Transforming Into a Data-Driven Smart Factory via the 'AI Temperature Prediction & Control Model'
Conquering 122 Million Cumulative Data Points to Achieve Autonomous Operation, Slashing Fuel Costs, and Boosting Workforce Flexibility
South Korea’s glass bottle manufacturing sector is evolving into a highly advanced smart factory ecosystem by embedding artificial intelligence (AI), a cornerstone of the Fourth Industrial Revolution. Through the successful development of the “AI Temperature Prediction and Control Model” at the Gunsan Plant, the facility has achieved remarkable milestones—optimizing furnace temperature control while significantly boosting overall operational efficiency.
■ Shifting the Core: From Veteran Intuition to AI-Driven Systems
Traditionally, furnace operations heavily relied on the experience and intuition of veteran workers to manage energy input tailored to specific product lines. However, this human-dependent approach led to variations in energy efficiency across shifts and hindered overall operational productivity, as operators could rarely leave their posts due to the ultra-high-temperature environment.
To overcome this bottleneck, the Gunsan facility benchmarked POSCO’s renowned smart blast furnace model and collaborated with AI specialists to build a cutting-edge AI system specifically customized for glass bottle manufacturing.
■ Achieving Stable Control Through a Rigorous Three-Phase Evolution
The development of this AI model has been executed with meticulous precision through sequential, data-driven stages since 2022:
Phase 1 (Prediction Model Development): Analyzed the melting process and trained the AI on 27 million data points to develop and visualize predictive models for critical variables, including internal temperature and oil flow rates.
Phase 2 (Model Optimization): Expanded the training dataset to 81 million data points, integrated real-time booster power metrics, and utilized OCR (Optical Character Recognition) technology to digitize dashboard gauges, drastically improving prediction accuracy.
Phase 3 (Stabilization & Integration): Leveraging a massive cumulative database of 122 million data points, the plant implemented PLC (Programmable Logic Controller) integrated operations, successfully achieving up to 4 hours of continuous autonomous operation and establishing a seamless real-time control architecture.
[Source: Smart Manufacturing Innovation Report 2026 / Deployed AI System & Quality Data Measurement Device]
■ Driving Both Energy Efficiency and Superior Quality
The impact of the AI system is clearly proven by empirical data. Most notably, temperature fluctuations have been drastically minimized. Under conventional manual control, the furnace frequently drifted outside the optimal target temperature of 1,370°C. In contrast, the new AI system leverages predictive data to proactively adjust to conditions 5 minutes in advance, maintaining steady operations strictly within the target range. This stability has successfully reduced fuel costs, extended the operational lifespan of the furnace, and sharply suppressed product defects.
Furthermore, the workplace environment has undergone a massive transformation. Daily temperature monitoring tasks—which previously required up to 3 hours of intense labor—have been minimized, allowing the plant to optimize shift staffing from 3 operators down to 2, thereby securing exceptional workforce flexibility.
■ Overcoming Key Challenges and Future Outlook
Because glass manufacturing involves simultaneous raw material melting and fuel combustion, any failure in temperature prediction carries the risk of large-scale manufacturing defects. The Gunsan team conquered these sophisticated engineering hurdles by successfully decoding the complex correlations between temperature sensors and burners, alongside resolving uncertainties in booster operation patterns.
“This is a profound evolution into a data-driven precision manufacturing system that goes far beyond simple automation,” a plant official stated. Moving forward, the facility plans to further advance the model by integrating it with real-time quality measurement data—such as monitoring for bubbles and seed defects—to systematically unlock the absolute optimal operating conditions.