Model-based control in the manufacturing of advanced composite materials is one approach to improve the overall quality and performance of the final part. Frequently quantitative nondestructive methods are not available for feedback control, and therefore a predictive controller is required. This study investigates a neural network based control system for the automated thermoplastic composite tow-placement process. This non-autoclave process incorporates an advanced controller that predicts desired part quality at minimal cost. The control system combines a neural network (NN) based model, numerical and neural network optimization, as well as an infrared thermal image camera to sense the temperature profile on the part surface. The optimization utilizes the NN process model to calculate the optimum inputs for the desired quality, whereas the image camera is used for temperature feedback control.