The independent claim is where the value lives, and Honda's grant US12019122B2, "Learning method, state estimation method, and state estimation device for state estimation model of secondary battery" (issued June 25, 2024), splits its value across two: the learning method that trains the model, and the estimation method that applies it. The CPC is pure battery-diagnostics — G01R 31/367 (battery testing), 31/388 and 31/392 (state-of-health estimation) — with no vehicle-control classes, which keeps the claim tightly on estimation.
The strategic content is the move toward learned estimation. Classical battery state estimation uses physics-based equivalent-circuit models; the industry has been migrating to data-driven models that capture aging and nonlinearity the physics models miss. A claim that recites the training method as well as the inference method is the more valuable structure, because it can read both on whoever trains such a model and on whoever runs it. Claiming the learning step extends the reach upstream.
On scope, the two-part claim is a deliberate breadth play, but it must clear dense prior art — learned battery-state estimation is an active research area. The defensible novelty is the specific training-and-inference technique, and the dependents that pin down the model structure, the features, and the loss are the moat. As a granted B2, the scope survived examination, which suggests the limitations are concrete enough to distinguish over the general idea of an ML battery model.
Honda holding this in June 2024 fits the broader pattern of automakers internalizing battery-management IP rather than leaving it to cell suppliers. State-of-health estimation underpins warranty, resale value, and safety, so owning the estimation model is a genuine competitive asset. The teardown verdict: a focused, examined grant whose breadth comes from claiming both training and inference — read the model-structure dependents to see how narrowly the learned method is actually pinned down.