A network-based fuzzy inference system for sonodisruption process of re-assembled casein micelles

Abstract The mathematical modeling of sonodisruption process of re-assembled casein micelles is very difficult due to its ill-defined nature. An adaptive network-based fuzzy inference system (ANFIS) was therefore used to model the process descriptively. Initially, two primary networks were obtained either by grid partition or subtractive clustering. Although grid partitioning led to an ANFIS with higher recognition capability, we continued to improve the ANFIS obtained by subtractive clustering because it described the process with so fewer rules. Increased number of epochs resulted in a model with higher recognition capability while maintaining its simple structure. When the squash factor increased from 1.25 to 1.45, recognition capability of model was further increased and its rule base became simpler (16 fuzzy rules). Finally an ANFIS was generated by applying a lower range of influence. It possessed the highest recognition capability amongst the models but with a comparatively more complex rule base (41 fuzzy rules).