Reproduce “Physics-informed distribution of relaxation times estimation and latent-space condition monitoring of solid oxide fuel and electrolysis cells from electrochemical impedance spectroscopy” on OpenResearch

About this paper

Žan Gorenc, Žiga Gradišar, Felix Mütter, Vanja Subotić, Pavle Boškoski

A physics-informed convolutional autoencoder estimates Distribution of Relaxation Times (DRT) and monitors conditions in solid oxide cells using Electrochemical Impedance Spectroscopy (EIS). The framework achieves consistent, physically constrained DRT estimates and an interpretable latent space that tracks operational events and degradation across diverse datasets without requiring architecture modifications or hyperparameter tuning.

Key insights

• The integral relationship between DRT and impedance can be reformulated as a convolution, enabling convolutional neural networks to solve the ill-posed inverse DRT problem with inherent physical constraints.
• Training a single, shared autoencoder across an entire dataset provides consistent implicit regularization, ensuring that observed differences in DRT estimates genuinely reflect electrochemical changes rather than fitting artifacts.
• The learned latent representation spontaneously organizes itself according to electrochemical relaxation timescales, providing an interpretable basis for condition monitoring without explicit supervision.

Summary by alphaXiv.