"Prof. Fabio Miani University of Udine, GB/Z Standards on AI 4 Steel — for my students"
A respectful technical overview and educational roadmap on the newly published Chinese GB/Z National Technical Guidelines for Artificial Intelligence in Steel Manufacturing.
Executive Summary & Welcome Address
Dear Colleagues from East Asia, esteemed academic partners from Northeastern University (NEU, Shenyang), POSTECH (Pohang), and Kyushu University, as well as our distinguished alumni from the Master in Metallurgical Engineering at the University of Udine and early arrivers of the AIST European Steel Forum 2026:
It is my profound privilege to welcome you to this introductory dialogue. As we stand on the threshold of the AIST European Steel Forum 2026—hosted here in Udine in close collaboration with Danieli—we observe an epochal paradigm shift in global metallurgy. The integration of Industrial Large Language Models (LLMs), Multimodal AI, and Autonomous Intelligent Agents into ironmaking, steelmaking, and rolling processes is no longer speculative; it is actively being codified into national engineering standards.
Purpose of this Discussion Paper
- To offer a respectful technical breakdown of the newly published Chinese GB/Z National Technical Guidelines on AI for Steel.
- To build an educational bridge for our University of Udine metallurgical engineering students and young industry researchers.
- To prepare the ground for global East-West academic cooperation prior to our gathering at the Castle of Udine dinner.
The Emergence of GB/Z Guidance Documents
In the Chinese standardization ecosystem managed by SAC (Standardization Administration of China), GB/Z denotes National Technical Guidance Documents. Unlike mandatory GB standards or recommended GB/T standards, GB/Z guidelines serve as rapid-response frameworks designed for fast-evolving technological frontiers where consensus is forming.
Traditional Automation (Legacy)
Deterministic Level 1 / Level 2 PLC models, static physical-chemical equilibrium tables, human-in-the-loop manual adjustments, fragmented sensor silos.
Industrial Foundation Model Era (GB/Z)
Cross-domain multimodal foundation models, closed-loop process intelligence, real-time edge-cloud coordination, and multi-agent dynamic optimization.
Deep-Dive: The First 3 GB/Z AI Technical Guidelines for Steel
China's authorities have issued three seminal national technical guidelines aimed specifically at structuring AI deployment across metallurgical plants. Below is an overview tailored for our academic discussion and student study:
Dataset Requirements for Steel Industry Large Models
Establishes mandatory protocols for data hygiene, tokenization, time-series alignment, and security in high-temperature, harsh industrial environments.
- Multimodal Ingestion: Standardizes data streams from Blast Furnace (BF) thermal imaging, Electric Arc Furnace (EAF) optical emission spectroscopy, continuous casting vibration sensors, and cold rolling surface inspection cameras.
- Data Curation & Privacy: Provides unified schema for handling proprietary metallurgical recipes, operational logs, and trade-secret alloy compositions while enabling safe model fine-tuning.
System Architecture of Industrial Foundation Models for Steel
Defines the reference computing topology needed to bridge enterprise cloud computing with mill-floor millisecond-level execution.
- Edge-Cloud Synergy: Heavy foundational model pre-training occurs on centralized cloud/data centers, while lightweight distilled models run on edge servers near the production line.
- Deterministic Control: Guarantees sub-10ms response times for critical safety and quality parameters during high-speed rolling and liquid metal pouring.
Reference Architecture & Agent Applications in Steel Manufacturing
Focuses on operational deployment of specialized AI Agents capable of reasoning, predictive maintenance, and autonomous carbon emission minimization.
- Autonomous Process Control: AI Agents dynamically adjust slag chemistry, oxygen blowing schedules, and furnace energy inputs to compensate for raw material quality fluctuations.
- Green Steel & Decarbonization: Models optimize hydrogen injection parameters and scrap metal ratios to minimize carbon intensity without compromising mechanical properties.
Strategic Implications for Metallurgical Education
For our students at the University of Udine Master in Metallurgical Engineering, these developments signify that metallurgy is rapidly evolving into a hybrid discipline combining thermodynamics, physical metallurgy, and data science.
New Curriculum Paradigm
Integrating industrial data hygiene and neural network calibration into standard metallurgical thermodynamics coursework.
East-West Standards Dialogue
Fostering standard interoperability between Asian manufacturing centers (China, Japan, Korea) and European equipment builders like Danieli.
Decarbonization Enabler
Utilizing AI foundation models to accelerate the transition to DRI-EAF zero-carbon steel production models globally.
Open Discussion Topics & Question Framework
To guide our morning online panel with NEU, POSTECH, and Kyushu University, as well as our afternoon interactive session with student alumni, I submit the following structured questions:
For NEU (Shenyang), POSTECH, and Kyushu University Colleagues:
"How do your academic institutions evaluate the trade-off between black-box deep learning agents and deterministic physical-chemical metallurgical models when establishing standards for critical mill operations?"
For University of Udine Master's Alumni & AIST Participants:
"In what ways can European equipment manufacturers and plant operators leverage standard datasets like GB/Z 231-2026 to ensure seamless digital integration in international greenfield and brownfield projects?"