An Autonomous Platform for Electron Paramagnetic Resonance Spectra

The comprehensive characterization of spin species continues to be a formidable challenge in the fields of chemistry, materials science, and biology. Traditional methods for Electron Paramagnetic Resonance (EPR) spectroscopy, while offering high precision, are impeded by significant time requirements and a dependency on extensive expert knowledge, which restrict their practicality and widespread application. Here, we introduce a hybrid approach that combines conventional computational techniques, machine learning, and an automated measurement system for the analysis and characterization of open shell species. Our methodology incorporates a multi-channel feature transformation alongside a deep learning model and a multi-grain iterative optimization method to accurately identify parameters in EPR spectra. Furthermore, our system utilizes a comprehensive, literature-derived EPR database, enabling rapid and accurate identification of spin species in EPR spectra in real catalytic systems. Our approach not only aligns with the accuracy of human experts, maintaining a margin of error within 0.1 Gauss, but also greatly enhances analysis speed by automating parameter adjustments and species identification. By integrating our spectral recognition system into an automated EPR measurement setup, we have successfully achieved the measurement and characterization of 36 samples within one hour, thereby streamlining the workflow and increasing throughput significantly. This advancement represents a pivotal development in EPR spectroscopy, bridging the gap between high-throughput demands and the need for precise, reliable analytical techniques.