Model-based control incorporating neural network optimization of the automated thermoplastic tow-placement process.

摘要

This dissertation demonstrates the use of on-line optimization algorithms to calculate optimum process set points for manufacturing processes such as the automated thermoplastic tow-placement (ATP) technique. These algorithms are implemented in the ATP process and utilize neural network-based process models to predict material quality as a function of process set points. The set points are computed by maximizing the throughput and maintaining a desired minimum quality. A new neural network-based predictive optimization scheme is developed to ensure non-linear optimization over a wide range of processing inputs. Process history can greatly affect the final part quality and, therefore, is an integral part of the optimization. Eventually, the set points (heat input, consolidation forces and deposition rates) have to be controlled in order to adjust to process noise or set point changes. Here, a novel adaptive predictive controller is utilized to maintain a optimal heat input into the substrate. A Cerebellar Model Arithmetic Controller (CMAC) learns to adapt to the heat transfer model on-line with feedback from a head-mounted thermal camera. A second numerical optimization adjusts two stepper motors, which change the distance of two hot nitrogen gas torches that governs the heat input into the substrate. Ultimately, the desired heat profile is reached and the temperature is controlled.; A unique in-situ non-linear optimization technique based on artificial neural networks has been developed. It is validated for the highly non-linear ATP process and successfully predicts optimum processing parameters. Test panels manufactured with the ATP process compare part quality manufactured with and without the intelligent control system. Overall, the analysis shows improved part quality with the control system. This research leads to a model-based predictive controller which can decrease costs by increasing the throughput and improving part quality. The developed approach is applicable to many other manufacturing processes where process simulations exist and conventional control techniques are lacking.