Degradation Modeling Considering Spatial Correlations Based on Distance Indicators

Purpose: In hydrogen fuel cell stacks, degradation in one cell can accelerate degradation in neighboring cells. For this reason, accurately predicting the lifetime of a stack requires accounting for both individual cell data and spatial influences among adjacent cells. Therefore, this study aims to propose a degradation model that incorporates spatial correlation information.<br/>Methods: The proposed approach applies a degradation model that explicitly incorporates spatial correlation among individual cells within the stack, employing exponential covariance functions based on Euclidean and Mahalanobis distances.<br/>Results: The proposed method was applied to degradation data from 18 cells of a stack and demonstrated a lower prediction error compared to an independent cell model. Estimated correlations were stronger among physically closer cells, supporting the presence of localized inter-cell interactions and degradation propagation effects.<br/>Conclusion: This study suggests that modeling spatial dependence among cells significantly improves the accuracy and reliability of lifetime predictions for hydrogen fuel cell stacks. Extending this approach to a full multivariate and potentially mixed-effects spatial framework across all cells is a key direction for building more robust prognostic systems.