Industry collaboration research projects

Identification of Point of Sale Insights by Mining Social Media

Patties Food PTY Limited, 2019-2021

Lead Investigator: Dr Lan Du


Machine learning to support the civil construction industry to create a safer future for employees-Platformers-Human-in-the-loop Analytics GRIP

GRIP, 2019-2022

Lead Investigator: Dr Lan Du


Data analytics for risk and safety data collected from construction sites

Funded by Vertical Matters and Department of Industry, Innovation and Science, 2017

Lead Investigator: Prof. Geoff Webb

The Monash team explored the relation of risk and safety data to work activities and incidents from construction sites looking for insights, correlations and causal influences, such as sites and injury types. The results are being used to establish a pilot prototype of an industry workforce risk and activity planner system to develop a true benchmark of safety and performance on a construction site. The aim - a reduction of safety injury incidents and claims.

This project exploration and insights on the relation of risk and safety data to work activities from construction sites. This includes specific reference to identification of risks and the management of such via safe work method statements (swms). The overall objective of this project is to establish a pilot prototype that will serve as the foundation of the development of an industry workforce risk and activity planner system. This system will operate in real time to provide a measure of a company culture indices in order to determine the true benchmark of safety and performance in Construction. Thus, the prototype will provide models and insights that result in a reduction of safety injury incidents/claims. The project focus is to provide the following:

  • Comprehensive statistics about risk and safety data at scale (for stakeholders, construction sites, contractors, etc.)
  • Exploratory data analysis for risk and safety data.
  • Insights of the data for correlations and causal influences , such as sites, injury types, etc. That also includes analysis of planned work activities and actual work activities whenever accidents occurred.

Meromiktik Data Lake

Funded by Agilent Technologies

The Monash Data Science team are creating a data lake of powerful information sourced from production and field data and used to perform data mining and analytics work. They have discovered new patterns and associations that will minimise time, effort and costs in encountering and troubleshooting problems in referencing large amounts of data. There are significant benefits and learnings from this project: minimising customer complaints and warranty issues to improve service delivery, and creating a design philosophy that facilitates focused data analytics from the electronics and firmware of future platforms.

Scientific instrument platforms create significant analytical, diagnostic and performance data, most of which is collected and analysed on a one-off basis.  Some is never referenced until a fault occurs and an attempt is made to diagnose the issue(s).  Currently value is gained from this data by manually examining it or using basic analytics tools (often Excel). This value can accrue to both customers and Agilent.

This project creates a data lake of powerful information from production and field data that is utilised to perform data mining and analytics work. The project aims to build models of the platform to find insights in production and field data in order to discover new patterns and associations, which therefore can minimise domain expert effort in encountering and troubleshooting problems that occur in production and later in deployment.  Automating the discovery and prediction of consequent errors encountered both in production and the field will minimise customer quality and warranty issues and improve service delivery.

Another significant benefit of the learnings out of this project is the opportunity to create a design philosophy that facilitates focused data analytics from the electronics and firmware of future platforms. This should allow modelling of future instruments, components and subsystems. Models associated with particular analysis domains and types can conceivably also be created. The focus will initially be atomic spectroscopy.


Polychromator characterisation and optimisation

Funded by Agilent Technologies, 2018

Our work is developing a higher quality polychromator through an automated process to understand why one might perform significantly better than others. One objective of this work is to understand why this happens and what parts of the supply chain, design and aligning process impact the overall performance.

The polychromator in a 51x0 instrument is one of two highly influential components that impact the results that a customer gets from analysis on the platform. The other is the sample introduction workflow. The intent of this work is to characterise, understand and gain insights on the process of manufacturing the polychromator, the supply of the components into it and then feed this information back into the NPI process to develop a better performing and higher quality polychromator.

The process in manufacturing used to be human tuning of the design. This process changed to being an automated one in which a multi-axis robot tunes four polychromators at the same time generating a large amount of information in the process documenting the process, the state of the polychromator and operation of the robot.

