Application of AI-based Intelligent Control Methods for Enhancing Product Quality in Glass Manufacturing
Omar Musazade *
Department of Instrument Engineering, Azerbaijan State Oil and Industrial University, Baku, Azerbaijan.
Stanislav Aghamatov
Department of Instrument Engineering, Azerbaijan State Oil and Industrial University, Baku, Azerbaijan.
*Author to whom correspondence should be addressed.
Abstract
Aims: This research aims to advance industrial glass manufacturing by implementing a next-generation autonomous control architecture. The study focuses on maximising energy efficiency and product quality by integrating Digital Twin technology, Deep Reinforcement Learning (DRL), and Explainable AI (XAI) to overcome the limitations of legacy control systems.
Study Design: This is an analytical and simulation-based research study focused on the cognitive optimisation of industrial float glass production processes.
Place and Duration of Study: The study was conducted in the Department of "Instrumentation Engineering" (Intellectual Measurement and Control Systems), Azerbaijan State Oil and Industry University (ASOIU), between September 2025 and June 2027.
Methodology: A high-fidelity Digital Twin of a float glass furnace was developed to simulate production dynamics. A DRL-based agent was implemented for real-time furnace regulation, allowing for continuous self-optimisation of thermal zones. The system integrated a Convolutional Neural Network (CNN)-based visual inspection layer for defect detection, complemented by an XAI module that provides transparent, logic-based justifications for autonomous operational adjustments. Theoretical evaluations of hydrogen-natural-gas blending were also conducted to assess sustainability impacts.
Results: The proposed DRL-driven architecture achieved improved thermal regulation, maintaining stability within ±0.5°C compared with traditional Fuzzy-PID models. The combustion strategy, optimised via reinforcement learning, produced a reported reduction in fuel consumption of more than 6% (P < 0.05). The XAI module provided real-time interpretability of system decisions, while the integration of hydrogen-enriched combustion pathways indicated a potential decrease in carbon emissions of approximately 8-10% and maintained high combustion efficiency compared with standard natural gas use.
Conclusion: The transition from localised control to an AI-driven autonomous framework, supported by Digital Twins and Explainable AI, provides a transparent approach to modernising glass manufacturing. These systems may reduce operational risks and environmental impacts, thereby supporting intelligent and sustainable industrial automation.
Keywords: Artificial intelligence, float glass manufacturing, digital twin, deep reinforcement learning, explainable artificial intelligence, convolutional neural network, thermal control, defect detection, energy efficiency, intelligent automation