Neural Network Application for Short Shot Prediction in Injection Molding

With the advancement of technology, many injection molding factories currently employ Automated Optical Inspection (AOI) technology to detect product defects. However, the high cost associated with establishing AOI inspection mechanisms has led smaller-scale injection molding factories to continue relying on manual methods for product quality inspection, hindering the overall industry's ability to continuously improve. In this study, we utilized edge computer to collect machine data during the injection molding process, which were then stored in a database. The collected data underwent correlation analysis to identify parameters that are more closely related to short shots, and subsequent data processing was performed. We then employed TensorFlow to build a neural network classification model for training purposes, achieving an accuracy of 91.2% in predicting short shots. Furthermore, we employed SHAP (SHapley Additive exPlanations) as a model interpretation tool, which allows us to utilize the analysis results of SHAP as a basis for troubleshooting when short shots occur. By inputting real-time machine data into the model to predict product quality defects, injection molding factories can establish automated product quality inspection mechanisms at a lower cost, thereby enhancing inspection efficiency and reducing labor costs.