Out of the box NLP processes are well established for use on multi-sentence texts. However their use on domain specific unstructured engineering texts, specifically maintenance notifications, is under-explored. This project proposes the development of pre-processing and post-processing code to adapt the annotation and knowledge graph pipelines developed by the UWA NLP-TLP team from maintenance work order short text to long text notifications. The aim is to extract valuable contextual factors and potential causes for maintenance from the long text.
A typical extract from long text is “20.05.2020 08:46:18 UTC Joe Bloggs. Oil level switch inspection, disconnection and reconnection complete. 21.05.2020 00:20:33 UTC Jill Cloggs. The gearbox was replaced with the spare unit, this came pre-filled with gear oil that was tested and left in for run-up, the level was drained down to half way on the sight glass. The gearbox flange bolt holes needed to be tapped out due to rust and paint still in the holes, this delayed the installation by several hours. The old gearbox was drained for ABC.”
This project requires strong coding skills, particularly in Python. Good performance in the NLP unit CITS 4012 or other relevant deep learning algorithm experience.