Accelerating Composite Cure Cycle Optimization with Combined Probabilistic Machine Learning and Finite Element Process Simulation

Abstract

Advanced composites are widely used in industries such as aerospace. To certify aerospace structures, the manufacturing process of composites is tightly controlled to ensure compliance with process specifications. For example, in autoclave curing of thermoset composites, process specifications often limit the maximum temperature in the composite part due to exothermic reaction, maximum thermal lag between the part and autoclave temperature cycle, maximum part porosity, and minimum degree of cure. Given all the requirements in process specifications, optimizing the temperature cycle in autoclave curing to reduce cycle time and increase production throughput is often challenging. While finite-element process simulation is used for coupled analysis of heat transfer and thermo-chemical curing reactions during the curing process, optimization across the entire design spectrum is often infeasible because of the long simulation time of high-fidelity models and commonly relies on trial and error and engineering insight. In this study, we present a novel framework based on combined probabilistic machine learning and finite-element simulation to accelerate composite-process optimization. The probabilistic machine-learning model is developed using the underlying theory of curing and the mathematical foundation of Gaussian Process Regression. Instead of trial and error or a time-consuming grid search, this approach relies on targeted finite-element evaluations of selected cure cycles by the machine-learning model, such that an optimized solution is obtained through a step-by-step simulation approach. While minimizing the effort required to optimize cure-processing conditions, the approach also enables evaluation of nontraditional cycles, such as multiple ramps and holds, to minimize autoclave cycle time. For validation, the framework is applied to optimize the cure cycle of a thick HEXCEL AS4/8552 composite laminate cured on a typical Invar tool. A validated finite-element model is used for process simulation along with the machine-learning model for targeted evaluation. Results show that an optimized solution satisfying process specifications while minimizing cycle time can be obtained with only a few analysis steps.