en.Wedoany.com Reported - During the 2026 World Artificial Intelligence Conference, China Unicom launched the co-construction of the "Industrial Master Craftsman" experience dataset, converting the operational expertise of senior technicians into machine-readable corpus to accelerate the intelligent transformation of the manufacturing industry.
On July 17, China Unicom held the "Model-Data Synchronization, Intelligent Manufacturing Resonance" AI-empowered New Industrialization Development Forum, focusing on the sedimentation of tacit knowledge in manufacturing and the construction of high-quality industrial data, initiating this dataset co-construction project. The forum pointed out that the global industrial intelligent manufacturing market is expected to exceed 3.5 trillion yuan by 2035, with Siemens, BMW, and others fully advancing AI-native factory transformations. Chinese manufacturing enterprises generally face the challenge that tacit experience, such as welding judgment, equipment maintenance, and process parameter adjustment, which relies on the practical perception of master craftsmen, is difficult to convert into data corpus. Moreover, with the retirement of senior technicians and the contraction of frontline talent, core know-how is at risk of loss. The traditional master-apprentice model struggles to meet the training needs of large-scale industrial large models, embodied intelligence, and industrial intelligent agents, while differences in operational methods among different teams raise the standardization threshold for datasets.
This co-construction involves 13 entities, covering equipment, shipbuilding, nuclear power, heavy industry, and "AI+Manufacturing" service providers, including Shanghai Electric, ZPMC, Waigaoqiao Shipbuilding, Shanghai Aircraft Manufacturing, China Unicom Shanghai Branch, and Kupasi Technology. The parties are tackling ten categories of high-value experience datasets, including equipment diagnosis, welding defect judgment, ship design review, and human-machine collaboration behavior. ZPMC has collected massive multimodal data from gas-shielded welding of steel bridge project plate units. After tuning large and small models, the recognition accuracy for porosity, undercut, and lack of fusion defects exceeds 98%. Waigaoqiao Shipbuilding's "AI R&D Design Assistant" has shifted shipowner and classification society comment analysis from passive response to intelligent assistance. Shanghai Electric's Shanghai Turbine Plant has piloted a blade process generation intelligent agent, which automatically generates processes, action sequences, and key parameters through 3D model feature extraction and process corpus learning, reducing the design cycle from 30 days to 14 days.
The Shanghai Municipal Commission of Economy and Informatization has introduced supporting policies. The latest "Several Measures to Further Promote the Development of 'AI+Manufacturing' in Shanghai" includes specific clauses to support the development of master craftsman experience data governance methodologies and the creation of open-source collection toolchains. After the dataset is implemented, industrial robots and intelligent agents can replicate the judgment logic of master craftsmen. New employee training in enterprises can shorten the growth cycle using digital corpus, and process knowledge can continuously iterate industry-specific large models, forming a sustainable intelligent manufacturing data loop.
