Real-time process monitoring and prediction of flow-front in resin transfer molding using electromechanical behavior and generative adversarial network

Abstract

Lightweight materials have been utilized for several decades, offering advantages in industries such as aerospace, automotive, and wind turbine manufacturing. Among these, fiber-reinforced plastics are widely utilized owing to their excellent mechanical properties. Resin transfer molding—a method for manufacturing thermoset composites—is susceptible to dry spots, which degrade the mechanical properties. Accurately identifying and predicting the flow front can enhance process robustness, ensuring defect-free composites. This study presents a novel approach for identifying flow fronts in real time and predicting flow-front scenarios using a tree model and a generative adversarial network (GAN). First, the changes in electrical resistance during infusion were investigated by examining the electromechanical behavior. Subsequently, by leveraging the electrical resistance data, linear equations were formulated to identify the locations of flow fronts between the electrodes. Finally, possible flow-front configurations were identified across three sections, encompassing 17 scenarios, using the developed identification model. Flow-front prediction was conducted using the tree model, which evaluated all 17 scenarios and tracked the most relevant scenarios according to probability. Additionally, the GAN generated more realistic flow-front configurations, enhancing both the identification and prediction of the flow. This model can reflect the racetracking effect without considering the permeability of the fiber preform, significantly reducing the computational cost compared with numerical simulations. Moreover, the flow-front prediction model effectively mirrored the experimental results, outperforming numerical simulations in both adaptability and speed. By utilizing this model, operators can identify defects such as dry spots in real time and predict their locations using the predicted flow-front configuration.