Curing curve prediction using genetic algorithm optimized back propagation neural network

Abstract

Thermoset composites constructed with monolithic moulding can significantly reduce the number of parts and connectors, improve the reliability of the composite structure, and reduce production costs. For such composite components, the degree-of-cure (DoC) curve is normally considered to evaluate the curing process, so that the curing process-induced deformation and residual stress can meet the manufacturing requirements. In this paper, a method is proposed to predict the DoC curve based on the neural network. First, the DoC curves for AS4/3501-6 composite structures were obtained from ABAQUS and HETVAL subroutine, and the results were verified by published values. Next, the curves of temperature over time and DoC over time were generated to train the neural network. Finally, the back propagation (BP) neural network is optimized using the genetic algorithm (GA), and the DoC curve prediction model is created. The results demonstrate that, with a maximum error of 3.24%, the simulated curves essentially correspond with published values. Evaluation and validation analyses show that the prediction of the DoC curve using the GA-BP neural network yields high computational efficiency and accuracy.