Machine Learning Approaches for Defect Detection in Injection Molding: A Comprehensive Review of Methods, Challenges, and Industrial Integration

Abstract

Injection molding has become indispensable in the mass production of thermoplastic components, serving key roles across automotive, electronics, and consumer goods industries. However, guaranteeing consistent part quality remains a significant challenge due to the intricate and variable nature of process parameters. Recent advancements in machine learning (ML) and artificial intelligence (AI) have revolutionized defect detection in injection molding, enabling more adaptive, predictive, and interpretable quality assurance systems. This review paper synthesizes the latest research on ML-based defect detection, with a focus on supervised learning algorithms, statistical distance-based modeling, and deep generative modeling techniques. We analyze findings from studies employing classifiers such as Support Vector Machines, Random Forests, Light GBM, and neural networks, as well as unsupervised approaches like Mahalanobis Distance and variation Auto Encoders. Special attention is given to model interpretability, with methods such as SHAP (Shapley Additive Explanations) enhancing transparency and actionable insights for process engineers. Results from recent benchmarking efforts using real-world datasets, such as the Korea AI Manufacturing Platform, demonstrate that model performance is highly dependent on specific part types, data distributions, and the presence of class imbalance. Hybrid models and transfer learning approaches are shown to further boost predictive accuracy and generalizability. Despite these advances, challenges persist in dataset standardization, real-time deployment, and robust model adaptation across diverse manufacturing contexts. This review provides practical recommendations for selecting and deploying ML models in Industry 4.0-enabled injection molding, supporting the ongoing digital transformation of quality control in smart manufacturing.