Temperature prediction and regulation for complex curved parts during automated fiber placement combining FE simulation and machine learning

Layup temperature is the most sensitive process parameter that impacts the prepreg tack and placement quality. Multi-physics-based process modeling for laying temperature on complex curve structures is time-consuming and notoriously difficult due to the interaction between process conditions and material parameters. This paper develops a hybridized model, combining a FE model (FEM), and a direct and inverse data-driven machine learning model (DDMLM), that can be utilized to simulate the heating process of AFP and control the material temperature for complex curved structures. In it, the dataset obtained from the FEM is first utilized to inform a direct data-driven machine-learning model that can obtain the relationship between layup temperature, heating power, and head speed through training, testing, and validation. Then, an inverse machine learning model is established to estimate the heating power for the defined layup temperature. Finally, the hybridized model is exemplarily executed on a winglet mold to confirm the benefits of such an integration. The results validate that the model can improve the temperature prediction efficiency and realize temperature control accurately.