Abstract
<title>Abstract</title> <p>In this paper we propose a modular multi-agent framework for the semantic analysis and interpretation of production processes and supply chains through the integration of process mining, dynamically updated Knowledge Graphs (KGs), and retrieval-augmented Large Language Models (LLMs). Emphasizing the use of agent-based Generative Artificial Intelligence (GenAI) for event log-driven analysis, this research addresses the challenges of Industry 4.0 and prepares for the transition toward Industry 5.0, which demands a closer synergy between humans and machines. Unlike traditional process mining approaches, the proposed framework integrates agent-level explainability, semantic graph reasoning, confidence-aware response generation, and dynamic graph updating mechanisms to improve transparency and contextual interpretation of industrial event logs. The approach automates the mapping and analysis of industrial and logistics processes, transforming structured and unstructured industrial logs into semantically linked graph representations that can be queried and interpreted through coordinated AI agents. This architecture uses a multi-agent framework to support industrial data interpretation, facilitate semantic exploration of process information, and assist human decision-making. The proposed solution was tested on different data sources, including logs from ERP, semi-structured logs from MES and unstructured reports of manufacturing facility specializing in high-precision components. Experimental observations qualitatively indicate a reduction in manual analysis effort together with improved accessibility to contextual industrial information. The primary scientific contribution of this work is the integration of hierarchical agent-based log segmentation, semantic Knowledge Graph construction, and explainability-aware retrieval mechanisms within a unified industrial analytics framework capable of supporting explainable industrial decision support.</p>