banner

How to Build Industrial Datasets? Luculent Shares Comprehensive Implementation Experience

Industry trends 2026.07.31

English Translation

On July 29, the special seminar on the construction of high-quality datasets for the high-quality development of the new power equipment and new energy industry in Jiangsu Province was successfully held in Zhenjiang. Li Li, Deputy Director of the Information Technology Development Department of the Ministry of Industry and Information Technology (MIIT), Chi Yu, Deputy Director of the Jiangsu Provincial Department of Industry and Information Technology, Yin Weimin, Vice Mayor of Zhenjiang Municipal People’s Government, Zhou Ping, Chief Engineer of the Electronic Information Research Institute of MIIT, Fan Ji’an, Chief Scientist of Big Data of China Unicom, and Zhang Guopeng, Deputy General Manager of China Unicom Jiangsu attended the event. Also present were relevant persons-in-charge from the Artificial Intelligence & Big Data Industry Division and the New Energy Industry Division of the Jiangsu Provincial Department of Industry and Information Technology, over 80 representatives including member entities of the pilot consortium for high-quality dataset development for the new power equipment and new energy industry, key enterprises in industrial clusters, relevant digital service providers and industrial organizations.

Luculent was invited to participate in the seminar. Xu Mian, General Manager of the Ecological Cooperation Headquarters of Luculent, delivered a keynote speech entitled Pioneering Practices and Integrated Application of Industrial Data Infrastructure in the Coal Machinery and New Energy Equipment Sectors. He systematically analyzed the pain points in building high-quality industrial datasets, the comprehensive implementation framework and benchmark industrial practices.

“A high-quality dataset serves as premium raw material for industrial AI. Without unified standards, dedicated tools, adapted models and implementable methodologies, data remains merely ineffective figures trapped in information systems,” Xu Mian noted in his speech. Drawing on practical experience accumulated across multiple industries including coal mines and wind power, Luculent has developed an integrated framework consisting of one set of high-quality datasets, a portfolio of practical and specialized models, a full toolchain for data collection, aggregation and utilization, and a complete digital-intelligent innovation methodology. Centered on business scenarios, the framework establishes a cyclic optimization mechanism linking data, models and applications to smooth the whole chain of data value conversion.

One Set of High-Quality Datasets: Building the Underlying Raw Material Repository for AI

Accurate fault prediction and production optimization powered by industrial AI hinge on clean, standardized and multi-dimensional data. Luculent adopts a dataset development model featuring top-level planning at the group level and practical governance at the plant level.At the group level, unified systems for data management, security and sharing are formulated, a dedicated data governance organization is established, and universal standards covering data collection, transmission and sharing are issued to break data silos among various plants and subsidiaries and unify data specifications across the group.At the plant level, targeted specifications for segmented data access are implemented focusing on production, equipment, environmental protection and safety operations, aggregating multi-modal data including time-series operational data, offline inspection data, video footage and specification documents.

This tiered framework functions like a standardized warehousing system: the group defines unified data entry rules, while individual plants categorize and accumulate equipment operating condition, fault, maintenance and process data as needed. It enables unified data traceability and ensures data usability, guaranteeing dataset quality from the source.

A Portfolio of Practical and Specialized Models: Developing Scenario-Specific Intelligent Brains

Generic models decoupled from industrial process mechanisms are prone to false alarms and inconsistent with on-site reality. Adhering to scenario-driven development and synergy between data and mechanism, Luculent builds a matrix of practical and specialized models tailored to real production scenarios. It integrates mechanism-based small models, vertical industrial large models and dedicated intelligent agents to cover business scenarios such as equipment health management, safe production, process optimization and industrial supervision.

Small models deliver precise identification of physical equipment failures such as bearing wear and gear degradation. Vertical industrial large models assimilate massive operation and maintenance cases and equipment specifications to automatically conduct root cause fault analysis and generate maintenance solutions. Supporting intelligent agents undertake end-to-end tasks including data verification, fault reasoning and work order preparation, striking a balance between professional accuracy and practical implementability.

