GenAI-Driven Occupational Safety Analysis in IoT-Enabled Environments

Abstract

This paper presents a Generative Artificial Intelligence (GenAI)-driven workflow for analyzing occupational safety in IoT-enabled industrial environments by integrating Large Language Models (LLMs), Vision-Language Models (VLMs), and Model-Driven Engineering (MDE) to deliver comprehensive, multimodal insights. The approach addresses the challenges posed by heterogeneous and unstructured data by using LLMs to interpret textual sources and VLMs to extract information from visual artifacts such as process flows and operational diagrams. For analytical reasoning, the workflow incorporates MDE techniques and rule-based modeling to support the formal validation of workflows and the identification of potential safety risks. The framework focuses on process extraction and analysis, generating structured representations using PlantUML activity diagrams to capture workflows, dependencies, and interactions across system components and stakeholders. The approach is demonstrated using a representative IoT-enabled industrial scenario, showing its effectiveness in improving transparency, identifying hazards, and supporting informed decision-making.