Huawei and Nanjing Iron & Steel say their YuanYe industrial AI system is now deployed across multiple steelmaking operations, with reported gains in prediction, inspection and energy use. It is a concrete example of AI moving from pilot language into production, though the performance figures come from the companies’ account.
Huawei AI Watch analysis
What happened
NISCO and Huawei developed YuanYe with ecosystem partners. The system combines AI models, metallurgical models and operational data, drawing on millions of data points across 26 business domains, according to Automation World’s report.
The report says blast-furnace models forecast temperatures and identify problems up to two hours ahead, with reported accuracy above 90%. In rolling, a prediction model responds in milliseconds; NISCO says its hit rate rose from 88% to 98%. Automated metallographic inspection is reported to exceed 95% accuracy and cut sample inspection time from ten minutes to three to five.
The companies also say a gas and flue-gas control agent predicts sulfur risks an hour ahead, nearly halves natural-gas co-firing and reduces emissions-related downtime by more than 25%. By October, NISCO had launched 25 business agents, the report says.
Why it matters
These are production tasks with direct consequences for steel quality, energy use and plant operations, not a general-purpose chatbot being given a hard hat. The account describes AI models working alongside industrial data and metallurgical mechanisms, with people still responsible for higher-level decisions.
The claimed gains are substantial, but they are company-reported results rather than independently assessed comparisons. That makes the deployment noteworthy; it also makes clear evaluation methods and results from other plants important next evidence.
Our read
This is one of the more tangible industrial-AI stories: specific processes, operational uses and reported outcomes. The strongest test will be whether the gains hold up beyond NISCO’s sites and can be reproduced without turning every factory into a bespoke integration project. Steelmaking is a demanding place to make a model earn its keep.
What to watch
- Whether NISCO publishes independent evaluations or methods behind the reported accuracy and efficiency figures.
- Whether YuanYe expands to more processes or sites, and what results follow.
- Whether the companies proceed with their stated plans to explore dedicated data centres for smart steel plants.
Discussion spark: For industrial AI, should companies have to publish independently checked performance results before describing a deployment as a success, or are operational results reported by the operator enough to make the case?
Sources and evidence
not affiliated with or endorsed by Huawei AI