Advanced State Estimation of Battery Cells and Packs

Battery energy storage systems play a central role in renewable energy integration and transport electrification. Their safe and efficient operation relies on accurate estimation of state of charge (SOC) and state of health (SOH). Reliable pack-level estimation remains challenging because cell-to-cell heterogeneity, thermal gradients, and limited sensing obscure internal behaviour. These challenges are particularly relevant to emerging sodium-ion systems, for which cell-level characterisation and pack-level estimation are not yet as well established as they are for lithium-ion systems.

This thesis bridges cell-level characterisation with pack-level estimation, with a focus on sodium-ion cells and optical-fibre-assisted sensing. First, it presents a protocol-aligned characterisation study of three commercial sodium-ion chemistries and two lithium-ion reference chemistries. For the main layered-oxide sodium-ion chemistry, variability is quantified across 33 cells. Temperature-dependent relationships among capacity, impedance, and open-circuit voltage are established, showing chemistry-dependent polarisation behaviour and stronger sub-zero capacity retention than the tested lithium-ion reference cells under the present protocol. Casing-level strain responses are also characterised, with clear cyclesynchronous signatures in the layered-oxide cells.

Second, the thesis develops an optical-fibre-assisted SOC estimation method for seriesconnected packs. A strain-charge sensitivity method is used to identify representative cells, and representative cell strain together with pack voltage is incorporated into Gaussian process regression models within an adaptive unscented Kalman filter. Validation on a small lithiumion series pack shows accurate reduced-sensing SOC estimation under constant-current and dynamic conditions.

Third, the thesis presents a capacity-based SOH estimation method for a parallelconnected sodium-ion pack under fixed-capacity partial cycling. By fusing rest-window SOC anchor points, pseudo-capacity observations, and a cycle-level Gaussian process-Kalman filter with smoothing, the method reconstructs latent capacity fade with quantified uncertainty. Applied to a five-cell parallel-connected sodium-ion pack (1S5P) over 373 cycles, it reveals substantial underlying degradation despite the externally measured delivered capacity remaining almost unchanged.