Welcome to an update on CTMTDS activities Jan - Jun 2021.
Team Updates #
Changes to the members of the Centre in the first six months of 2021 include welcoming:
Two new PhD students:
- Sandy Spiers joined the Centre in February. Sandy will complete his PhD at Curtin University within Theme 3, supervised by Ryan Loxton and Hoa Bui.
- Braden Thorne joined the Centre in March. Braden will complete his PhD at UWA within Theme 2 supervised by Michael Small and Débora Corrêa.
A new Research Fellow, Mojtaba Heydar joined the Centre at end of June to support Theme 3.
Aaron Cahill has joined the board representing Roy Hill. We thank Mike Lomman for his excellent support of the Centre and look forward to Aaron participation
Chris Rowlands has left Roy Hill; we thank him for his support in significantly increasing our engagement with Roy Hill. Laks Bhatti and Mike Manning have joined us on the Operating Committee as Roy Hill representatives.
Knight Kulkaew, BHP Maintenance and Engineering Centre of Excellence and Liam Hussey, BHP Nickel West have joined us in the Operating Committee and PhD Support Panel as BHP representatives. We thank Jose Ricce for his support in both the Operating Committee and PhD Support Panel.
We have also celebrated the success of PhD students meeting academic milestones:
- PhD candidates in Research Theme 1, Ziyu Zhao and Tyler Bikaun, completed their confirmation of candidacy at UWA.
- Tim Pesch, PhD candidate in Research Theme 2, has had his PhD research proposal accepted.
CTMTDS Activities #
Knowledge Graph Construction from Maintenance Work Orders #
Throughout the first six months of 2021, Theme one researchers Michael Stewart and Tyler Bikaun, in partnership with Liam Hussey, Lead Data Analytics at BHP Nickel West, have deployed a Knowledge Graph Application (Echidna) on the BHP servers. This will enable the NiW Kwinana team to access the application from their BHP systems and directly use the tool.
Maintenance Work Order Annotation #
Researchers in Theme 1 have consistently annotated subsets of work orders each week, with 5 annotators per work order. They have even co-opted Natasha and are using this dataset to train a model to construct knowledge graphs from work orders from BHP Nickel West and are now working on a data set from Roy Hill.
Early Detection of Failure #
Ayham Zaitouny, working with the BHP strategy and innovation team, has built a mathematical approach for the early detection of failure. They have developed a new mathematical platform to predict early failures (changes) in temporal signals. They are currently testing this on a data set.
Development of Optimisation Models for Scheduling #
Hoa Bui has developed a model designed to solve dimensionality challenges with one of our industry partners. This model is now being tested with industry data and showing itself to be quicker and more efficient than the existing methods in the literature for problems of this type. This not only results in positive outcomes for the project but also brings exciting academic results and publications.
Data Fit Organisation (DFO) model #
Keyao (Eden) Li, in collaboration with CORE Innovation Hub, is working on a model to identify the data science workflow(s) within an organisation. This includes the capabilities of different roles needed along the data workflow to support the organisation's data science innovation journey. Eden's research consists of a new project proposed to conduct an industry pilot to validate the Data Fit Organisation (DFO) model.
A data-driven approach for labelling process plant event data #
An essential requirement in any data analysis is to have a response variable representing the aim of the analysis. Much academic work is controlled, and the ground truth is clearly defined. This is seldom the reality for equipment performance in an industrial environment. It is common to find issues with the response variable in industry situations. We have developed a case study where the problem is to detect an asset event (failure) using the data available. No ground truth is known from historical records. It is essential to know the "how-to" label of the event of interest in the case study. Different labelling strategies will generate different models with a direct impact on the in-service fault detection efficacy of the resulting model. In collaboration with Kylie Hollins and Shreyas Srikanth from Alcoa, Débora Correa, Adriano Polpo, Michael Small, and Melinda Hodkiewicz submitted this case study to the International Journal of Prognostics and Health Management. This is the first paper the Centre has written as a collaboration of industry partners and academics.
