NISCO Deploys World's First Swarm-Intelligence Steel Model at Scale
Nanjing Iron & Steel transforms manufacturing through a dual-drive, multi-model AI architecture.
1Swarm Intelligence in Steel
Nanjing Iron & Steel Co. launched the YuanYe Steel Large Model, the first of its kind using swarm intelligence. By integrating Huawei’s Ascend and Kunpeng infrastructure, the company transitioned from experience-based operations to data-driven intelligence. This architecture fuses metallurgical mechanism models with probabilistic computing, creating a dual-drive system for complex manufacturing environments.
2Dual-Brain Model Architecture
The YuanYe model features a unique dual-brain design. The right brain manages probabilistic computing, while the left brain handles precise, scientific decision-making. Aggregating data from millions of operational technology points across 26 business domains, the system enables real-time digital twin mapping, allowing for precise control and intelligent assessment of the entire steelmaking process.
3Blast Furnace Proactive Control
In the ironmaking stage, the model analyzes internal sensor data to perform real-time dynamic simulations. Operators can now identify furnace health issues two hours in advance with over 90 percent accuracy. This shift from reactive remediation to proactive pre-control has increased prime-grade molten iron yield by more than 10 percent annually.
4Precision Rolling Performance
The rolling-force prediction model achieves millisecond-level response times to manage dynamic operating conditions. By narrowing thickness tolerance to ±0.2 mm and reducing deviation to 1.9 percent, NISCO reached world-leading standards. The rolling-force hit rate improved from 88 percent to 98 percent, supporting an annual output exceeding 2.5 million tons of steel plates.
5Automated Metallographic Inspection
The smart metallographic inspection system utilizes robots and AI to automate sample grinding, polishing, and image acquisition. Inspection accuracy now exceeds 95 percent, while per-sample processing time dropped from 10 minutes to 3–5 minutes. This 80 percent efficiency gain allows human operators to focus on anomaly analysis rather than repetitive manual tasks.
6Environmental Compliance Optimization
The S-Link Intelligent Agent manages gas and flue gas sulfur control by integrating 602 process parameters. It issues sulfur-risk warnings one hour in advance with 95 percent accuracy. By dynamically allocating emission quotas via game-theoretic models, the system slashed natural gas co-firing consumption by 50 percent and reduced emission-related downtime by 25 percent.
7Economic and Operational Impact
NISCO has achieved full digital connectivity for all key equipment, shortening R&D cycles by 50 percent. These operational improvements have reduced annual costs by over CNY 100 million. By October 2026, the company launched 25 business agents to handle diverse tasks, ranging from production scheduling to complex operations management across the plant.
8Human-Centric AI Paradigm
NISCO views digital transformation as a human-centric endeavor aimed at unleashing employee potential. The company has deployed over 300 AI models and 1,000 low-code agents created by frontline staff. This scalable integration across production and management sets a new global benchmark for the intelligent transformation of the traditional steel manufacturing industry.