DEMONSTRATION
Predictive Maintenance
Water Pipe Failure Prediction & Optimisation
THE CHALLENGE
A water utilities company wanted to start taking a proactive approach to pipe maintenance.
Replacing a pipe before it breaks is significantly cheaper
and far easier to manage operationally
Fixing already burst pipes was highly expensive, time
consuming, disruptive and damaging to their reputation
Which pipes should be replaced first to maximise the financial benefit?
THE DATA
Pipe metadata
The age, material, location, size, soil type, distribution
zone, water level, etc. for each pipe in the network
Maintenance history
The date, time, work type, location, pipe id, cost etc. for
every pipe fix in the last 20 years.
Data is from an open dataset on water pipeline infrastructure.
OUR SOLUTION
User portal
The central hub where users can see output from the
optimisation and PIPE ML in an actionable format
Financial Optimisation
The system that uses predictions from PIPE ML to find the financially optimal
maintenance strategy, given engineer availability constraints and cost weightings
PIPE ML – Predictive model for pipe failure
Using the data available, we used machine learning to predict
the likelihood that a given pipe would fail in the next 3 months
OUR APPROACH
maintenance data
transform
& clean
Predictions
Insights
Model metadata
user portal
Optimal Maintenance
Plan
Engineer Allocations
Financial Impactpipe metadata
additional data
(e.g. IoT sensors)
MODEL EVALUATION
➔ Given a limited capacity, there are two main strategies a company could follow to schedule predictive maintenance tasks:
Oldest pipes first or using the PIPE ML risk score
Oldest Pipes First
Giving predictive maintenance to pipes
according to their age.
This method is a step towards predictive
maintenance, but still only uses one
variable to prioritise works.
Attending pipes with the highest risk
scores according to the PIPE ML model.
This method utilises all the variables
available to identify the pipes with the
highest odds of breaking.
PIPE ML Risk Score
MODEL EVALUATION
➔ Using the model to prioritise which pipes to fix is 9 times more accurate than fixing the oldest pipes first
0.23% 0.30%
2.70%
Random Oldest PIPE ML Risk
Accuracy for the top 1000 pipes
1.3x
9x
FINANCIAL MODELING
➔ Through an exploratory analysis of the map we found a sample of areas where different pipe widths correlated with the
number of households. Wider pipes act as main arteries for carrying water through the city, while narrower pipes serve a
smaller number of households
Red is widest and
blue is narrowest
FINANCIAL MODELING
➔ Pipes were clustered into 5 groups according to their diameter, which is the variable we used to estimate
their importance to the network and the additional cost a breakage would represent
- Maintenance work done on a pipe
before it breaks
- Cost: 1x
PROACTIVE
- Work done on a pipe after a breakage
- Cost: 2x
REACTIVE
WORK TYPES PIPE IMPORTANCE
CLUSTER
PIPE
DIAMETER (CM)
HOUSEHOLDS COST
1 48 15 1x
2 99 50 3.3x
3 148 300 20x
4 224 1000 66.6x
5 300 5000 333.3x
FINANCIAL MODELING
➔ PIPE ML allows the company to save 9.5% in total maintenance cost compared to the ‘oldest pipes’ methods
£323
£290
Oldest
PIPE ML risk
Cost of reactive maintenance
(£m)
Executive Overview
Current status of the network as a whole
High level risks and daily summary
Map View
Immediate visual feedback on risk areas
User selected colour coding
Engineer Allocation
Output from the optimisation model, showing
how engineers should be allocated to tasks
Financial Analysis
A financial breakdown of the benefits for
proactively maintaining pipes
Pipe View
Drill down into individual pipe data to see
actionable insights and metadata
Machine Learning Monitoring
Output metrics from the ML model can be
monitored in the platform by data scientists
KEY FEATURES
Our London based team has a proven track record delivering bespoke data
projects for a wide range of ambitious clients.
We’re a data science consultancy building innovative AI solutions.
We favour a highly communicative approach - we take the time to meet with
your team and establish the optimal way to utilise your data.
On delivery of each project, we ensure a smooth transfer of knowledge to
internal stakeholders, both from a technical and non-technical perspective.
If you are interested in finding out more about our services and how they can
transform your business, get in touch and we'd be happy to tell you more.
ABOUT US
hello@adsp.ai

Predictive Maintenance

  • 1.