ANZIAM (Australia and New Zealand Industrial and Applied Mathematics) #
Hoa Bui spoke at a one-day conference in July organised by the New South Wales branch of ANZIAM (Australia and New Zealand Industrial and Applied Mathematics), a division of the Australian Mathematical Society. Hoa presented her optimisation models for scheduling, completed in collaboration with Alcoa and Nickel West. (https://austms.org.au/event/nsw-anziam-mid-year-virtual-meeting-2021/)
Hoa Bui is hosting an internationally recognised Variational Analysis and Optimisation Webinar with other researchers in UniMelb, UNSW and FedUni on behalf of the Mathematics of Computation and Optimisation (MoCaO) special interest group of the Australian Mathematical Society (AustMS). The primary purpose of this webinar is to connect mathematicians who work on optimisation. Every week, there are around 50 researchers from all over the world who connect to discuss optimisation
UIIN University Industry Innovation Network #
Eden Li presented a paper at the 2021 The University-Industry Interaction Conference. This is the annual conference of the University-Industry Innovation Network (UIIN), which brings together up to 1000 managers, practitioners and researchers in the field of entrepreneurial universities, collaborative innovation and university-industry interaction each year. Eden presentation was titled ‘Nurturing Collaboration? A Skill Set for University-Industry Collaboration Champions’. Eden elaborated on the capability requirements of university-industry collaboration champions. Based on the data collected from interviews, required support and barriers to project-based industry placement programs were identified. Eden developed a skill set for university-industry collaboration champions with six skill requirements categorized under four types: foundation, transformational, network, and integration. . This study highlights the need for constant investment in the skills and capabilities for the future of university-industry engagement.
NIST Technical Language Processing Community of Interest (TLPCOI) #
Melinda Hodkiewicz, Michael Stewart and Tyler Bikaun presented their research at the NIST Technical Language Processing Community of Interest (TLPCOI) Model-Based Enterprise Summit. The summit was a fantastic opportunity to present research to a broad audience of both industry and academics. Approximately 100 attendees were online for the NIST TLP community of Interest workshop.
Australian Mathematical Sciences Institute (AMSI) Winter School 2021 #
PhD Researchers Ryan Leadbetter and Gabriel Gonzalez attended the nationwide Australian Mathematical Sciences Institute (AMSI) Winter School 2021. The event covered specialist topics in statistical data science, including Bayesian statistics, advanced Markov Chain Monte Carlo methods, likelihood-free inference, dimension reduction for high dimensional data, and modern neural networks. The course ran for two weeks full time. The applied Bayesian statistics knowledge Ryan and Gabriel acquired aligns well with their PhD topics.
Workshops organised #
Master Classes - March series #
Deterministic Dynamics, Machine Learning & Tipping Points by Prof Michael Small #
Deterministic dynamical systems include any physical system which behaves predictably. We measure time series for these systems and use them to build a model of the behaviour of the system. Using these time series measurements, Michael constructs a mathematical model of the dynamical behaviour of the system. The tools used to build these models are techniques from machine learning, and this model acts as a simulation of the system (known as a "digital twin"). Michael compares the future behaviour of the model to the observed behaviour of the system. When these two differ, they differ because the physical system is no longer in the same regime. This behaviour change can be characterised as a tipping point . Mathematical techniques can then be applied to these models to predict imminent nascent tipping.
This master class provided a broad introduction to the fields of applied dynamical systems, recurrence quantification analysis and complex networks. Michael discussed the computational tools that can be derived from these methods to analyse time-series data. Participants can incorporate these computational tools into a feature extraction framework as a complementary/alternative to traditional feature extraction techniques. Michael also reviewed some of the transition (change-point) detection methods derived from the above methods. They are helpful to track changes in the behaviour of time series. Participants should be able to implement these change-point detection techniques in asset health-monitoring signals to enhance the predictive analysis performances.
