ChatCNC: Conversational machine monitoring via large language model and real-time data retrieval augmented generation

Abstract

Human-Centric Smart Manufacturing (HCSM) has become a central theme of Industry 5.0, promoting collaborative interactions between humans and intelligent systems. Nevertheless, HCSM technologies have been struggling to reach their full potential due to the lack of digital literacy among manufacturing workers, particularly in real-time machine monitoring. Current monitoring systems with rigid interfaces limit the operators to handle Industrial Internet of Things (IIoT) systems to directly query manufacturing data for further analysis without external technical support. To address such bottlenecks, we propose ChatCNC, a conversational machine monitoring framework that integrates Large Language Models (LLMs) to enable natural language-driven interactions with real-time Computer Numerical Control (CNC) machine data. Leveraging LLM-based multi-agent collaboration and Retrieval-Augmented Generation (RAG), ChatCNC interactively retrieves data from real-time IIoT database while also supporting context-aware responses based on the collected data, which reduces reliance on technical support from software engineers. As ChatCNC allows rapid adaptation of LLM Application Programming Interfaces (APIs) via prompting techniques, its performance is evaluated across multiple versions, each combining different LLMs and prompts, using various types of questions. Notably, our framework demonstrates its reliability in human-data interaction for industrial applications, achieving 93.3% accuracy in responding to complex queries that require advanced data inference like production tracking. Furthermore, possible failure modes are thoroughly analyzed based on interaction scenarios among multiple LLM-based agents. Such results highlight the potential of the framework as a cornerstone for HCSM.