Step-by-step instructions can form a critical component of industrial maintenance, contributing to process reliability, safety and knowledge transfer. Structural and content-related requirements for such instructions are well established in the literature and relevant standards. But little is known about how practitioners assess the relative importance of these features or how different formats influence their delivery. In this paper a literature review is combined with the results of a survey conducted among 52 industry professionals. Ten quality features for maintenance instructions are defined and evaluated in terms of importance. Furthermore the suitability of different media types e.g. paper/PDF, video, app or AR to support these quality features is assessed. Based on these findings, the application of generative artificial intelligence (AI) for the creation of step-by-step instructions is explored. The focus is placed on existing generative pretrained transformer (GPT)-based large language models. An experimental use case using the GPT-4o model and involving two machine tools and 22 sub-tasks was conducted to assess structural compliance and content quality of AI-generated step-by-step instructions. The results show a high degree of individual variation in how users assess the importance of quality features and suitable media formats to support those quality features. Furthermore they show that the assessed generative AI model is capable of reproducing detailed structural templates for step-by-step instructions. Obtaining sufficient content quality in the generated instructions failed during the conducted use case.
Diedrich et al. (Thu,) studied this question.