Pattern Recognition and Change Point Detection by Dr Ayham Zaitouny #
Transition detection methods are used to locate tipping points (change-points) in signals in various applications. In addition to the essential applications of detecting transition points in time series of dynamical systems, such detection algorithms can also be used for non-temporal signals such as image analysis and spatial data. Identifying these change-points allows a better understanding of the system and the recognition of the different modes or patterns. The detection of these transition points is a core step for better classification and more reliable predictive analytics. In terms of maintenance, these change-points may refer to the transition in the health-monitoring signal of an asset between different health modes: healthy mode, deterioration mode or failure mode.
In this master class, Ayham reviewed some of these transition detection techniques and how they can be used for maintenance applications to identify different patterns and modes.
Rethinking Technology-Led Maintenance Transformation Implementation People are the Key to Success by Dr Keyao Li #
Effective and efficient maintenance management ensures that assets are productive and meet business needs. However, the mining industry's maintenance practices are slow to evolve, and asset downtime remains stubbornly high. In light of the innovations in edge computing, artificial intelligence and Industrie 4.0, the Centre for Transforming Maintenance through Data Science (CTMTDS) transforms maintenance by conceiving a new digitally-driven maintenance management system. Although technological innovations have clear benefits, their total value depends on successful adoption and implementation. It has been reported that technological innovations fail if they are not supported by end-users and other stakeholders. Some factors that influence users' attitudes and intentions towards new technology include its perceived usefulness and ease-of-use, the skill level of the users, work design and team structure within the organisation, effects of trust and more. Thus, understanding that people are critical to successfully implementing technology-led maintenance transformations and being aware of workforce capabilities required for maintenance is of great value.At the end of this masterclass, participants developed an appreciation for:
- The critical factors of data science innovation and how to leverage them to drive data science implementation in the workplace;
- Digital capability requirements for promoting data science in the workplace; and
- Strategies to develop digital capability.
Master Classes - April series #
Knowledge Graphs Demystified, Dr Michael Stewart #
Have you ever wondered how to ask questions across multiple data sources such as SAP and delay accounting systems? According to McKinsey & Company, many companies are using graph databases to achieve greater flexibility and maintain a competitive edge. Knowledge graphs are employed by a wide range of top companies – eBay, Walmart and Volvo, to name a few.
But what is a knowledge graph? Why are leading companies actively building knowledge graphs, and how is one created?
In this master class, Michael provided an overview of knowledge graphs, highlighting their unique advantages compared to structured data models such as relational tables. He motivated participants to understand the need for knowledge graphs in maintenance via a simple maintenance-specific example. Michael demonstrated how work orders can be transformed into a knowledge graph to visualise historic asset data and allow domain experts to make informed business decisions.
Complex Time Series Modelling by Dr Débora Correa #
The possibility of representing time series as complex networks has attracted much attention in the last decade. In a complex network, the entities describing a complex system are represented as nodes, and edges define the intrinsic relationship among interconnected entities. Debora investigated how patterns in these interactions bring insights into the dynamical properties of the studied system. This idea has been extensively applied to social systems. Debora also illustrated how these patterns are applied to time series data. A complex network can be constructed from a time series to create a template or represent the shape of that time series.
Using that idea, many network representations of time series have been proposed, which try to capture the essential information about the system from different perspectives.
In this master class, Débora reviewed some of these representations to see how the structural properties of the networks can be used to characterise dynamic aspects of time series.
Solution Methods for Practical Scheduling Models by Dr Hoa Bui #
Maintenance planning and scheduling are critical in any business. Having an effective schedule will bring a range of benefits to the company. Therefore, having a scheduling optimisation model which can partially automate the scheduling process and reduce errors becomes more and more important, especially in dynamic organisations. In general, building an optimisation scheduling model involves two key steps: (1) creating a realistic mathematical model to represent the scheduling problem; and (2) developing an algorithm to solve the designed model.Some effective algorithms can solve a wide range of optimisation problems, but dimensionality is usually a big challenge. Therefore, customising or designing suitable solution methods is always a critical step in optimisation. The designed algorithm needs to provide practical solutions. Still, we also need to ensure stability in terms of speed to update the model when new data becomes available. In this master class, Hoa took participants through how to build a good scheduling model and several solution methods to tackle the problems of high dimensionality.
