Stories of High-Quality Development | 'AI chef' revolutionizes special steel making

Narrators: Li Jinyan, Director of AI Research Institute, Nanjing Iron & Steel Co., Ltd; Shi Mao, President of Steel and Non-ferrous BU, Huawei

When AI models are applied to industrial scenarios with stricter requirements and lower fault tolerance, can they still be effective?

At Nanjing Iron & Steel Co., Ltd., the adoption of "AI chefs" transformed the traditional research and development (R&D) paradigm by replacing experience-based trial and error with intelligent, data-driven decision-making in steel production.

Take carbon-manganese low-temperature steel, a key material in manufacturing high-end vessels such as liquefied gas carriers (LPG/NH3 ships), as an example. During the material R&D process, too much carbon makes the steel brittle, while excessive manganese affects welding performance, with dozens of variables influencing one another. In light of this, we came up with an idea of leveraging large AI models to discover "recipes" for steel making. However, challenges such as the complexity of the steel manufacturing process, diverse and fragmented data modalities, and the difficulty in eliminating AI models' "hallucinations" became key obstacles to overcome.

Instead of building a versatile AI model covering all the scenarios and processes, we ultimately adopted an approach of training a group of specialized scenario-specific models with domain expertise. Some focused on researching whether steel could meet strength design requirements, while others studied the low-temperature toughness of steel materials, allowing them to spontaneously collaborate, thereby achieving the prediction and calculation of various performance indicators.

It can be said that digitalization makes "data visible," while intelligence enables "better decision-making." Today, industrial AI models have been applied across multiple industry scenarios including home appliance manufacturing and cold chain logistics, empowering various industries and helping them progress from "seeing clearly" to "performing better" and "thinking innovatively."