Neural network-based optimal curing of composite materials

Polymer-matrix composites using thermosetting resins as the matrix are increasingly finding use in several applications. A widely-acknowledged impediment to their widespread commercial use is the high cost associated with their manufacture, arising from the long processing cycle times. Towards improving the manufacturing affordability of composite materials, several research efforts have been directed at the use of simulation models to obtain optimal cure cycles for the manufacture of composites. However, the computational tedium associated with a rigorous numerical process simulation hampers the practical effectiveness of a numerical simulation-based optimization endeavor. With the objective of alleviating the computational tedium, this paper presents the use of an artificial neural network in conjunction with a nonlinear programming technique for determining optimal cure cycles for the fabrication of thermosetting-matrix composites. The neural network is trained in terms of non-dimensional groups formed of the process and product parameters, which provides for better incorporation of the physical relationships among the parameters, for minimization of the training variables, and for generalization of the network training across material systems. Optimal cure cycles are reported for a wide range of practically-relevant processing conditions.