The outcome of the automated process can still leave an instrument as performing over a range above the minimal expectation or requirement. The concept of a “hot” instrument is one that performs significantly better than others in the view of our customers. One objective of this work is to understand why this happens and what parts of the supply chain, design and aligning process impact the overall performance.

This information is then intended to be fed back into the NPI process to allow the design of the polychromator to be more consistent around the level of the better performing instruments, or "hot" instruments.

This information can then be used to gain a deep understanding of the variation across the polychromators built in OF and when combined with data that we gather from the use of the instrument over its lifetime (internally used and when customers allow us data) give us insights on how our instruments perform over their lifetime. These insights can be related to dealing with maintenance issues, overall performance degradation and opportunities to provide better analytical results to our customers.


Cirrus: an automated mammography-based measure of breast cancer risk

Project lead: Dr.Daniel Schmidt

Mammograms are routinely collected in Australia and overseas as part of standard screening programs for the early detection of breast cancer. These scans show masses of the tissue but can also predict the future risk of developing the disease by determining the amount of "bright" pixels in a mammogram. This technique is a semi-automated labour intensive, subjective and inappropriate for routine use in a clinical setting. To overcome these obstacles and produce a risk measure for a woman directly from her mammogram, the research team created Cirrus - a predictive model that achieves a level of accuracy close to the state-of-the-art genomic risk prediction systems. The Cirrus project uses image processing and machine learning techniques to automatically discover the textural features of a mammogram that are most predictive of future risk of breast cancer. Further work is required to improve its robustness to the differences between mammography machines, as well as integrating it into the Volpara medical imaging system that is currently deployed in the United States. The Cirrus project is joint-research undertaken in collaboration with Faculty of Information Technology (Dr. Daniel Schmidt), the Centre for Epidemiology and Biostatistics, The University of Melbourne (Prof. John Hopper, Dr. Enes Makalic) , BreastScreen St Vincent's (Dr. Helen Frazer) and Volpara Solutions (Dr. Ralph Highnam), and is supported by an NHMRC grant (APP1147764), a Cancer Council Victoria grant-in-aid and a National Breast Cancer Foundation Investigator Initiated Research Scheme grant (IIRS-18-093).


Movember15: Improving men's access to care: a national ambulance approach to reduce suicide and to improve the mental health of men and boys

(Movember Foundation, 2015-ongoing)

Project lead: Prof.Geoff Webb

A world’s-first program presenting a unique opportunity for delivery of an early, effective and sustainable public health response for male mental health issues that has national coverage and reach.

Males frequently suffer with mental health problems, yet they are often reluctant to seek professional help. When males do present for help, this is most often in the context of an acute physical illness or crisis. Given their limited contact with health providers, it is important that when males do make contact, their mental health needs are appropriately identified and early, targeted interventions are applied. A key frontline emergency service frequently accessed by men and boys across Australia is the ambulance service in each jurisdiction, with around 1.2 million attendances to male patients annually. However, such contact does not necessarily result in linkage to ongoing care. This presents a unique opportunity for enhancing help-seeking and access to effective treatment for men and boys by building an innovative national public health response. Through a national assessment of male mental health presentations to ambulance, emergency departments and hospitals, this project will comprehensively map the needs of men and boys identifying key intervention points for linkage to appropriate care. This work will inform the development of targeted training aimed at enhancing the skills and confidence of paramedics to support men with mental health problems, as well as a suite of low-cost population-level interventions that will be trialled in multiple jurisdictions.


An Innovative Integrated Algorithms for Cost-Effective Management of Water Pipe Networks

Project lead: Dr. David Albrecht

We are observing the buried water pipe networks of urban Victoria. Most located in congested cities. All are over 50 to 100 years old. A water main failure incurs significant economic and societal costs and is a major challenge to water authorities. We are working to facilitate effective management of these vital assets by developing integrated pipe failure prediction and asset management algorithms, and modules for pipe renewal and rehabilitation.

