In many organisations, maintenance procedure documentation (used to guide technicians through complex manual tasks) is still paper-based. Caitlin's research combines Human-Computer Interaction and Ontology Engineering to transform the way that organisations manage and use procedure documentation to support maintenance technicians.
Integrating the representative ability of deep neural network with probabilistic logical reasoning is still an open and a challenging problem. It brings expensive computation cost due to the exponentially large searching space in knowledge representation and knowledge compilation. Ziyu's research interest is to combine deep learning with logic and probability theory for linguistics computing, taking into account knowledge representation and knowledge reasoning.
Tabular information exists ubiquitously inside PDF documents. They can encompass an entire document, or appear concurrently with text and images. The purpose of this project is to extract information into a machine readable format, while preserving implied relationships such as headers and titles irrespective of domain, through a hybrid computer vision and rule based approach.
Technical language corpora which express information about the same scenario commonly convey this information in markedly different ways. This project aims to devise methods for measuring the similarity of content in technical language corpora and evaluate the effectiveness of these methods.
Majiga's research interests include knowledge graph construction and knowledge discovery from natural language text using natural language processing, deep learning and network analysis. Automatically extracting key information, which is buried within a large amount of textual data, through knowledge graph construction is vital to support intelligent applications for knowledge discovery.
Knowledge graph rectification is using domain or high-level understanding to add, remove or alter predicted links in knowledge graphs. Given the advances of knowledge graph embeddings and their use in link prediction, codified logical understanding at a class level can be used to further improve these links and associated knowledge graphs.
Tyler's research project involves investigating how technical language in the form unstructured industrial maintenance records differs from common English corpora in terms of linguistics, data treatment, and potential for publicly available data sets and tasks. Tyler's research aims to build a foundation for consistent, reproducible and resource efficient application of TLP in industrial settings.
Failure modes and effects analysis and maintenance strategy development are key contributors to the design of any asset management program. This CEED Project aims to apply technical language processing to identify problems, causes and remedies in maintenance work orders in order to improve asset life estimation.
Determining mean-time-between-failure (MTBF) estimation for in-service assets is an essential process. We emulate the process of end-of-life event detection using a natural language processing pipeline followed by statistical parameter estimation to produce MTBF values for inservice assets from maintenance work order data.
The source code is available online.
The resources sector captures vast data about the history of its assets held in structured and unstructured text fields. This project aims to construct knowledge graphs from maintenance work orders in order to visualise historic asset data, provide decision support to reliability engineers, and allow for easy identification of failure modes.