AI Improves EV Battery Health Estimates

Researchers advanced EV battery-pack engineering across materials, structural design, sensing, and prediction, producing a topology-optimized carbon-fiber-reinforced polymer (CFRP) enclosure prototype weighing 4.78 kg that is compression-molded with rapid 10–15 minute curing and exhibited deformations within acceptable limits. Structural-safety studies show battery packs face complex inertial and cyclic loads that can cause insufficient stiffness and fatigue — including box cracking at lifting lugs — and combining simulation with targeted local redesigns can restore safety and reliability. On the sensing and estimation side, a frequency-domain–enhanced Transformer that fuses FFT-derived features with time-domain statistics and residual connections substantially improves lithium-ion capacity estimation, achieving markedly lower RMSE/MAE and higher R2 on MIT fast-charging data. For trip-level energy management, the IFS-Former — a transformer pre-trained across heterogeneous fleets with trainable missing-feature embeddings and placeholder masks — adapts to inconsistent feature spaces and, after transfer learning, attains R2 ≈ 0.96–0.97 and MAE ≈ 1.19–1.31 while fitting onboard at about 32 MB. These technical advances enable more reliable on-board AI for battery health and trip energy estimation, helping address consumer concerns such as range anxiety and safety. Industry demand for lithium remains critical, positioning major suppliers such as Ganfeng to benefit from accelerating vehicle electrification.