Researchers from the University of Pittsburgh, including Brian Lee, Linna Qiao, Samuel Gleason, Guangwen Zhou, Xiaohui Qu, Judith Yang, Jim Ciston, and Deyu Lu, have developed a machine learning model to improve the analysis of X-ray absorption spectroscopy (XAS) and electron energy-loss spectroscopy (EELS) data. These techniques are crucial for studying the redox states and structures of materials used in energy applications, such as batteries and catalysts. The research was published in the journal Nature Communications.
The team’s model combines an autoencoder, which standardizes the spectra, and a transformer model, which predicts the oxidation state and Bader charge of copper directly from L-edge spectra. The model was trained on a large dataset of FEFF-simulated spectra and tested on both simulated and experimental data. The results showed highly accurate predictions across different types of spectra, including simulated XAS, experimental XAS, and experimental EELS.
This advancement is significant for the energy industry as it enables more efficient and accurate analysis of copper redox processes under various conditions. The model’s ability to handle diverse experimental conditions and mixed valence materials makes it particularly useful for in situ and operando studies, which are essential for understanding the real-time behavior of materials in energy devices. By improving the quantitative analysis of these spectra, the model can help accelerate the development and optimization of energy materials and technologies.
This article is based on research available at arXiv.