    DEMONSTRATION Predictive Maintenance Water PipeFailure Prediction & Optimisation
  • 2.
    THE CHALLENGE A waterutilities company wanted to start taking a proactive approach to pipe maintenance. Replacing a pipe before it breaks is significantly cheaper and far easier to manage operationally Fixing already burst pipes was highly expensive, time consuming, disruptive and damaging to their reputation Which pipes should be replaced first to maximise the financial benefit?
  • 3.
    THE DATA Pipe metadata Theage, material, location, size, soil type, distribution zone, water level, etc. for each pipe in the network Maintenance history The date, time, work type, location, pipe id, cost etc. for every pipe fix in the last 20 years. Data is from an open dataset on water pipeline infrastructure.
  • 4.
    OUR SOLUTION User portal Thecentral hub where users can see output from the optimisation and PIPE ML in an actionable format Financial Optimisation The system that uses predictions from PIPE ML to find the financially optimal maintenance strategy, given engineer availability constraints and cost weightings PIPE ML – Predictive model for pipe failure Using the data available, we used machine learning to predict the likelihood that a given pipe would fail in the next 3 months
  • 5.
    OUR APPROACH maintenance data transform &clean Predictions Insights Model metadata user portal Optimal Maintenance Plan Engineer Allocations Financial Impactpipe metadata additional data (e.g. IoT sensors)
  • 6.
    MODEL EVALUATION ➔ Givena limited capacity, there are two main strategies a company could follow to schedule predictive maintenance tasks: Oldest pipes first or using the PIPE ML risk score Oldest Pipes First Giving predictive maintenance to pipes according to their age. This method is a step towards predictive maintenance, but still only uses one variable to prioritise works. Attending pipes with the highest risk scores according to the PIPE ML model. This method utilises all the variables available to identify the pipes with the highest odds of breaking. PIPE ML Risk Score
  • 7.
    MODEL EVALUATION ➔ Usingthe model to prioritise which pipes to fix is 9 times more accurate than fixing the oldest pipes first 0.23% 0.30% 2.70% Random Oldest PIPE ML Risk Accuracy for the top 1000 pipes 1.3x 9x
  • 8.
    FINANCIAL MODELING ➔ Throughan exploratory analysis of the map we found a sample of areas where different pipe widths correlated with the number of households. Wider pipes act as main arteries for carrying water through the city, while narrower pipes serve a smaller number of households Red is widest and blue is narrowest
  • 9.
    FINANCIAL MODELING ➔ Pipeswere clustered into 5 groups according to their diameter, which is the variable we used to estimate their importance to the network and the additional cost a breakage would represent - Maintenance work done on a pipe before it breaks - Cost: 1x PROACTIVE - Work done on a pipe after a breakage - Cost: 2x REACTIVE WORK TYPES PIPE IMPORTANCE CLUSTER PIPE DIAMETER (CM) HOUSEHOLDS COST 1 48 15 1x 2 99 50 3.3x 3 148 300 20x 4 224 1000 66.6x 5 300 5000 333.3x
  • 10.
    FINANCIAL MODELING ➔ PIPEML allows the company to save 9.5% in total maintenance cost compared to the ‘oldest pipes’ methods £323 £290 Oldest PIPE ML risk Cost of reactive maintenance (£m)
  • 11.
    Executive Overview Current statusof the network as a whole High level risks and daily summary Map View Immediate visual feedback on risk areas User selected colour coding Engineer Allocation Output from the optimisation model, showing how engineers should be allocated to tasks Financial Analysis A financial breakdown of the benefits for proactively maintaining pipes Pipe View Drill down into individual pipe data to see actionable insights and metadata Machine Learning Monitoring Output metrics from the ML model can be monitored in the platform by data scientists KEY FEATURES
  • 12.
    Our London basedteam has a proven track record delivering bespoke data projects for a wide range of ambitious clients. We’re a data science consultancy building innovative AI solutions. We favour a highly communicative approach - we take the time to meet with your team and establish the optimal way to utilise your data. On delivery of each project, we ensure a smooth transfer of knowledge to internal stakeholders, both from a technical and non-technical perspective. If you are interested in finding out more about our services and how they can transform your business, get in touch and we'd be happy to tell you more. ABOUT US hello@adsp.ai