Researchers automate process diagrams with LLMs and GAs

Researchers from Skolkovo Institute of Science and Technology (Timur Zakarin, Sergei Voitov, Sergei Shumilin, Evgeny Burnaev) have developed *P&ID Pilot*, an end-to-end AI pipeline that automates the creation of industrial process diagrams—from optimal Process Flow Diagrams (PFD) to validated Piping and Instrumentation Diagrams (P&ID). The system leverages a hybrid approach combining genetic algorithms (GA) and large language models (LLMs) to synthesize PFD topologies with the lowest loss values while adhering to engineering constraints.

In the second stage, an LLM-based agent transforms the generated PFD into a source-grounded P&ID using a restricted engineering software development kit, achieving 100% execution success while maintaining compliance with domain-specific rules and reference graph structures. This unified pipeline offers a feasible path toward end-to-end process design automation, substantially reducing manual engineering effort and enabling exploration of numerous design topology options at scale.