| The physical properties of carbon fibers depend upon the many operating conditions under which they are produced, but much of this dependence is functionally unknown. Also, there is a large time lag (on the order of several weeks in Clemson's pilot-scale process) between spinning and final product quality measurement. Therefore, an advanced control strategy, such as feedforward control, would be required to satisfactorily optimize and control the production process. The objective of this research was to determine the feasibility of developing a computer-implemented process model for the production of mesophase pitch-based carbon fibers which could be used as the basis of an advanced control system.;A successful integrated neural network process model was developed. The model is a combination of fundamental equations, empirical equations, and feedforward neural networks (FNNs). The equations were applied to sparse process data that were compiled from previous researchers in order to transform and simplify the data. Then, the transformed data were used as inputs to the FNNs. Various input configurations and numbers of neurons in the hidden layer were investigated, and the best FNNs were chosen using several criteria. The model was then used in several applications to demonstrate its utility and potential. |