Researcher Catch-up #
Our Researchers present to our industry partners monthly. In 2021 the following presentations have been delivered:
An optimisation model for sixteen-week Maintenance Planning in Alcoa #
In February Hoa Bui and Andrew Harrison presented our first industry/academic presentation.
Maintenance planning and scheduling are critical in any business. Having an effective schedule will bring a range of benefits to the company. To produce maintenance schedules each week, it is essential to have an accurate sixteen-week plan so that the resource allocations, external contractors, activities prioritisation, major overhaul tasks, and assets activities are planned weeks ahead. Moreover, in a complex refinery site, the assets are interrelated. Therefore, an excellent sixteen-week plan must consider asset maintenance activities to reduce the bad clashes and increase desired alignments between each maintenance activity. Currently, most maintenance schedules are prepared manually, which is inefficient and time-consuming. This project will develop optimisation models to produce a sixteen-week plan, taking into account all requirements in Alcoa (time window, resource limitation, prioritisation) and at the same time maximising good alignments and reducing clashes.
This theme was continued by Melinda Hodkiewicz.
Use of T3 scheduling expertise to inform down day planning for timing, budget and resource needs #
Melinda presented a case study to develop a down day scheduling support tool co-developed with BHP NW refinery by Theme 3. Melinda described: a) the technical development and b) value from stakeholder input to the process. The complexity of building any scheduling tool for a complex process plant down day planning was discussed.
In March, the presentations were based around Theme 1 research #
Tyler Bikaun presented Semi-automated Estimation of Reliability Measures from Maintenance Work Order Records #
Determining reliability measures such as mean-time-between-failure (MTBF) for in-service assets is an essential process. Statistical distributions of end-of-life values are used to assess asset reliability performance and the effectiveness of maintenance strategy. However, identifying the end-of-life event for each instance of functional failure is an arduous manual process dependent on structured and unstructured fields in the maintenance management system and rules used by individual reliability engineers. Tyler presented a methodology for emulating the process of end-of-life event detection using a natural language processing pipeline and a proposal for statistical parameter estimation to produce MTBF values for in-service assets from maintenance work order data was discussed and evaluated.
Michael Stewart presented Redcoat: A Collaborative Annotation Tool Supporting Technical Language Processing Research #
Technical language such as maintenance work orders contains a wealth of information, such as indicators of failure modes and end of life events. Knowledge graphs provide a powerful solution for unlocking this knowledge, allowing historic asset data to be visualised and queried to make informed business decisions. Michael provided a live demo of his knowledge graph for technical language and demonstrated how it can capture both structured and unstructured information. He also explained how the knowledge graph is constructed via a live demo of Redcoat, an annotation tool. Redcoat is used extensively in Theme 1 to build a labelled training dataset for technical language processing.
Theme 4 Presentations in May #
Dr Eden Li presented: Are we ready to transform? Leadership capability for adopting data science solutions #
One of the objectives of theme 4 research is to investigate the capabilities required for both individuals and organisations to unleash the potential of data science innovation, thus promoting maintenance transformation. To fulfil this objective, a placement project with the industry partner Roy Hill is currently being conducted. The project is to link the capabilities of maintenance supervisors with the uptake of data science solutions within their team. The Roy Hill Maintenance Work Management Process establishes the workflow by which workers identify, prepare, and complete all maintenance activities. Work management provides the foundations for maintenance strategy execution and continuous improvement, contributing to stable equipment reliability and availability for production. The skills and capabilities required for frontline workers to understand and conduct the work management process with the application of advanced data science solutions have not been clearly identified. We need to ensure that our workers know what they need to do in their role and have the skills to meet their business's objectives. Given future development and business transformation in the new digital age, it is essential to deepen the understanding of the current skill requirements and the capabilities needed to thrive in the digital future.
