Neural Networks and Genetic Algorithms Used for Modeling and Optimization of the Siloxane‐Siloxane Copolymers Synthesis

Abstract

This paper presents the use of neural networks and genetic algorithms as tools for modeling and optimization applied to a complex polymerization process–synthesis of statistical dimethyl‐methylvinylsiloxane copolymers. A feed forward neural network models the dependence between the conversion of monomers and copolymer composition (output variables) and working conditions (temperature, reaction time, amount of catalyst and initial composition of monomers–input variables). The training and validation data sets are gathered by ring‐opening copolymerization of the octamethylcyclotetrasiloxane (D₄) with 1,3,5,7‐tetravinyl‐1,3,5,7‐tetramethylcyclotetrasiloxane (D₄ ^V), with a cation exchange (styrene‐divinylbenzene copolymer containing sulfonic groups) as a catalyst, in the absence of solvent. This model is included into an optimization procedure based on a scalar objective function and solved with a simple genetic algorithm. The genetic algorithm computes the optimal values for the control variables and for the weight coefficients attached to the individual objectives. An inverse neural network modeling, that is the identification of reaction conditions leading to a desired value for copolymer composition, is presented as particular variant of optimization. The genetic algorithm and neural networks prove to be good and accessible tools for solving an optimization problem performed with a multi‐objective scalar function and provide important information for the experimental practice.

Keywords:

• neural networks
• direct and inverse modeling
• genetic algorithms
• optimization
• siloxane copolymer
• polydimethylmethylvinylsiloxane