Feature extraction and physical interpretation of melt pressure during injection molding process

Melt pressure is crucial in injection molding. A variety of pressure sensors have been installed in injection molding machines and molds to collect melt-pressure information. Many methods have been developed to extract the features of melt pressure. However, challenges exist for these feature extraction methods because they are, inevitably, extremely dependent on manual operation. In this study, an unsupervised feature extraction method using a sparse autoencoder is proposed to extract the features of melt pressure during injection molding. An injection molding model was applied to reinterpret the network structure of a sparse autoencoder to better understand the network structure. The feature curve, which is defined as the curve of weights between input and hidden units, was proved to be a significant indicator to measure how melt compression causes variations in injection pressure. Over 10,000 shots were conducted to verify the proposed feature extraction method. The experimental and simulation results show that the feature curve effectively extracts switching points (gate and velocity-pressure switchover), important indicators (viscosity index, holding pressure, and peak pressure) in the injection molding process and shows how the geometric features of a part affect the melt flow.