Physics-informed distribution of relaxation times estimation and latent-space condition monitoring of solid oxide fuel and electrolysis cells from electrochemical impedance spectroscopy
A physics-informed convolutional autoencoder estimates the distribution of relaxation times (DRT) directly from electrochemical impedance spectroscopy (EIS) data without spectrum-specific tuning. The model resolves overlapping relaxation processes in synthetic two-ZARC spectra and reconstructs measurements from three independent solid oxide fuel and electrolysis cell datasets with range-normalised errors below 1.1%. Decoder-probe analysis shows the learned latent representation is organised according to relaxation timescale, and distances in this latent space capture operating changes, hydrogen-shortage events, and long-term degradation.
Researchers propose a physics-informed convolutional autoencoder that estimates the distribution of relaxation times (DRT) from electrochemical impedance spectroscopy (EIS) data. A discretised relation between impedance and DRT is embedded in training, constraining the network to produce impedance-consistent distributions. The model resolves overlapping relaxation processes in synthetic two-ZARC spectra and accurately reconstructs measurements from three independent solid oxide fuel and electrolysis cell datasets, with range-normalised errors below 1.1%. Decoder-probe analysis shows that the learned latent representation is organised according to relaxation timescale. Distances in this latent space capture operating changes, hydrogen-shortage events, and long-term degradation. The same lightweight architecture is applied across all datasets without modification, providing consistent DRT estimation and an interpretable basis for condition monitoring.
The approach embeds a physics-based forward model (discretised impedance-DRT relation) into the training loss of a convolutional autoencoder, ensuring that estimated DRTs are consistent with measured impedance. This physics-informed constraint reduces sensitivity to regularisation choices and enables direct DRT estimation without per-spectrum tuning. The latent space learned by the autoencoder is organised by relaxation timescale, allowing unsupervised condition monitoring via latent distances.
For solid oxide fuel cells and electrolysis cells, this method offers a consistent, automated DRT estimation pipeline that can be applied across different datasets without modification. The interpretable latent space enables detection of operating changes, hydrogen-shortage events, and long-term degradation, which could support real-time condition monitoring and predictive maintenance in energy systems.
The method reduces the need for expert tuning in DRT analysis, potentially lowering operational costs for fuel cell and electrolysis system monitoring. Early detection of degradation and hydrogen-shortage events could improve system uptime and safety, creating value for manufacturers and operators of solid oxide cells.
Next signals to watch include validation on larger and more diverse fuel cell/electrolysis datasets, integration into real-time monitoring systems, and extension to other electrochemical devices such as batteries. The lightweight architecture suggests potential for deployment on edge hardware for online diagnostics.