Recipe Based Anomaly Detection with Adaptable Learning: Implications on Sustainable Smart Manufacturing

Area of Science:

• Manufacturing Engineering
• Artificial Intelligence
• Quality Control

Background:

• Industry 4.0 drives AI adoption in manufacturing, but challenges remain in handling diverse, irregular datasets.
• Current injection molding quality inspection at the batch level limits individual product defect identification and AI application.
• Existing AI models struggle with varied manufacturing settings, leading to inaccurate quality assessments.

Purpose of the Study:

• To propose a novel anomaly detection framework for injection molding processes to enhance product efficiency and quality inspection.
• To address limitations in AI implementation caused by diverse manufacturing settings and batch-level inspection.
• To improve defect prediction accuracy and enable continuous, data-driven process optimization in smart factories.

Main Methods:

• Recipe-Based Learning: K-Means clustering classifies injection molding data into setting-specific recipes, ensuring data normality.
• Statistical validation using the Kruskal-Wallis test confirms the necessity of recipe-based classification for varying settings.
• Anomaly detection using Autoencoders trained on normal data per recipe; KL-Divergence for adaptable learning with unseen settings.

Main Results:

• The proposed AI framework predicted 61 defects, significantly outperforming the existing 41 defects and an integrated model's 2 defects.
• Adaptable Learning outperformed integrated and additionally trained models, achieving continuous prediction for new settings without retraining.
• The data-driven approach demonstrated superior quality inspection and enhanced process management capabilities for smart factories.

Conclusions:

• The novel anomaly detection framework effectively handles diverse manufacturing settings in injection molding, improving defect prediction.
• Recipe-Based Learning and Adaptable Learning offer a robust AI solution for enhancing quality control and process optimization.
• This AI-driven approach significantly boosts productivity and decision-making in smart manufacturing environments.