applications of multi-layer perceptron artificial neural networks for polymerization of expandable polystyrene by multi-stage dosing initiator

In this research, expandable polystyrene (eps) polymerization with conventional and multi-stage initiator dosing (mid) methods is simulated by multi-layer perceptron (mlp) artificial neural networks (ann). in order to optimize mid method, an efficient algorithm was employed for optimal training of the neural network. an algorithm was used to train the mlp networks more rapidly and efficiently than the conventional procedures. the main objective of mid method implementation is to reduce the time of the polymerization and because of that, by having different tests (first stage polymerization at 4, 3.5, 3, 2.5 hours and different amounts of used initiator at common state 100, 80, 75, 70 percent and the different number of dosings 12, 10, 8, 6) it was found that in an optimal state, the first stage polymerization time can be 3 hours and amount of the used initiator can be reduced to 70% in comparison to common state and number of dosings can be 6 times. the results of the simulation showed that the time of the first step of the polymerization has been reduced, the amount of the used initiator has been optimized and the count of the dosing times reduced to half, and therefore the time of the eps polymerization is reduced to 60% of the conventional method.