Our Research

Predictive Failure Analysis

Water main failure prediction is the cornerstone of Data61’s water industry suite of solutions. The Data61 team has collaborated with 27 utilities around the world on failure prediction and related issues, and has tuned and refined its machine learning techniques.

Our non-assumptive, prediction techniques allow us to assess large populations of assets using readily available data to predict which assets are most likely to fail. We do this by allowing machine learning to take the guess work out of the equation. Letting the power of modern computing do the work.

Predictive Analytics Use Cases

The Data61 Team continues to innovate and invest significantly to re-shape its award winning predictive maintenance capabilities.
The primary areas of investigation that are of significant benefit to the water industry include:

  • Water main failure prediction
  • Sewer corrosion prediction
  • Sewer pipe choke prediction
  • Active leakage detection
  • Water demand analysis
  • Intelligent network optimisation

Water Main Failure Prediction

Data61’s adoption of nonparametric, Bayesian and stochastic point process methods for failure prediction has resulted in outcomes that have been proven to be more than twice as accurate as the currently heralded ‘industry standard’, the Weibull Method.
To take advantage of this capability, Data61 has established a clear and simple approach to execution in the field. Failure prediction and risk assessment projects are delivered across the following phases:

  1. Data pre-processing and visualisation
  2. Multi-factor analysis including water main attributes, geographic locations, weather, soil, pressure, water quality, unstructured data, etc
  3. Data-driven short/long-term water pipe failure prediction with confidence estimation
  4.  Optimisation and risk assessment based on incorporation of constraints such as geography, constructability, cost and consequence of failure, and budget constraints

The primary deliverables of this process are:

  • A listing of all pipes in your network ranked by probability of failure over a prediction period that can be defined by the end user
  • A ‘risk-rated’ ranking incorporating end user defined cost, consequence, budget and other constraints deemed to be key to allocation of scarce resources
  •  GIS representation of risky pipes and regions, with the ability to overlay critical client or community assets and locations
  • Clarity as to the factors that are most influential in the effective prediction of failure in your specific network
  • Identification of the next set of your data, or external data (e.g. weather, soil, foliage etc.) that if sourced will likely increase the accuracy of failure prediction
  • Configuration and deployment of Data61’s cloud-based Failure Prediction and Risk Assessment Platform
  • Establishment of an ongoing data management process that enables the end user to take control of timely update of your failure prediction and risk assessment process.

Sewer Corrosion Prediction

Leveraging Data61’s award winning main failure prediction work, Data61 developed a similar spatiotemporal model to predict sewer corrosion. The model has been specifically designed to overcome the problem of insufficient failure observations and reduces the prediction uncertainty of sewer corrosion.
This new data analytics model enables:

  •  H2S (and other parameters) estimation
  • Spatiotemporal corrosion prediction
  • Chemical dosing optimisation
  • Optimal sensor deployment
  • This solution helps water utilities to improve efficiency and save costs in chemical dosing, sewer pipe rehabilitation and sensor deployment.

Sewer Pipe Choke Prediction

Wastewater pipe blockages can cause significant social-economic cost. They have become one of the thorniest challenges for water utilities.
Based on the insights derived from wastewater pipe blockage prediction factor analysis, Data61 has developed a data-driven stochastic point process-based model for blockage prediction.
The technology developed by Data61 is designed to improve current reactive maintenance strategy, reduce maintenance cost and improve the reliability of continuous wastewater services.
Active Leakage Detection
Leakage is a concealed failure that occurs underground – unnoticed until it develops into an observable state resulting in high economic loss. Water utilities invest heavily to tackle leakage, and in the timely repair of pipes.
This solution helps to prioritise water network zones where an active leak detection program might add the most value. This is normally in pipes of less than 375 mm in diameter, with the following three major challenges being addressed:
 Factor complexity – identification of the most influential factors causing the leakage e.g. pipe age, material, pressure, traffic load etc.

 Probability of leakage – ranking the probabilities of each pipe leaking assists in the determination of which pipes or zones to inspect
 Zone selection with constraints – ranking pipes and zones based on a balance of failure prediction and other practical and budgetary constraints
The key deliverables from an active leakage detection project include:
 Quantitative correlation of single/multiple factors and leakage
 Probability of leakage ranking for each pipe
 Consequence and constraint ranking of pipes to support pipe inspection prioritisation
 Colour coded GIS maps
 Decision support tools

Water Demand Analysis

Data61’s collaborative work with water utilities in this area is designed to improve understanding of customer water use patterns and forecasting of customer water consumption in the future.
More specifically to help:

  • Provide analytic insights into use patterns and the most influential demand drivers
  • Understand water use pattern change following the lifting of restrictions and the potential for demand bounce back in the future
  • Conduct customer segmentation for water demand analysis based on historical consumption behaviours
  •  Improve water demand forecasting through machine learning techniques.

The team uses data-driven machine learning techniques to investigate the following water demand forecasting elements:

  • Dynamic customer segmentation – each customer’s membership over the segments can change over time in response to its consumption behaviour changes
  • Soft-membership – each customer can be assigned to all the segments in proportion
  • Time series analysis and potential applications for climate correction

Intelligent Network Optimisation

Data61 has worked with water utilities to process a variety of data generated within water networks to discover hidden patterns.
The primary objective of this work is to increase reliability of data-driven, operational decision making – with direct focus on dynamic optimisation of network operations to delicately balance key performance indicators in four areas:

  • Continuity of supply
  • Water quality
  • Energy consumption
  • Cost of operations
  • Data61 has used data-driven machine learning approaches to investigate challenges of the water quality analysis, chemical dosing estimation, and energy savings to:
  • Construct predictive models for water quality analysis
    • Spatial-temporal analysis
    • Water age estimation
    • Impact of water age on water quality
    • Impact of operational changes on water quality
    • Complex spatial-temporal dependencies towards chemical dosing estimation
    • Additional sites optimisation
    • Chemical dosing optimisation
  •  Energy saving at subsystem and whole system scales
    • Demand prediction
    • Factor analysis
    • Risk analysis
    • Joint optimisation