AR-TFRA: An AR-Assisted Typical Fault Retrieval and Analysis System for Efficient Power Grid Communication Operation and Maintenance Training

Smart power grids desire fast fault detection and analysis. Though modern AR-assisted training paradigms provide more realistic pre-defined operation guidance, it is difficult to integrate all typical faults with AR and almost impossible for trainees to well learn all fault analysis methods. How to build an AR-assisted typical fault retrieval and analysis system is of great value in practice yet under-explored. To accomplish this, we propose AR-TFRA. Specifically, we first build a vector database of typical faults based on a multi-modal foundation model to allow multi-modal query and retrieval. Then, both text and audio are realized for database query. After combining with AR, AR-TFRA provides real-time feedback and interaction of typical fault analysis during communication operation and maintenance training. We believe such a system is broadly applicable to various training scenarios.