Talking like Piping and Instrumentation Diagrams (P&IDs)
This work addresses a domain-specific problem for engineers by providing a tool to interpret process information in P&IDs using AI, though it is incremental as it applies existing graph and LLM techniques to a new application area.
The paper tackles the problem of enabling natural language communication with Piping and Instrumentation Diagrams (P&IDs) by representing them as labeled property graphs using the DEXPI data model and integrating these graphs with Large Language Models (LLMs) through graph-based retrieval augmented generation (graph-RAG), resulting in a system that allows users to query P&IDs in natural language and helps mitigate LLM hallucinations.
We propose a methodology that allows communication with Piping and Instrumentation Diagrams (P&IDs) using natural language. In particular, we represent P&IDs through the DEXPI data model as labeled property graphs and integrate them with Large Language Models (LLMs). The approach consists of three main parts: 1) P&IDs are cast into a graph representation from the DEXPI format using our pyDEXPI Python package. 2) A tool for generating P&ID knowledge graphs from pyDEXPI. 3) Integration of the P&ID knowledge graph to LLMs using graph-based retrieval augmented generation (graph-RAG). This approach allows users to communicate with P&IDs using natural language. It extends LLM's ability to retrieve contextual data from P&IDs and mitigate hallucinations. Leveraging the LLM's large corpus, the model is also able to interpret process information in PIDs, which could help engineers in their daily tasks. In the future, this work will also open up opportunities in the context of other generative Artificial Intelligence (genAI) solutions on P&IDs, and AI-assisted HAZOP studies.