Identifying LLM applications for predictive maintenance

Abstract

Predictive maintenance (PdM) offers significant potential for reducing downtime and costs, but in small and medium-sized enterprises (SMEs), its implementation often fails due to heterogeneous equipment landscapes, limited and inconsistent data, and restricted data science resources. At the same time, maintenance-relevant knowledge is distributed across documents and service reports, making it only partially accessible and difficult to use effectively. This preliminary study analyzes the roles that large language models (LLMs) can assume as an assistive technology along the PdM process. Methodologically, the paper is based on a qualitative framework analysis of relevant literature in order to structure the problem space, systematize key fields of application for LLMs in the PdM context, and derive design implications as well as potential development paths toward integrated and agentic support systems.