Beyond Bruggeman: Tomography-Driven Anisotropic Transport Tensors with Calibrated Uncertainty for Li-Ion Battery Electrodes
- SSRN Electronic Journal
Résumé
Porous-electrode battery models — from single-particle to Doyle–Fuller–Newman formulations — overwhelmingly parameterise transport via scalar Bruggeman corrections that cannot capture the microstructural anisotropy of real composite electrodes. Using finite-volume homogenisation (NASA PUMA) on voxelised three-phase NMC cathode microstructures, we compute full anisotropic effective-transport tensors (ionic/electronic conductivity, thermal conductivity, tortuosity) and quantify the electrochemical penalty of the Bruggeman shortcut. For a calendered NMC cathode, Bruggeman overestimates through-thickness effective diffusivity by and underpredicts 4C voltage loss by ; overpotential decomposition reveals ionic ohmic losses dominate (67 % of total at 4C), while graded-porosity architectures shift the penalty to activation overpotential. Validation against a multi-source real-tomography benchmark (NREL/NLR and Toda; NMC1/NMC2; calendered/uncalendered states) yields uncalendered through-thickness tortuosities –3.0 in quantitative agreement with EIS measurements, and establishes a 24-voxel minimum RVE edge length. A 22-case parametric sweep qualified by five-seed 95 % confidence intervals identifies near-percolation design regions (CBD fraction , coarse particle-size distributions) where and single-seed design maps are statistically unreliable. A Gaussian-process surrogate trained on 97 PUMA-labelled cases achieves mean leave-one-out with 93.9 % calibrated predictive-interval coverage at the speed of direct simulation. The workflow links voxel-scale electrode architecture to uncertainty-qualified, surrogate-accelerated transport tensors for physics-informed battery electrode design.Porous-electrode battery models — from single-particle to Doyle–Fuller–Newman formulations — overwhelmingly parameterise transport via scalar Bruggeman corrections that cannot capture the microstructural anisotropy of real composite electrodes. Using finite-volume homogenisation (NASA PUMA) on voxelised three-phase NMC cathode microstructures, we compute full anisotropic effective-transport tensors (ionic/electronic conductivity, thermal conductivity, tortuosity) and quantify the electrochemical penalty of the Bruggeman shortcut. For a calendered NMC cathode, Bruggeman overestimates through-thickness effective diffusivity by and underpredicts 4C voltage loss by ; overpotential decomposition reveals ionic ohmic losses dominate (67 % of total at 4C), while graded-porosity architectures shift the penalty to activation overpotential. Validation against a multi-source real-tomography benchmark (NREL/NLR and Toda; NMC1/NMC2; calendered/uncalendered states) yields uncalendered through-thickness tortuosities –3.0 in quantitative agreement with EIS measurements, and establishes a 24-voxel minimum RVE edge length. A 22-case parametric sweep qualified by five-seed 95 % confidence intervals identifies near-percolation design regions (CBD fraction , coarse particle-size distributions) where and single-seed design maps are statistically unreliable. A Gaussian-process surrogate trained on 97 PUMA-labelled cases achieves mean leave-one-out with 93.9 % calibrated predictive-interval coverage at the speed of direct simulation. The workflow links voxel-scale electrode architecture to uncertainty-qualified, surrogate-accelerated transport tensors for physics-informed battery electrode design.
Mots-clés
lithium-ion battery electrodes, NCM cathode, microstructural anisotropy, PUMA, tomography, effective diffusion coefficient, tortuosity, graded porosity, calendering, representative volume element, Kriging surrogate model, uncertainty quantification