Semantic Quality Assurance of Industrial Maintenance Procedures

Abstract

Maintenance technicians in industry follow procedures that guide them through inspection, repair, and service tasks. Organisations seek to convert procedure documentation to machine-readable formats as their digital capabilities improve and regulatory requirements tighten. In this paper, we consider the opportunity for semantic quality assurance of digital procedures. We demonstrate a configurable and repeatable workflow containing three modules. The completeness module makes implicit information in procedures explicit using OpenAI’s Generative Pre-trained Transformer (GPT) model. The consistency module creates Resource Description Framework (RDF) triples that are aligned with, and checked against, the axioms of the open-source Ontology for Maintenance Procedure Documentation (OMPD). Finally, the correctness module performs closed-world checks on the RDF triples using the Shapes Constraints Language (SHACL). Each module can be used in isolation, or together, to realise an end-to-end semi-automated quality assurance workflow. Pre-processing of the raw maintenance procedure documents to extract entities (tools, materials and activities) and relations is achieved in a novel manner using prompt engineering with OpenAI’s GPT-3.5 Turbo model and few-shot learning. This end-to-end workflow enables organisations to perform quality assurance such as assessing the correct order for task sequences, and checking that all maintenance procedures have at least one maintenance task. We demonstrate this workflow on six procedures from the iFixit repository. The outputs of this workflow support maintenance technicians, planners and engineers by realising high-quality procedure documentation and automated procedure management update processes. The code and data used in this work is publicly available at https://github.com/equonto/quokka/.

Keywords

• Industrial ontology
• ontology templates
• OpenAI GPT
• OTTR
• SHACL
• technical language processing