Dry-spot prediction in glass fiber/Epoxy composites during resin infusion process based on electromechanical analysis and long short-term memory

Abstract

In fiber-reinforced plastic manufacturing, dry spots are typically formed during resin infusion and significantly degrade mechanical properties. Ensuring reliable process monitoring is therefore essential for achieving high-quality composites. This study proposes a resistance-based finite-grid framework for predicting dry spots in glass fiber/epoxy composites during vacuum-assisted resin transfer molding (VARTM). Unlike conventional sensing techniques that rely on embedded or localized sensors, the proposed approach enables area-based sensing during resin infusion using a minimal number of electrodes. A finite resistive-grid model was developed to quantitatively correlate resin impregnation between electrodes, enabling its application to intrinsically non-conductive glass fiber composites, where the resin acts as the primary conductive medium. Experimental measurements demon strated close agreement with the grid model predictions. Based on this validated model, resistance datasets were generated through simulation and used to train a Bidirectional Long Short-Term Memory (BiLSTM) network for dry-spot classification. The trained model achieved a classification accuracy of 89.5 %, successfully identifying both the presence and location of dry spots from time-resistance sequences. By combining deep learning with the developed finite-grid resistance model, this study provides a foundation for data-driven process monitoring in glass fiber composite manufacturing and offers insights applicable to other non-conductive fiber systems during VARTM process.