In recent years, the Nuclear Power Plant Repair Service at the Tianwan NPP in China’s Jiangsu province has been promoting the integration of artificial intelligence (AI) with nuclear plant repair scenarios, relying on platforms such as Hezhi Longling. Currently, intermediate results have been achieved in such areas as personnel profiling, formation of work orders, compliance testing and auxiliary generation of repair plans for electric valve drives. A platform for intelligent control of lifting and rigging work and a robot for cleaning offshore engineering tunnels are also being developed.
In terms of personnel management, a seven-dimensional model for assessing personnel competencies has been created, establishing differentiated assessment indicators for various repair positions and forming a job profile based on data of employee competencies. The system is capable of handling historical data, supports storage of data for different periods and comparison of dynamics, and displays the distribution of staff competencies using radar charts, bar charts and trend graphs.
Once the data is loaded in batches, the system automatically performs normalisation, multivariate assessment and ranking, generating materials that include a general overview, job comparison, thematic analysis of responsibilities and recommendations for improvement, which serves as the basis for allocating resources for training job selection and differential management.
To solve the problem of long-term preparation of outfits for repair work and dependence on manual experience, a team based on the intelligent agent platform Longhuitong has developed an intelligent outfit training assistant, implementing a solution according to the scheme “One intelligent agent, one skill, seven categories of specialised knowledge bases”.
This system combines a database of historical outfits, a database of repair regulations, an equipment accounting database, a database of safety measures and risk management, a database of spare parts and tools, a personnel qualification database and a database of management procedures. Using historical data on work performed and the requirements of regulations, it can dynamically generate the content of the work package, automatically fill out the contents of that work package and issue a draft of the work package in Excel format available for download and interface with the Enterprise Project Management (EPM) system.
In terms of quality control of repairs at Tianwan NPP, an AI agent has been developed to verify the compliance of repair work. This agent, based on a large language model and search amplification technology (RAG – Retrieval-Augmented Generation), is capable of receiving repair regulations, lists of preventive maintenance programs and repair completion reports in batch mode, automatically analysing document content and conducting semantic comparison.
The system can gradually check the coverage of the stages of the repair procedure by the regulations, while simultaneously identifying problems such as missing points, data anomalies, lack of signatures and logical contradictions in completion reports. The structured verification report it generates makes it possible to trace each item to a specific procedure step or report line. It also contains recommendations for eliminating shortcomings.
Intelligent support is also beginning to be used when repairing key equipment. Taking into account the large number, variety of models and high requirements for the repair of electric valve drives at a NPP, Tianwan has developed a system for auxiliary generation of electric drive repair plans. After the engineer enters the equipment information and the fault description, the system refers to the electrical repair knowledge base, historical plans and equipment data and automatically generates a repair plan draft. This covers the analysis of causes, repair steps, technical features, preparation of spare parts and tools, and precautions. According to repair engineers, the plans are mainly suitable for use on site and can save preparation time during emergency repairs.
In addition to the applications already implemented, Tianwan NPP is also promoting the creation of an intelligent rigging management platform. Given the high specialisation and many stages of control when performing lifting operations at a nuclear plant, this platform provides support for both the web interface and API (Application Programming Interface). It also enables control of the development, training and monitoring processes of AI models.
After entering cargo parameters and site conditions, the platform will be able to help plan lifting operations, recommend equipment and sling points, warn about the risks of overload and collision. It will also be able to formulate a lifting plan based on information about the cargo, provide procedures, necessary materials and contact information for those responsible in emergency situations. It can be used to form a bank of educational questions, analyse records of work performed and prepare reports on eliminating deficiencies.
In future, the station will continue to expand scenarios for the use of AI in the field of repair, taking into account industrial practice, helping to increase the level of digitalisation and intellectualisation of repair activities. This will strengthen cybersecurity management when introducing intelligent applications, providing support for the safe and stable operation of power units.
Tianwan NPP comprises a mix of technologies – Russian supplied VVER-100 units and Chinese ACPR-1000s. In 2018, Russia and China signed four agreements, including for the construction of two VVER-1200 reactors for Tianwan units 7&8. Construction of unit 7 began in May 2021, and unit 8 in February 2022. Tianwan 7 is scheduled to be commissioned in 2026 and Tianwan 8 in 2027. Tianwan 1-4, with VVER-1000 reactors, commissioned between 2007 and 2018, are already in operation. Units 5&6 with Chinese ACPR-1000s were commissioned in 2020 and 2021.
