Development of a Vision-Image-Based Quality Prediction Neural-Network Algorithm for an Injection Molding Machine Considering Cavity Sensor and Vibration Data

Abstract This research develops a neural-network-based algorithm for predicting the quality levels of injection molding products to handle quality regarding problems caused by occurrences of low-quality products. With an assumption that vibration data and temperature and pressure data of the cavities in mold generated from the injection molding machine would be different according to the products’ quality, main objective of this research is to predict the quality grade for each product utilizing the vibration data of the machine and temperature and pressure data of each cavity collected during a product is processed. Among diverse features that can represent quality of injection molding products, we especially focus on the features that could be driven from vision images of the products. We firstly explain how the infrastructure is constructed for collecting the vibration data, cavity sensor data, and vision-image data. Then, for the vibration and cavity sensor data, statistical features that stand for specific patterns of each data utilized as independent variables are derived. Quality grades of each product are also distinguished by two indicators: flection of the product’s housing and alignment of pinholes derived from the vision images of products utilizing the Canny-Edge algorithm. Finally, utilizing statistical features of the vibration and cavity sensor data as independent variables and distinguished quality grades of each product as dependent variables, a neural-network-based quality prediction algorithm is developed, and the performance of the algorithm is evaluated.