The electrode formulation has a significant effect on the performance of lithium ion cells. The active material, binder, and conductive carbon all have different roles, and finding the optimum composition can be difficult using an iterative approach. In this study, a design of experiment (DoE) methodology is applied to the optimisation of a cathode based on lithium iron phosphate (LFP). The minimum LFP content in the electrodes is 94 wt%. Seventeen mixes are used to evaluate adhesion, resistivity, and electrochemical performance. The coating adhesion increases with binder content, and the coating conductivity increases with carbon nano-tube content. The best coatings achieve 5C:0.2C capacity ratios above 50%, despite the relatively high coat weight. Models based on just the component mixture do not replicate the discharge capacities at high rates. However, a combined mixture + process model can fit the data, and is used to predict an optimum formulation.

The size of a lithium iron phosphate (LFP) cathode mix was increased by a factor of thirty, and the capacity of the cells produced with it by a factor of three-hundred. As well as rate and cycling tests, the coatings were also characterised for adhesion and resistivity. The adhesion and total through-plane resistance were both dependent on the drying conditions during coating. The discharge capacities at high rates and the pulse resistances showed much less influence from the drying temperature. The mix formulation contained 97 wt% LFP, and was based on an earlier design of experiments (DoE) study, using relatively high active material contents. Overall, the mix exceeded the performance predicted by the modelling study.

Lithium ion electrodes with carbon fibres can show X–Y anisotropy in their electronic conductivity. The interface resistance between coatings and foils is much larger than the coating resistance; for cathodes the former is reduced by calendering.

Join counting, a standard technique in spatial autocorrelation analysis, has been used to quantify the clustering of carbon, fluorine and sodium in cross-sectioned anode and cathode samples. The sample preparation and EDS mapping steps are sufficiently fast for every coating from two Design of Experiment (DoE) test matrices to be characterised. The results show two types of heterogeneity in material distribution; gradients across the coating from the current collector to the surface, and clustering. In the cathode samples, the carbon is more clustered than the fluorine, implying that the conductive carbon component is less well distributed than the binder. The results are correlated with input parameters systematically varied in the DoE e.g. coating blade gap, coating speed, and other output parameters e.g. coat weight, and electrochemical resistance.

An abstraction can be used to relate two structural causal models representing the same system at different levels of resolution. Learning abstractions which guarantee consistency with respect to interventional distributions would allow one to jointly reason about evidence across multiple levels of granularity while respecting the underlying cause-effect relationships. In this paper, we introduce a first framework for causal abstraction learning between SCMs based on the formalization of abstraction recently proposed by Rischel (2020). Based on that, we propose a differentiable programming solution that jointly solves a number of combinatorial sub-problems, and we study its performance and benefits against independent and sequential approaches on synthetic settings and on a challenging real-world problem related to electric vehicle battery manufacturing.