| Abstract Scope |
Lately, the increasing use of fiber-reinforced composites in critical load-bearing structures within an aircraft, along with the trend such as larger wind turbine blade sizes, necessitates stricter tolerances to produce composite parts. To gain deeper insight into the composite manufacturing process and address potential sources of out-of-tolerance issues, there is a growing interest in simulating composite manufacturing process, particularly in composite curing, as curing-induced distortion is a primary cause of out-of-tolerance issues.
A standard composite curing simulation often involves thermo-chemical coupling for temperature and degree of cure prediction, and thermo-mechanical coupling for predicting curing-induced distortion. It often requires to be run in a transient analysis, which could be computationally intensive, posing challenges in executing various design cases. In this study, our objective is to enhance the efficiency of curing simulation by introducing a methodology that integrates simulation with machine learning, facilitating rapid analysis of composite curing processes. Using a C shape composite plate as an example, multiple curing simulations are conducted. Each simulation involves a different combination of plate curvature and layup designs. A deep learning model is then developed and uses the generated simulation results for training. Once the training process is completed, the model can be used to predict the key curing results such as curing induced distortion, degree of the cure, and maximum residual stresses for a new design within seconds. |