Graph Learning in Machine-Readable Plant Topology Data

Abstract

Digitalization shows that data and its exchange are indispensable for a versatile and sustainable process industry. There must be a shift from a document-oriented to a data-oriented process industry. Standards for the harmonization of data structures play an essential role in this change. In engineering, DEXPI (Data Exchange in the Process Industry) is already a well-developed, machine-readable data standard for describing piping and instrumentation diagrams (P&ID). In this publication, industry, software vendors, and research institutions have joined forces to demonstrate the current developments and potentials of machine-readable P&IDs in the DEXPI format combined with artificial intelligence. The aim is to use graph neural networks to learn patterns in machine-readable P&ID data, which results in the efficient engineering and development of new P&IDs.