A Cognitive Framework for Pressure Equipment Inspection Based on Multi-Agent AI Systems and Vllm Models
TL;DR
It is demonstrated how specialized AI agents—including agents for visual analysis, technical diagnostics, compliance verification, and interactive communication—can autonomously identify defects, calculate the remaining service life, and generate valid reports for registries such as CROPP.
Abstract
: This paper investigates the transformation of the traditional pressure equipment (PE) inspection process into an intelligent digital workflow through the application of advanced artificial intelligence technologies. The research focuses on overcoming the issues of inspector cognitive overload and the inefficient management of extensive documentation during the interpretation of the Pressure Equipment Directive (PED) and national regulations. An innovative multimodal architecture is proposed, based on a Multi-Agent System (MAS) and Vision-Large Language Models (VLLMs) utilizing a Retrieval-Augmented Generation (RAG) mechanism for dynamic compliance validation. Through a case study of a Liquefied Petroleum Gas (LPG) tank inspection, it is demonstrated how specialized AI agents—including agents for visual analysis, technical diagnostics, compliance verification, and interactive communication—can autonomously identify defects, calculate the remaining service life, and generate valid reports for registries such as CROPP.