ASDR: automatic system to diagnose and recognize electrical drawings

This paper presents an automated system to diagnose and recognize Metering and Relaying Diagrams (MRDs) in electrical drawings. These diagrams play a crucial role in defining protection schemes for new substations, so ensuring their accuracy is essential to avoid delays and additional costs in the construction process. The study focuses on four primary stages: data collection, annotation, model training, and application development. Advanced machine learning models such as YOLOv5, RTM-Det, and Faster-RCNN are explored. To enhance flexibility, the study utilizes the MMYolo toolbox and the MMDetection toolbox and showcases adequate dataset preparation with Albumentations, emphasizing the importance of transfer learning for efficient resource utilization. The resulting Automatic System to Diagnose and Recognize Electrical Drawing (ASDR) empowers experts to scrutinize drawings, identifying missing components and potential missing areas. The ASDR web application, developed using Next.js, Flask, and Postgresql, provides a user-friendly and scalable solution.