a crucial physical descriptor, a parameter that helps characterise the catalytic performance for the oxygen reduction reaction activity of perovskite oxides1.
Figure 2. The overall workflow diagram. Machine-learning model training and material screening, experimental verification and density functional theory (DFT) analysis. θ is X-ray incidence angle
First, the team consolidated a small dataset, which contains oxygen reduction reaction activities of different perovskite oxides as well as other features of the metal ions (ionic electronegativity, ionic radius, ISA values, ionisation energy, and tolerance factor) as potential physical descriptors to train the machine learning models. The models were trained to learn the underlying composition-activity association of the perovskite oxides. The team designated the polarisation resistance of the catalyst, in terms of low area-specific resistance (ASR) at lower temperatures of around 600–750°C, as the active indicator for determining the optimum parameters of fuel cell performance. Several machine-learning algorithms, including linear and non-linear methods, were employed to fit the models. Among all methods, the artificial neural network (ANN) model achieved the best-fitting results, and it was used to rank the importance of features among potential physical descriptors. Ni’s team identified that the Lewis acid strength (ISA) of metal ions is efficient in predicting the active indicator. Along this line, thousands of distinct perovskite oxide compositions were screened by the models, and among them, four perovskite oxides (SCCN, BSCCFM, BSCFN, and SBPCFN) were found to have promising features. To verify the predictions, Ni’s team synthesised the four catalysts and conducted characterisation and electrochemical tests. The results indicate that the four discovered catalysts have outperforming activity metrics compared to the benchmark material (BSCF). In particular, SCCN exhibits excellent oxygen reduction activity with an extremely low ASR. It was verified with density functional theory (DFT) calculations, a quantum mechanical modelling method used to investigate the electronic structure information (Figure 3).
Figure 3. Density functional theory (DFT) calculation of electronic structure evolution. The model (a) and differential charge density (b) of BSCCFM-m for illustration. The yellow region represents charge accumulation, whereas the blue region represents charge reduction
The team has successfully demonstrated the use of machine learning techniques in discovering highly active fuel cell cathode catalysts. By predicting the features of compositions based on molecular formulas, the model shows promise in efficiently screening a vast number of compositions. It is also more efficient to synthesise only those molecular models with promising predicting features for model verification and testing. With emerging data initiatives to describe the oxygen reduction reaction activity of different perovskite compositions at lower temperatures, the process of discovering advanced catalysts could be accelerated in the future.
The research is supported by National Natural Science Foundation of China (grant no. 51827901, H.X.), Project of Strategic Importance Program of The Hong Kong Polytechnic University (grant no P0035168, M.N.), Sichuan Science and Technology Department (grant no. 2020YFH0012, H.X.), Program for Guangdong Introducing Innovative and Entrepreneurial Teams (grant no. 2019ZT08G315, H.X.), as well as the Natural Science Foundation of Guangdong Province (grant no. 2020A1515010550, H.X.). The data source is available at: https://www.nature.com/articles/s41560-022-01098-3#Sec21, while the code for the study is available at: https://github.com/nlpcui/MLORR.
Reference |
---|
1. Zhai, S., Xie, H., Cui, P., Guan, D., Wang, J., Zhao, S., Chen, B., Song, Y., Shao, Z., Ni, M. A combined ionic Lewis acid descriptor and machine-learning approach to prediction of efficient oxygen reduction electrodes for ceramic fuel cells. Nat Energy 7, 866–875 (2022). https://doi.org/10.1038/s41560-022-01098-3
Prof. Meng NI in the Department of Building |