In June, our researcher presentations were delivered by our UWA PhD candidates in Theme 2 #
Tim Pesch presented his PhD research topic Sequential Order Statistics for Non-Identical Component Lifetimes #
Various technical systems are composed of multiple components. A system can fail once one of the components fails (series system), all components fail (parallel system), or a specific number of components fail (n-out-of-k system). The main aim of reliability analysis is to better understand the lifetime behaviour of the individual components and the whole system. . Estimating average lifetimes and predicting future failures governs how maintenance schedules, warranty bonds, or bulk replacement strategies are deployed. Tim presentation introduced a highly sophisticated model for technical systems consisting of components with different lifetime behaviours which share a common workload. Load-sharing systems operate under the assumption that the failure of one component can alter the lifetime expectation of other components. This dependence structure will be represented using sequential order statistics. The distributional results will yield reliable estimators to better predict future component and system failures
Braden Thorne presented his research interests Reservoir Computing Approaches to Parameter Extraction with Applications #
Braden presented the task of determining parameters of dynamical systems from their time series, which he calls parameter extraction. He presented a series of models inspired by reservoir computing that map from some original time series to some static, high dimensional vector or feature. These random (in the sense of arbitrarily selected) feature maps (RFM) allow him to separate signals with different parameter values. Braden explored several RFM variations, presenting models in the time and frequency domain and considering models with and without time series embedding. For simulated systems throughout stable and chaotic regimes, he achieved accurate parameter extraction with significant robustness to key hyper-parameters of the models. He then assessed the RFM performance on an engineering application; cavitation detection in centrifugal pumps from vibration data. While this application introduces several problems for the models, Braden still observed some ability to separate signals with different underlying parameters (here, the system tends to cavitate).
In July, the presentations were focused on Theme 3, introducing our new Research fellow Dr Mojtaba Heydar. #
Dr Mojtaba Heydar presented A stochastic model for job assignment problem with random arrivals and processing time #
Mojtaba considered a stochastic assignment problem in which requests or jobs arrive randomly. Due to the nature of jobs, the processing times follow a known probability distribution. This assignment problem may arise in hospitals where patients must be assigned to the next and best available beds. He used a Markov decision process to model this problem. Mojtaba formulated a dynamic program recursion to optimise an objective function, calculated the optimal decision variables, and discussed helpful simulation techniques when the size of the problem is too large. He illustrated the theory with some numerical examples.
Ponpot Jartnillaphand, our 2021 honours student, presented his honours research. Crew Rostering Optimization in Maintenance Operations - Models and Solution Methods #
The dynamically changing workplace environment has resulted in the need for rostering algorithms to accommodate changes within an organisation. Determining crew and task scheduling is essential for any large and complex organisation, thereby becoming an active research topic in Operations Research. In his talk, Ponpot consider a generic crew scheduling model for maintenance operations. It involves various constraints, such as resourcing, timing, ordering and rostering that may suit the maintenance operations. The objective of the model is to maximise the weighted summation of completed jobs. For the solution methods, Ponpot is investigating the logic-based decomposition algorithm and its variants, shown to be efficient to deal with problems of this kind (large scale binary linear programming).
IN THE SPOTLIGHT #
Congratulations to Sandy Spiers on being awarded the 2020 Best Mathematics and Statistics 4th Year/Honours Student Award at Curtin University. Sandy completed his Honours Thesis with the Centre, under the supervision of Prof. Ryan Loxton and Dr Asghar Moeini.
We are very pleased Sandy has joined the Centre to complete his PhD under the supervision of Prof Ryan Loxton. Sandy's PhD focus is on maintenance optimisation for network-connected assets
Stay tuned for our next issue where we will cover:
- Industry placements
- New publications
- Research updates