On the use of large language models for methods-time measurement

Manual operations remain a significant part of modern manufacturing, maintaining the role of Methods-Time Measurement (MTM) as a framework for improving productivity and standardising work processes. Despite its usefulness, conducting MTM analysis is a labour intensive task requiring specialised expertise. Recent advances in Large Language Models (LLMs) provide a promising path for streamlining MTM analysis. This study establishes a performance baseline for LLMs in this domain by assessing their capabilities when provided with meticulously detailed work descriptions and a comprehensive MTM knowledge base. To this end, we evaluated the analyses generated by DeepSeek-V3.2-reasoner, GPT-5.2-pro, Gemini-2.5-pro, and Qwen3-14B by comparing them against expert-validated ground truths across eleven working scenarios. Our findings suggest that LLMs can accurately perform MTM analysis when supplied with detailed work descriptions and domain knowledge. GPT-5.2-pro and Gemini-2.5-pro proved especially capable, perfectly analysing five and six scenarios, respectively, across ten experimental runs. Despite this success, challenges remain when tasks involve subtle rule applications or steps that are implied rather than stated. Although LLMs significantly reduce analysis time and effort, expert validation remains critical for reliable deployment. These results demonstrate the viability of LLMs as assistive tools for MTM practitioners.