Deep-learning framework breakthrough: AI-guided heterogeneous cathode design supercharges lithium-ion battery kinetics

Demand for higher energy density and faster charging capabilities in lithium-ion batteries has driven researchers to engineer thicker electrodes. However, as cathode thickness increases, severe kinetic limitations arise: sluggish ionic and electronic transport leads to non-uniform reaction distributions, increased polarization, and incomplete active-material utilization.

To overcome these structural bottlenecks, a research team from Huazhong University of Science and Technology, alongside collaborators from South China Normal University, has developed an innovative deep-learning-guided inverse design framework. Published in ENGINEERING Energy, the study demonstrates that optimizing heterogeneous porosity distributions and electrode thickness can significantly boost the discharge capacity of LiNi0.6Co0.2Mn0.2O2 (NCM622) cathodes across a wide range of discharge rates (0.5 C to 4 C).

The researchers established a physics-based pseudo-two-dimensional (P2D) half-cell model to construct a database containing nearly 100,000 data samples linking electrode thickness, discharge rate, and spatial porosity distribution to voltage-capacity responses. Using this dataset, they trained a conditional generative adversarial network (cGAN) to serve as a high-fidelity surrogate model, which was subsequently integrated with particle swarm optimization (PSO) for rapid inverse structural design.

Key Highlights and Core Research Findings

• P2D-cGAN-PSO Integrated Framework: The cGAN surrogate model accurately predicted discharge voltage-capacity curves with exceptional precision (MAE = 0.0077 mAh, RMSE = 0.02 mAh, R2 = 0.99, and a mean relative error of only 0.49%), reducing the computational burden of traditional physics-based simulations during high-dimensional structural optimization.
• Enhanced Kinetic Capacity via Porosity Engineering: At a fixed electrode thickness, optimizing heterogeneous porosity distributions improved discharge capacity across all tested discharge rates compared to uniform baseline structures:
  • 0.5 C: Capacity increased by 1.74% (from 5.06 to 5.15 mAh).
  • 1 C: Capacity increased by 1.39% (from 4.66 to 4.73 mAh).
  • 2 C: Capacity increased by 3.39% (from 2.99 to 3.10 mAh), representing the largest performance enhancement.
  • 4 C: Capacity increased by 1.66% (from 1.60 to 1.63 mAh).
• Unveiling Kinetic vs. Thermodynamic Mechanisms: P2D numerical analysis revealed that while NCM622's sloped equilibrium-potential curve helps homogenize local reaction rates at low discharge rates, transport limitations dominate at high rates. High discharge rates amplify spatial disparities in reaction rates, rendering active materials near the current collector underutilized.
• Rate-Dependent Structural Trade-offs: PSO inverse design confirmed that optimal electrode thickness decreases as discharge rates rise (shifting from ~150 μm at low rates to ~130 μm at 2 C and ~72 μm at 4 C) due to narrower porosity tolerance windows and intensified polarization constraints.

"Electrode thickness primarily controls active-material loading, whereas spatial porosity distribution provides a critical secondary degree of freedom to mitigate transport limitations and enhance active-material utilization," noted the research team. "Our cGAN-PSO framework bridges the gap between data-driven acceleration and physical electrode-scale optimization, providing clear simulation-based guidance for next-generation high-performance battery cathode design."

· Journal: ENGINEERING Energy

· Read the full article for free: https://rdcu.be/zFhPKsJsMKWa

· Cite this article: Wang, S., Bai, X., Lei, Y., Pei, J., Ma, L., Long, R., Liu, Z., & Liu, W. (2026). Deep-learning-guided heterogeneous cathode structure engineering for enhanced kinetics in lithium-ion batteries. ENGINEERING Energy, 20(5), 10910. https://doi.org/10.1007/s11708-026-1091-z

· Follow us on X and LinkedIn