| Processing thick-sectioned composites (>1/2″ ) can be difficult due to the exothermic nature of the resin and the low thermal conductivity of the composite. This is particularly true for the Resin Transfer Molding process examined here. The temperature profile used to polymerize the resin, otherwise known as a "cure cycle," must be carefully chosen to reduce thermal gradients within the composite while ensuring satisfactory processing times. Instead of trial and error methods that are expensive, time consuming, and non-optimal, we propose a knowledge-based optimization strategy.;In order to be effective, the optimization strategy requires an accurate simulation of the process and supplementary heuristic information. A 1-D finite difference cure simulation was used to simulate the process. Some of the simulation's model input parameters were found directly through experimentation. Other input parameters were identified using a least-squares approach to match the simulation to experimental data from seven test composites. Because an accurate residual stress model for the process was unavailable, a heuristic for predicting the quality of a composite, based on the progression of resin cure within the composite, was used.;Four different global optimization schemes were studied: Random Walk, Simulated Annealing, Genetic Algorithms, and Evolutionary Strategies. The optimal cure cycle suggests heating the composite to initiate cure followed by a cooling stage to ensure inside-out curing. An extension of the Evolutionary Strategies optimization method was developed to account for the known variability in the simulation's input parameters and generate an optimum that can withstand some batch to batch variations.;In addition to the suggested cure cycle, we propose to implement adaptive control. Through a sensitivity analysis of the optimal cure cycle, the conditions and manner of altering the cure cycle were identified. Furthermore, instead of relying on intrusive sensors for control, heat flux sensors were installed within the mold to provide a more effective, reusable source of feedback. A simulated implementation of the optimized control strategy showed remarkable success, reducing the production of bad parts from 59% to 1%. However, heat loss through the mold and sensor noise during the actual experiment did not allow for its successful experimental implementation. |
