Classification of Functional Types of Lines in P&IDs Using a Graph Neural Network

In recent years, with the use of computer-aided design software for most plant projects, intelligent piping and instrumentation diagrams (P&IDs) have become the default format for P&IDs. However, most of the previous P&IDs are often in image format, and it is necessary to convert them to intelligent P&IDs. This study solves the problem of classifying the functional types of lines in the process of converting image P&IDs into intelligent P&IDs. The challenge is to classify whether each line in a P&ID is a piping or signal line. To solve this, the objects connected to the line, not the type of line, need to be considered. First, the connection relationships between symbols and lines in a P&ID are represented as a graph. The problem is then modeled and solved as a node classification problem using a graph neural network based on the mean aggregator convolutional layer of GraphSAGE. In addition, a dataset was generated from 19 real-world P&ID drawings to train the graph neural network and the optimal model was selected through hyperparameter tuning. The implementation and experiments of the proposed method demonstrate an accuracy of 99.53%.