Foundation models and intelligent agents in chemical engineering: advances, challenges, and perspectives

Abstract

Chemical engineering spans scales of molecules, catalysts, reactors, units, and plants, governed by physical laws and subject to stringent safety, operability, and reliability requirements. General-purpose large language models, although increasingly capable of scientific reasoning, often lack the domain grounding, physical consistency, and engineering reliability required for industrial deployment. We argue that reliable chemical engineering AI will require the closed-loop integration of domain-specific foundation models, agentic orchestration, and physics- and simulation-based validation. Within this framework, domain-specific models provide knowledge for task-specific reasoning, multi-agent systems support complex task orchestration and specialized tool use, and mechanistic models, engineering constraints, and process simulators assess feasibility and provide feedback for refinement. Existing studies largely demonstrate individual components rather than a fully integrated engineering intelligence lifecycle. This Perspective reviews recent progress in domain-specific foundation models, agent-based process design and optimization, process safety analysis, and operational decision support. We outline challenges and opportunities across three dimensions: strengthening the domain-specific foundation models, developing coordinated agent platforms for multiscale engineering workflows, and establishing traceable evidence and controlled updating mechanisms that distinguish task-level feedback, offline version updates, and online adaptation. Together, these directions define a pathway from conversational chemical engineering models toward reliable, verifiable, and continuously improving engineering intelligence.

Keywords

chemical engineering / domain-specific foundation models / multi-agent systems / closed-loop validation / engineering intelligence

Rights & permissions

Higher Education Press 2027

Just Accepted

This article has successfully passed peer review and final editorial review, and will soon enter typesetting, proofreading and other publishing processes. The currently displayed version is the accepted final manuscript. The officially published version will be updated with format, DOI and citation information upon launch. We recommend that you pay attention to subsequent journal notifications and preferentially cite the officially published version. Thank you for your support and cooperation.