A Comparative Study of Neural Network Architectures Applied to the Manufacturing of Composite Materials

The current investigation focused on the comparison of two neural network architectures applied to intelligent control of resin transfer molding (RTM) processes. The objective was to develop an intelligent control scheme that is efficient, generic and applicable to real-time control of manufacturing processes. In an earlier study, the authors introduced a neural network based control scheme to accomplish this goal. The neural network used in that study was a forward model neural network which needed to be coupled with a search mechanism in order to achieve a forward model control method. In a recent study, the authors introduced an inverse model neural network based control which eliminated the need for such a search mechanism. However, it was shown that the mapping of the inverse model neural network was harder to learn than that of the forward model neural network and thus the performance of the inverse model control scheme was inferior. In the present study, the authors investigated an improved inverse model neural network architecture. The focus of this study was the control of progression of flow during resin transfer molding processes. It was shown that considering the slope information of the flow fronts resulted in an improved inverse model neural network architecture and the resulting inverse model control method performed comparably with the much less efficient forward model control scheme utilizing a forward model neural network and a rigorous search method.