Automatic Recognition System for P&ID Diagrams in Distributed Control Systems

Piping and Instrumentation Diagrams (P&IDs) are essential engineering documents for configuration design in thermal power plants. In modern coal-fired power generation, P&IDs are primarily delivered in CAD formats or as scanned PDFs, and their digitization still relies heavily on manual interpretation. This process is inefficient and labor-intensive due to the complexity of graphical symbols, dense textual information, and irregular spatial layouts. With the advancement of computer vision, the industry is increasingly exploring intelligent methods for digital design delivery and legacy diagram digitization. This study presents an intelligent recognition system based on a deep learning framework, specifically designed for the thermal power domain. The system combines cutting-edge object detection and optical character recognition (OCR) technologies to accurately extract equipment symbols, identifiers, and measurement point codes from P&IDs originating from multiple design institutes. Additionally, it enables structured output for downstream processing and system integration. To support model development, a customized dataset was constructed using real-world diagrams from thermal power projects. Experimental results on unseen samples show a symbol detection accuracy of 91% and a text recognition accuracy of 97%, demonstrating both high efficiency and strong applicability for engineering scenarios.