Adapting sigmoid functions for hydrogel swelling curve prediction with neural networks.

Stimuli‐responsive hydrogels are representatives of smart materials with enormous swelling capability. The prediction of discrete hydrogel swelling states by artificial neural network based on the processing parameters has been realized in our previous work. In the current study, we explore ways to enhance the prediction capabilities by integrating physical information of the swelling curves to the model: Instead of predicting discrete swelling states, we predict the mathematical parameters of the continuous swelling curve. We therefore assume that the swelling behavior of hydrogels is consistent with a sigmoidal tanh function, based on their physical properties. After predicting the parameters of the sigmoidal function with the trained model, we analyze the prediction accuracy in comparison to the initial discrete prediction. Moreover, the new approach then allows us to physically interpret the different material properties, which denote sensitivity, maximum achievable differential swelling, and the position of the reference point for derivation of a material model. [ABSTRACT FROM AUTHOR]