Data-Driven Modeling of Quality-Related Multiple Indicators in Injection Molding: An Integrated JMI-MSVR Technology

Abstract

Injection molding (IM) has been widely used in modern industrial production, which is a typical complex industrial process with multiple indicators of final products quality. Establishing the quantitative relationship between process parameters and multiple quality indicators is vital for the improvement of IM process. However, the actual IM process collects massive data of high-dimensional process parameters, resulting in high redundancy and irrelevant information in the raw dataset. Therefore, it has become an urgent problem to identify the critical process parameters from these raw data and establish a multi-output model that captures the coupling relationship between critical process parameters and multiple quality indicators. This study proposes a data-driven technology for quality-related multiple indicators in IM process, which elegantly integrates joint mutual information (JMI) and multi-output support vector regression (MSVR). Firstly, the critical process parameters of IM process are selected by a JMI-based sequential automatic search algorithm. This algorithm not only effectively filters out redundant information from the raw industrial dataset, but also provides the essential input features for training MSVR. Then, MSVR is employed to establish the quantitative relationship between critical process parameters and multiple quality indicators of IM process, where its hyperparameters is fine-tuned by the randomized search approach. Finally, this integrated JMI-MSVR technology is tested by the experiment on an industrial dataset of IM process.

Keywords: Industrial big data, intelligent manufacturing, quality modeling, critical parameters selection, support vector regressions