A model exists to serve a decision, and the right model for a charger microcontroller is not the right model for cell research. NiMH charge modelling spans a fidelity ladder from empirical voltage curves, through equivalent-circuit networks, to reduced electrochemical models and full physics-based simulations of porous electrodes; each rung buys insight or accuracy at the price of computation and calibration burden. This paper climbs that ladder, states what each model can and cannot predict about charging, and gives a selection framework that matches model fidelity to the decision it must support - termination, current planning, observer design or cell development - so engineering effort is spent where it changes the outcome.
The simplest models tabulate voltage, efficiency and temperature against SOC and current from test data, with little internal structure; the termination logic of Papers 6 to 10 is effectively an empirical model. They are cheap, robust and adequate for a fixed cell and fixed profile, but they cannot extrapolate to untested currents or temperatures, cannot separate causes, and must be rebuilt when the cell changes. They remain the right choice for cost-driven single-cell chargers with a frozen profile.
Their failure mode is silent extrapolation: an empirical table used outside its characterised grid predicts plausible but wrong behaviour, exactly in the fast-charge end band where safety margin matters.
Thevenin and Randles equivalent circuits - ohmic resistance plus RC blocks and a Warburg/diffusion element - add dynamic structure with a handful of physically interpretable parameters identified from pulse tests (Paper 17). They reproduce terminal voltage and heat in real time, embed naturally in Kalman-filter SOC and resistance observers, and run on modest microcontrollers; this is the workhorse fidelity for charge control and fuel gauging.
Their limit is that parameters lump many internal processes together: an RC time constant does not reveal proton concentration, gas generation or local current distribution, so ECMs can predict that voltage will peak but not redesign the electrode that determines why.
Reduced-order electrochemical models retain selected internal states - electrode stoichiometry, averaged concentration, a gas/oxygen balance and pressure - while spatial detail is aggregated or solved with a few nodes. They can predict the oxygen-onset SOC, internal pressure and charge acceptance that an ECM only curve-fits, at moderate computational cost, making them suitable for model-based fast-charge optimisation and for pressure-aware control of the traction-cell type studied in the literature.
They demand more calibration - electrode capacity balance, kinetic and diffusion parameters, recombination coefficient - and benefit from cell-design data a buyer may not fully possess, which is why they are most often built jointly by cell maker and advanced customer.
Full porous-electrode (P2D-style) and multi-physics models solve species, charge and energy conservation across electrode thickness with spatial profiles of concentration, overpotential, gas evolution and temperature; COMSOL-class NiMH equivalent-circuit and electrochemical libraries and published traction models sit here. They reveal local current density, gas pathways and the effect of thickness and formulation on fast-charge ceiling - the tools of cell design and of explaining why a grade behaves as it does.
Their cost is heavy parameterisation and off-line computation; they are not controller models but design models whose output - an oxygen-onset map, a reduced model's parameters - is distilled down to the lower rungs that run in product firmware.
The selection rule follows the decision: fixed-profile termination needs empirical/ECM fidelity; online SOC and health observers need an ECM plus an observer; model-based current planning and pressure-aware fast charge need a reduced electrochemical model; electrode and separator design, and root-causing a pressure or ageing problem, need full physics. Over-modelling a simple charger wastes calibration; under-modelling a fast-charge controller hides the very mechanisms that set its safety limit.
The first figure arranges the fidelity ladder against computational cost and predictive scope; the second maps specific charge decisions to the minimum sufficient model rung, the practical output of the comparison.
The efficient workflow uses each rung to feed the next: physics simulation and bench characterisation produce reduced models, reduced models produce calibrated ECMs and current envelopes, and ECMs run in the shipped controller with conservative margins back to the higher-fidelity results. Weijiang can supply the characterisation data and, for fast-charge programs, co-develop the reduced model parameters that let a customer's controller exploit a cell's real fast-charge ceiling rather than a generic table.
Models become especially powerful when fed by data rather than fixed parameters; the final estimation paper examines data-driven and machine-learning approaches to charge termination and SOC, and where they genuinely beat classical observers.
Weijiang Power designs and manufactures nickel-metal hydride cells, matched packs and charging-ready configurations for consumer, industrial, medical and mobility customers, and supports partners with charge-protocol guidance, IEC 61951-2 performance files, IEC 62133-1 safety evidence and charger co-validation. Share your cell format, charge rate, thermal envelope and cycle target and our engineers will specify a cell-and-charge combination that protects both runtime and service life. Review the range on the products page.
