Melinda Hodkiewicz (BHP Fellow, Engineering for Remote Operations ) and her colleagues at the University of Western Australia’s Natural & Technical Language Processing Group have just published a paper* on the use of knowledge graph technology to build and explore large maintenance data sets.
The authors have been studying how short text fields on maintenance
work orders (MWO) can be extracted and combined with structured data
into a knowledge base. This can be used to find information in historic
asset data for failure modes and effects analysis, maintenance strategy
validation and process improvement. A failure mode taxonomy was derived
from the ISO14224 standard.
The researchers have developed two open source tools to facilitate the work: MWO2KG uses deep learning supported by annotated training data to automatically build a knowledge graph, and Echidna an intuitive query-enabling interface that visualizes the historic asset data in the graph database. A demonstration of Echidna is available online and the source code for both tools is on GitHub.
The research is said to leverage the Semantic Web’s resource description framework (RDF) which has been used to build a triple store of maintenance data. Triple stores are particularly amenable to visualization in a graph, an approach which is said to be ‘increasingly used’ in the engineering sector (the ‘Bosch Materials Science Knowledge Base** was cited).
A supervised deep learning model driven by (expert) annotated data was used to extract named entities. Here the authors acknowledged that access to maintenance work order data is ‘problematic’ as being considered ‘commercial-in-confidence’. Also while it is desirable to have multiple annotators, ‘getting them to agree is a challenge’. Named entity recognition was performed with the open source Flair library. At the end of the pipeline, a Neo4j graph database was used for further data exploration. The researchers are now working on functionality that will enable data owners to upload their own datasets for analysis.
* Constructing and exploring knowledge graphs from maintenance data in Proceedings of the Institution of Mechanical Engineers, Part O: Journal of Risk and Reliability.
** The Bosh Materials Science KB appears to be something of an
Arlésienne. We found nothing on the Bosch website but there are
many references in the semantic web community to research in progress
on the BMSKB such as this one.
© Oil IT Journal - all rights reserved.