Work

Enhancing Assembly Instruction Generation for Cognitive Assistance Systems with Large Language Models

Year: 2025

Type: article

Abstract: Cognitive Assistance Systems enhance manual assembly by shortening learning cycles and allowing workers to handle a wider range of products. However, generating assembly instructions remains time-consuming, particularly in environments with high product variability. This paper presents a novel approach to automate and streamline this process using MTM-based standardized instruction texts and Large Language Models via the OpenAI API. By deriving instructions from MTM analyses, a unified syntax and structure can be realized, improving consistency and efficiency. The integration of GTP-4o further enables the automatic generation of context-specific warnings and error notifications. Model fine-tuning and prompt engineering play a pivotal role in this approach, allowing the generation of precise instructions. The evaluation, based on the BLEU and METEOR scores, focuses on the assessment of the technical functionality and the quality of the generated outputs and shows promising results that highlight the potential of this approach for improving the automated generation of standardized assembly instructions for Cognitive Assistance Systems.

Source: Procedia CIRP

DOI: 10.1016/j.procir.2025.03.010

Authors: Benedikt Kelm, Paul Hubert Haas, Simon Jochum, Lennard Margies, Rainer Müller

Institutions: Saarland University, Zentrum für Mechatronik und Automatisierungstechnik

Language: English

Topic: Manufacturing Process and Optimization

Subfield: Industrial and Manufacturing Engineering

Field: Engineering

Domain: Physical Sciences

SDG: Industry, innovation and infrastructure

Open Access status: diamond

Funder: Bundesministerium für Wirtschaft und Klimaschutz