Introducing a Machine Learning Approach to Uncover Fundamental Electrochemical Insights from in-Situ Spectroelectrochemical Data

The application of machine learning (ML) approaches in fundamental electrochemistry has been limited by several intrinsic challenges. These include the complexity of electrochemical data, difficulties associated with generating extensive and representative datasets, and limitations inherent in existing analytical methodologies1-2. In-situ spectroelectrochemical experiments, however, produce rich, multidimensional data, particularly thin-layer spectroelectrochemistry (TL-SEC), which offers diffusion-free conditions typically described by the Butler-Volmer (BV) model. State-of-the-art physics-based analyses of SEC data on multi-electron transfer systems involve using multivariate chemometric approaches such as Evolving Factor Analysis to extract redox species concentrations from potential-dependent spectral evolution, subsequently fitting these to the BV model to determine key thermodynamic and kinetic parameters3-4. While effective, these methods often involve extensive data preprocessing, spectral deconvolution, fitting calibration, and can be sensitive to experimental noise.

In this work, we introduce an innovative data-driven deep learning framework capable of directly deriving thermodynamic and kinetic information from SEC datasets, without the intermediate steps of conventional data deconvolution and fitting. Our approach leverages a convolutional neural network (CNN) trained using a hybrid dataset that integrates experimental SEC data with systematically scalable simulations derived from a physics-based engine. This multi-task learning CNN model achieves excellent validation and test performance with notably low MSE and MAE losses. Validation against unseen simulated and experimental TL-SEC data confirms the model’s high accuracy in predicting species concentrations, currents, and electrochemical parameters. Our hybrid ML methodology compares favorably with the physics-based algorithm and demonstrates potential generalizability to broader electron-transfer mechanisms.