<title>Neural network approach to robotic thermoplastic tow placement process control</title>

Abstract

Thermoplastic tow placement is an merging manufacturing technology that gives more flexibility in the design of parts and in cost reduction through on-line consolidation, compared with traditional labor- and time-intensive autoclave processing. This research is focused on developing an on-line control technique that is fast and reliable and can be used to maximize the process throughput. Recently developed process models are integrated into a neural- network-based optimization package, which is capable of locating optimum setpoints in minimal time. A control algorithm with feedback from an IR thermal camera has been developed and is used to achieve the desired process temperature. A feedforward neural network with a cascadien architecture is used to simulate the process. Simulation computations are now possible in less than a second, compared to around 20 minutes for the original FORTRAN simulation. This allows running of the process models on- line for control purposes. An optimization algorithm that suits the rough topology of the network has been written and tested. The algorithm is based on weighting the quality outputs of the neural network and finding the highest process speed for a given minimum quality. The output of the optimization is used as an input to the robot controller and to a temperature controller for the process.