Welcome to an update on CTMTDS activities. In this issue we will be focusing on two projects conducted by PhD students while on their industry placement.
Highlights #
Submission of Masters Theses #
Congratulation to Jaymin Moffatt and Toby Griffiths for submitting their Masters of Professional Engineering theses. Thank you to our industry partners for supporting their projects.
Placement Projects #
Timing of Fixed Plant Maintenance Interventions #
During her placement at Roy Hill, ITTC PhD student Yingying Yang analysed historical maintenance schedules and time periods for shutdowns.
Shutdowns are traditionally scheduled at fixed intervals, a strategy that works well in a controlled, planned environment, but is not ideal in rapidly-changing situations. Based on her experience at Roy Hill, Yingying is planning a PhD proposal focusing on optimising shutdown timings based on stockpile status, demand chain, maintenance requirements, and even market conditions, to introduce flexibility into the shutdown process and the timing windows of shutdown.
Yingying’ s work will potentially enable shutdown planners in the mining industry to make flexible shutdown schedules when certain situations arise, which would bring additional benefit for the industry.
Asset Health Management #
Ryan Leadbetter undertook a two-week placement at Roy Hill (RH) in March and continued participating in weekly meetings with the Roy Hill Smart Maintenance team. Ryan explored a dataset that was constructed by Toby Griffiths, containing sensor, alarm, and state information for Roy Hill production excavators.
As well as working with the excavator data, Ryan has also been contributing to ‘current state’ documents across the mobile asset and rail sections of Roy Hill, summarising what steps have been taken within the company towards data-driven decision-making tools. This will lead to more potential projects and help with identifying an area for his PhD proposal.
Workshops organised #
Researcher Catch-up #
Our Researcher Catch-up have continued fortnightly virtually. These meetings are arranged by our research fellows and consist of two 30-minute presentations given by researchers in the Centre.
They are attended by the Centre Chief Investigators, Research Fellows and HDR candidates associated with the Centre
In early May Hoa Bui hosted an online seminar with presentations from Toby Griffiths and Jaymin Moffatt
Managing Streamed Sensor Data for Mobile Equipment Prognostics #
Toby Griffiths (UWA) presented work he has completed for his Masters of Professional Engineering entitled ‘Managing Streamed Sensor Data for Mobile Equipment Prognostics’.
The ability to wirelessly stream data from sensors on mobile equipment provides opportunities to proactively assess asset condition. However, data analysis methods are challenging to apply due to the size and structure of the data, which contains inconsistent and asynchronous entries, and large periods of missing data. These methods also require expertise from site engineers to inform variable selection.
Toby project developed a data preparation method to clean and arrange this streaming data for analysis, including data-driven variable selection. Data is drawn from a mining industry case study, with sensor data containing 300M rows (20GB) for a fleet of 13 excavators during a nine-month period. Variables include numerical sensor data and binary indicators describing the conditions and status of different subsystems. Additional data is drawn from the fleet management system and work orders completed on the fleet, with a focus on hydraulic failures. The data cleaning process applies the OHLC (Open-High-Low-Close) method to the sensor data. This summarises data as the first (open), maximum (high), minimum (low) and last (close) values for the interval. Results using a one-hour interval show a data size compression of up to 96%, while retaining important features, such as peaks, and minimising the effect of missing data. Using this method, 15 of the total 21 hydraulic failures are present in the data, compared to 14 using the raw data. The extra visibility is due to the OHLC method ignoring missing data within each interval, provided there is some data to populate the four values. Without this summary method, the 15th failure would not have been detected because it occurred during a period where considerable sensor data was. The OHLC method also enables the use of candlestick plots for effective visualisations, and it can be used on any data presented in a similar form. A logistic regression model using L1 regularisation was developed based on the OHLC data, utilising a lagged data frame for numerical sensors and summarising the binary indicators. The model identifies variables that are most sensitive to specific faults and hence provide early warning of failure. Application to hydraulic failure detection shows that the data-driven selection agrees with a selection based on expertise from reliability engineers and maintenance planners. This shows that the methods developed in this paper provide a useful basis for predictive models while also compressing the data.
Detecting Asset Cascading Failures #
Jaymin Moffitt presented the work he has completed for his Masters of Professional Engineering entitled ‘Detecting Asset Cascading Failures’.
Experienced process plant personnel observe that corrective maintenance work on one asset may often be followed by corrective work on the same asset or related assets within a short period of time. This problem is referred to as a cascading failure. If cascading failures can be identified, preventative measures can be implemented to prevent those cascades, thereby eliminating unnecessary corrective work. Data is drawn from over 50,000 work orders for 5,655 pumps in a process operation over a 5-year period. A complex network is produced by connecting assets based on the frequency of co-occurrence of work. The method produces quantified measures, eigenvector centrality and betweenness centrality, which are visualised as properties of nodes of a complex network. This analysis identifies pumps that are super-spreaders and pumps whose failure leads to cascading failures of other pumps. The model can be tuned to different time windows, for example failures, within 1 or 7 days. From these insights, changes can be made to operational, maintenance and recording practices to prevent re-occurrence. Recommendations resulting from this work include investigation of self-loops, potentially due to work or operational quality issues, and identification of chronic hidden failure events in standby pumps.
In late May Michael Stewart hosted and presented an online seminar with a focus on natural language processing.
From Named Entity Recognition to End-to-End Entity Typing #
Entity recognition is an important technique in natural language processing (NLP), playing a crucial role in many NLP tasks such as information extraction and knowledge graph construction. In this presentation Michael provided a high-level overview of entity recognition, beginning with named entity recognition and concluding with his work on end-to-end entity typing. He introduced the Theme 1 “Maintenance Work Order Tagging” project, which involves constructing the first high quality labelled dataset for entity typing in maintenance in order to provide a solid foundation for future projects and research in Theme 1. The presentation concluded with a demonstration of his work on applying this project to automatically construct a knowledge graph for maintenance, allowing maintenance work orders to be visualised and queried in a powerful and novel way.
Prediction of Pump Maintenance using Markov Models on Free Text Work Orders #
Alex Rohl presented the work he is completing for his Honours research on the prediction of pump maintenance using Markov models on free text work orders.
In many industries, it is essential to have a reliable model that can predict the future maintenance of assets, in order to establish an optimal replacement policy. In his presentation, Alex discussed methods of deriving such a model from unstructured short-text work orders that relate to maintenance on a specific piece of equipment within a large collection of physical assets. This process is two-fold. Firstly, by applying natural language processing techniques, it is possible to translate each short-text into a generalised “concept”, which describes a group of similar work orders. Secondly, for each asset, the transitions from each concept to the next can be used to create a discrete-state continuous-time Markov model. This process was verified numerically by analysing an industry dataset that consists of 5,000 distinct pump assets and 80,000 work orders over a timespan of eight years.
In the spotlight #
Congratulations to Hoa Bui on being awarded the prestigious AustMS-WIMSIG Maryam Mirzakhani Award .
This award is designed to support international female students pursuing a postgraduate degree in mathematics in Australia. Each year the award is made on a competitive basis by a selection committee of distinguished mathematicians, appointed by the executive committee of WIMSIG. Only one Maryam Mirzakhani Award is awarded each year.
Stay tuned for our next issue where we will cover:
- New projects approved for the Centre
- CTMTDS Code Repository