An automated method for converting P&IDs in shipbuilding to structured graphs

The P&ID (Piping and Instrumentation Diagram) is an essential drawing used to understand the structure and interconnections of systems within a ship. It serves a critical role throughout the entire lifecycle, from design to maintenance. However, because P&IDs express complex engineering information using simplified symbols, lines, and text, manual interpretation is typically required, which poses challenges such as high expertise demands, time consumption, and inconsistency depending on the individual engineer’s interpretation. In this study, we proposed a method for automatically analyzing the interconnections between equipment and piping from P&IDs in both image and DXF (Drawing eXchange Format) formats, considering the distinct characteristics and limitations of each format, and converting them into a normalized graph structure. First, deep learning-based object detection was applied to image-formatted P&IDs to automatically detect various symbols and extract their classes and position information. Among the models tested, Cascade R-CNN demonstrated the best detection performance. Subsequently, entities such as lines and text were extracted from DXF-formatted P&IDs, and algorithms were developed to filter out significant lines and automatically determine branch relationships. Finally, the key components extracted from the P&ID were defined as nodes, and the interconnections between them were defined as edges to form a structured graph. Experimental results using test P&IDs confirmed that the proposed method could accurately identify interconnections and convert them into a structured graph even in environments that reflect real-world drawing conventions. As a result, the proposed method significantly improves upon the previously manual and repetitive interpretation process. It establishes a foundation for utilizing structured graph data in a wide range of future design and operational support applications by converting P&ID information into digital and normalized data. • We propose an automated method for converting shipbuilding P&IDs into structured graphs. • We employ a dual-format approach, utilizing both image and DXF formats for robust symbol detection and graph conversion. • Cascade R-CNN is chosen to detect symbols, showing high accuracy using synthetic training data. • Using algorithms specifically designed for DXF P&ID analysis, the method identifies interconnections, including branch relationships. • The method is applied to test P&IDs, demonstrating its effectiveness and enabling the development of intelligent design systems.