AI-Assisted Physics-Informed Predictions of Degradation Behavior of Polymeric Anion Exchange Membranes

Abstract

The global transition to hydrogen-based energy infrastructures faces significant hurdles. Chief among these are the high costs and sustainability issues associated with acid-based proton exchange membrane fuel cells. Anion exchange membrane (AEM) fuel cells offer promising cost-effective alternatives, yet their widespread adoption is limited by rapid degradation in alkaline environments. Here, we develop a framework that integrates mechanistic insights with machine learning, enabling the identification of generalized degradation behavior across diverse polymeric AEM chemistries and operating conditions. Our model successfully predicts long-term hydroxide conductivity degradation (up to 10,000 hours) from minimal early-time experimental data. This capability significantly reduces experimental burdens and may expedite the design of high-performance, durable AEM materials.

Figures (8)

• Figure 1: Schematic of the PENN architecture. Polymer genome fingerprints and environmental features are input to a multilayer perceptron (MLP), which predicts four physically meaningful degradation parameters: initial conductivity $\sigma_0$, limiting conductivity $\sigma_{\infty}$, characteristic time $t_0$, and decay shape parameter $\alpha$. These parameters are then passed through a mechanistic degradation equation and compared to experimental time series to guide training via a physics-informed loss function.
• Figure 2: Parity plots comparing predicted versus true hydroxide conductivity across all test samples using (a) PENN, (b) NN and (c) GPR models. All models show good accuracy and consistency across the range of predicted conductivity values when using the entire dataset for training.
• Figure 3: Representative degradation curves comparing PENN (blue), NN (orange) and GPR (green) predictions against experimental data (black) for six different AEM samples. Each model was trained on all available data. The top row depicts cases with more drastic degradation, while the bottom row depicts more moderate degradation profiles.
• Figure 4: Normalized degradation behavior across all AEM samples. The PENN-predicted degradation curves collapse onto a universal master curve defined by Equation \ref{['deg_normalized']}. The blue line represents the idealized form $y=\frac{1}{1+x}$. This agreement across chemistries and conditions reveals a shared empirical degradation mechanism and confirms the ability of the PENN to uncover universal trends.
• Figure 5: Distribution of PENN-predicted degradation parameters across all AEM samples. Histograms show the learned values of $\, \sigma_0$ (top left), $\, \sigma_{\infty}$ (top right), $\alpha$ (bottom left), and $t_0$ (bottom right). These distributions reflect the variability in conductivity behavior across different chemistries and testing conditions, highlighting materials with sharper or more gradual degradation.