Knowledge-Guided Mamdani-to-Sugeno Neuro-Fuzzy Transformation: A Battery Particle-Cracking Case Study

Particle cracking contributes to lithium-ion battery degradation, motivating modeling approaches that combine numerical adaptability with interpretable degradation knowledge. Mamdani fuzzy systems encode such knowledge through linguistic rules, but transformation into trainable Sugeno models can alter their original meaning. This study develops a knowledge-guided Mamdani-to-Sugeno neuro-fuzzy transformation, evaluated through a controlled case study of particle cracking in lithium-ion battery anodes. A zero-order Sugeno adaptive neuro-fuzzy inference system (ANFIS) learns Mamdani-generated crack-damage responses from C-rate and particle-size inputs, with semantic correction and adaptive blending guiding antecedent adaptation. Across 15 initializations, the proposed model achieved a mean test root mean square error (RMSE) of (1.601±0.093)×10−3. Scratch ANFIS achieved lower numerical error, whereas the proposed method showed stronger correspondence with the Mamdani source. Compared with a structurally aligned unconstrained baseline, rule-firing cosine similarity increased from 0.9012 to 0.9632, while tolerant boundary intersection-over-union (IoU) improved from 0.2672 to 0.4486. These findings demonstrate stronger preservation of interpretable source relationships while maintaining numerical fidelity within the controlled benchmark. Although physical predictive validity remains to be established using measured degradation data, the framework provides an interpretable pathway for converting linguistic degradation knowledge into adaptive numerical models and offers a foundation for extension to broader battery degradation mechanisms.