Currently, water authorities work with pipes above and below a 300 mm diameter. The size of the pipe determines a risk ranking and a pipe renewal and replacement approach. These separate approaches are natural since large diameter pipes, also known as critical pipes, fail less often but their failures can lead to major consequences. Modelling the failure of these larger pipes is complex. Our researchers have less failure data to work with and must consider more variables, as failures are impacted by internal (pressure and transients) and external (corrosion, traffic loads etc) factors around the pipe barrels and joints. There is more failure data for small diameter pipes, but the pipe failures depend significantly on reactivity of soils and prevailing climate.

The project will work in partnership with the current global project on Advanced Condition Assessment and Pipe Failure Prediction Project (ACAPFP).


fungalAI - Anti-fungal stewardship and infection control

Project lead: Prof. Geoff Webb

Monash Data Science is one member of a cross discipline research team that developed this platform to understand our local epidemiology, patient outcomes and improve our antifungal stewardship practice.

Invasive fungal infections are rare, neglected diseases that cause a life-threatening pneumonia in patients with impaired immunity such as those associated with chemotherapy or organ transplants. Treatments are limited, costly and have significant side effects. Healthcare systems cannot monitor the effectiveness of their work. Stewardship and infection control needs high quality data. Ideally, that data should arrive as quickly as possible to best drive improvements in care. This has not been possible, until now.

fungalAi makes real-time surveillance of fungal diseases possible using an expert system integrating lab and drug information together with natural language processing and deep learning based image analysis of chest CT scans and reports

Read more about this world-first research project


Improving the Quality of Primary Health Care for TAC Clients

Project lead: Prof. Reza Haffari

Advanced data science and artificial intelligence techniques for time series analysis and sequential decision making are leading the way for improved care of patients with chronic and complex health conditions acquired as a result of a car accident. Requested by the Transport Accident Commission of Victoria, the Monash team of DataScience researchers analysed patients' claim data to design approaches for modelling and recommending health service usage. As a result, the TAC now implements an AI-enabled strategy to patient care that is delivering better outcome, i.e. patient recovery with less time and cost.


Data Analysis of Victoria Police Incidents and Injuries Data

Project lead: Prof. Reza Haffari

Victoria Police work tirelessly to protect us all but they have a duty of care to protect their force, too. Working with Monash researchers specialising in data science and artificial intelligence techniques, this project investigated the patterns and causes of incidents and injuries within the Police force with the aim to reduce incidents. Unstructured injury data reports were analysed using advanced text mining techniques, to identify various events and their causal relationships. The Monash research supported Victoria Police to design and implement more effective prevention programs resulting in reduced incidents.


Datafication as a trans-disciplinary tool to refine, understand and address a range of complex real-world situations

Project lead: Prof. Wray Buntine and A/Prof. Henry Linger


Effective Multi-Task Learning for Neural Machine Translation

(Amazon Research Award, 2019-2020)

Project lead: Prof. Reza Haffari


Effective Training of Document Neural Machine Translation

(Google Faculty Research Award, 2019-2020)

Project lead: Prof. Reza Haffari


Yonder: Industry workforce Risk prediction and Activity Planner System

(Victorian State Government, 2019)

Project lead: Dr. Mahsa Salehi

The construction industry has one of the highest number of serious claims, injuries and illnesses due to the hazardous nature of the work. The aim of this project is to show the capability of machine learning techniques to provide useful predictive information about the risk of different projects in construction sites. We then will explore the possibility of proposing a global schema for further data collection, exploration and analysis.


Broadspectrum: Plant & Equipment Utilisation Review Project

(Broadspectrum, 2019-2020)

Project lead: Dr. Lan Du

An opportunity exists for Broadspectrum to collaborate with the Monash Department of Data Science and Artificial Intelligence for a one-year program aimed at advancing a range of unique machine learning and artificial intelligence techniques to optimise plant and equipment utilisation. The findings of this research will have the potential to improve, cost efficiency, capital allocation and plant and equipment utilisation across a plethora of industry sectors. The proposed project will focus on utilising various advanced machine learning and data mining technologies to analyse historical data that is available from systems such as IVMS and BRSIMS. The aim in analysing this data is to gain insights that can potentially benefit Broadspectrum in terms of optimising resource utilisation, and exploring the opportunity and viability of injecting further AI into its daily management.