Roboflow is a framework made for developers in computer vision field, which enables better data collection for pre-processing and model-training techniques.

In order to build a picture warehouse for AI model training which can identify objects, issues, and anomalies of equipment under grid inspection, Gia Lai PC (i) implemented the work of picture collection of grid equipment, (ii) utilized retrieving equipment for modelling equipment anomalies, (iii) set up temporary structures in various areas in order to generate diverse backgrounds, (iv) conducted physical impacts in order to simulate and record anomalies, issues of equipment on 110kV and MV power grid.

The work was implemented from 01/12/2022 to 15/01/2023 with 03 phases: Phase 1 from 01/12 to 15/12/2022, Phase 2 from 15/12 to 31/12/2022 and Phase 3 from 01/01 to 15/01/2023. Through the work, Gia Lai PC collected 18,138 pictures of equipment and pieces, including 3,239 insulators, 3,706 conductors and lightning wires, 903 lightning arresters and 10,290 other equipment and pieces via utilization of smart devices, camera drones for taking pictures of equipment, pieces with anomalies, defects. The collected pictures are in high quality, at least HD quality; objects are clear and defects are zoomed in within picture frame, ensuring that there is no shadow from other objects shrouding onto target objects; and data is saved to EVNPCPC’s shared database.

The interface which is used for picture storage for AI model training for grid inspection

The study, pilot application of AI in image processing is for site inspection from pictures in PMIS and for analysis of picture from camera drones in order to detect anomalies, minimize grid faults from violation of power corridor, defect of insulators, and other reasons. The application of AI has brought many benefits, such as: to enhance the professionalism in grid management, to early detect defects and to prevent grid faults, to improve reliability in power supply; to support operators in checking the whole power grid and all assets, equipment; to replace periodical inspection with a more flexible, active inspection plan....

However, in order to better, more effectively implement AI application, it is necessary (i) to invest and sufficiently equip specialized measuring devices, tools; (ii) to build a training procedure and technical teams with the capability of using camera drones, taking pictures, and building processing and detecting modules (Module AI) to analyze videos taken by camera drones, to detect anomalies...

Author: Văn Thiện