Multi-quality prediction of injection molding parts using a hybrid machine learning model

With the advantages of high efficiency and low manufacturing cost, injection molding is a primary method of polymer processing. However, comprehensive inspection of part quality is limited due to high costs of time, labor, and equipment, often hindering quality control. There is an urgent need to develop a rapid and low-cost inspection method that can perform various quality inspections on injection-molded parts. Accordingly, this study proposes a virtual measurement technique based on a multi-quality prediction neural network that combines with an autoencoder network (AE) and a multilayer perceptron network (MLP). The research focused primarily on extracting and reducing the dimension of captured data using machine perception, quality index, and automatic feature extraction technologies to aid the rapid training of a hybrid AE/MLP model. Experimental case studies demonstrated that the method instantly predicted the residual stress distribution, weight, and geometric dimensions of plastic parts, and the model prediction error (root mean squared error) was less than 5% of the total tolerance. In particular, the predicted residual stress distribution was highly similar to the actual image, providing a substitute for the actual measurement of the residual stress within the molded part.