Neural Network-Based Prediction and Optimization of Air Bubble Defects in Glass FiberReinforced Nylon 6 Injection Molding

Abstract Injection molding is a widely used manufacturing process, but defects such as air bubbles significantly impact product quality and mechanical properties. Traditional optimization techniques rely on trial-and-error methods, which are time-consuming and inefficient. This study presents a neural network-based approach for predicting and optimizing air bubble defects in glass fiber-reinforced Nylon 6 injection molding. A dataset of key process parameters—including nozzle temperature, barrel temperature, injection speed, pressure, cycle time, and cushion size— was collected through a Design of Experiments (DOE) approach. A feed forward neural network was trained to predict defect size based on these parameters, achieving an R-squared accuracy of 0.92. To further refine process conditions, an optimization algorithm was applied to minimize air bubble formation. The optimized parameters resulted in an 80% reduction in defect size, demonstrating the effectiveness of neural network-based process control. This approach provides a data-driven framework for improving product quality and reducing waste in injection molding.