Improving Industrial Injection Molding Processes with Explainable AI for Quality Classification

Explainable AI makes industrial quality control more practical: The paper demonstrates that XAI techniques (SHAP, Grad-CAM, and LIME) can reveal which factors actually matter for quality decisions, helping manufacturers understand why an AI system rejected a part rather than just accepting its verdict as a "black box."

You can have your cake and eat it too—fewer sensors, better performance: By reducing input features from 19 to as few as 6 using XAI analysis, the researchers maintained high classification accuracy while actually improving how well the model generalizes to new data, suggesting that simpler systems can be more reliable than complex ones.

Interpretability enables adoption in resource-constrained settings: Many factories lack comprehensive sensor networks, which has historically prevented AI deployment. This work shows that XAI-guided feature reduction makes quality control AI feasible even for manufacturers with limited sensor technology, democratizing access to these tools.

Faster inference speeds support real-time manufacturing: The reduced feature sets improved inference speed, meaning quality decisions happen quicker on the production line—critical for catching defects before they propagate and for maintaining manufacturing throughput.

Responsible AI requires transparency about trade-offs: The paper explicitly evaluates the balance between accuracy, speed, and interpretability rather than optimizing for just one metric, modeling the kind of thoughtful decision-making needed when deploying AI systems in high-stakes industrial environments.