A Full Toolchain for Data Collection, Aggregation and Utilization: An All-in-One Pipeline for Data Processing

Luculent has independently developed 12 multi-modal data toolkits, forming a complete “collection – fusion – modeling” toolchain equivalent to an automated industrial data production line:

  • Collection: Leveraging edge gateways and link monitoring tools to support communication protocols of all types of equipment and steadily collect high-frequency real-time data from wind farms and underground coal mines.

  • Aggregation: Adopting cleaning, restoration, annotation and vectorization tools to automatically process heterogeneous raw data, completing deduplication, desensitization and sample standardization.

  • Utilization: Equipped with modeling tools for mechanism simulation, artificial intelligence and intelligent agents, embedded with more than 300 industrial modeling components to support online model training and remote deployment.

Fully compatible with cloud-edge-end architecture, the tool suite features automatic operating condition classification and sample verification capabilities, significantly streamlining enterprise data governance and model development workflows.

A Complete Digital-Intelligent Innovation Methodology: Standardized Implementation Guidelines

High-quality dataset development is not a one-off project and requires long-term iterative operation. Luculent has developed a full engineering implementation methodology, forming a standardized workflow of application planning → industrial knowledge aggregation → model construction → application development.

Starting from business demands, the methodology promotes collaboration between OT process specialists and IT technicians. On one hand, professional knowledge bases are built by sorting out equipment mechanisms, industrial standards and experience of operation and maintenance experts. On the other hand, multi-modal data fusion and reasoning are realized through large model fine-tuning, knowledge graphs and hybrid retrieval. Supporting long-term operational mechanisms for data governance and cross-organizational authority management are put in place to foster a positive loop connecting data, models and business scenarios, preventing datasets from being left idle after construction and failing to generate value.

During the sharing session, Xu Mian presented tangible cases in the coal machinery and wind power sectors to demonstrate the industrial empowerment brought by high-quality datasets.

In the coal machinery sector, addressing challenges including high underground operation and maintenance costs, shortage of skilled technical personnel and difficulties in inheriting fault-handling experience, Luculent has built a full-lifecycle multi-modal dataset covering more than ten types of equipment such as shearers and scraper conveyors, and developed intelligent agents for early warning, maintenance and knowledge Q&A. The solutions have been deployed for leading coal enterprises including China National Coal Group, Huadian Coal Industry and Tiandi Benniu, connecting over 30,000 pieces of equipment. The system enables automatic fault diagnosis and intelligent generation of maintenance schemes, supporting equipment manufacturers in transforming toward “manufacturing plus services”.

Selected for national pilot program! The consortium led by Luculent has obtained recognition from MIIT.

In the wind power sector, to tackle obstacles such as diverse wind turbine manufacturers, fragmented data and difficult deployment of AI early warning solutions, Luculent has formulated three major industrial standards for unified measuring point naming, fault coding and unit status codes. It integrates SCADA and CMS monitoring data to build a comprehensive wind turbine dataset, and develops models for gearbox and pitch system fault prediction as well as unit energy efficiency optimization. The solutions help new energy enterprises build proprietary data centers, unlock value from wind turbine operational data, cut manual maintenance costs and boost power generation efficiency of wind farms.

Among the first provincial-level pilots: Luculent takes the lead in developing the “fuel repository” for wind power operation and maintenance AI.

Xu Mian stated that Luculent will continuously leverage the strengths of its national cross-industry and cross-region industrial internet platform. Building on 27 years of experience in industrial digitalization, the company will further refine its four-component dataset development framework. Luculent will deeply engage in the construction of the Jiangsu Provincial Pilot Consortium for New Energy Industry Datasets, consolidate the data foundation for local new power and new energy equipment enterprises, and drive the intelligent and green transformation of manufacturing through data elements.


Return to List Page
Share on WeChat Moments ×
Open WeChat and click on "Discover" at the bottom,
Use 'Scan' to share web pages on your social media circle。