2. Based in France, operating worldwide
33 years of engineering, consulting, and innovation
Water
City and land use
IT and innovation
French Engineering Grand Prize Solar Impulse Efficient Solution label
3. Ageing water networks
Massive leakage
Climate change
Fragile water resources
Demography
Growing water demand
Energy costs
Carbon emissions
Access to water
6. EFFICIENT network replacement, the next challenge for water utilities
Which pipes to replace? Where to start?
How to get the maximum benefit out of every $ invested?
How to reduce Non Revenue Water to the minimum?
7. Risk: 0,86
Score: 204
Priority: 1
Solutions exist! Quantitative models leveraging machine learning…
Failure history determines the likelihood of future failures for every pipe in the network
9. Failure records + GIS data
HpO® AI : machine learning from past failure DATA
Failure
Date
Location
Type
Cause
Pipe
Diameter
Material
Optional data
Position
Laying bed
Type of soil
Coating
Etc...
10. HpO AI : how does machine learning work?
Thousands of combinations of pipe attributes are tested to identify statistically
significant failure rate groups
Material
Ductile iron
Cast iron
PVC
PEHD
PEBD
Asbestos cement
…
Diameter
32
40
60
75
80
90
100
110
125
150
200
250
…
Water
quality
Traffic
Strategic
High
Regular
Weak
Rare
Nil
Pressure
Average
Amplitude
Length of mains
Soil type
Laying date Climate
Watertable
depth
Neighbourhood
Other custom datasets
Past
failures
11. HpO AI : specific algorithms for a special type of survival analysis
We know only a fraction of the lifespan and leaks of the networks
1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 2020 2030 2040
Observation frame on a fraction of the lifespan and leaks
100%
0%
Likelihood
of
failure
Years
Pipe survival curves
Likelihood of failure is not constant in time
Group 1: Ductile iron 100-300 1990
Group 2: Ductile iron DN100-300 1970
Group 3: Steel 200-600 1960
Group 4: PRV 150-250 2000
Group 5: PEHD 40-100 1990
Group 6: Mixed 40-200 before 1950
Group 7: Cast iron 100-300 1950
Group 8: PVC 63-110 1960
Today
12. HpO AI : the main outcome of the AI/machine learning process
One new criterion to become the core of the decision-making process
Likelihood of failure
Obtained from machine learning with confidence
13. HpO AI : is artificial intelligence reliable?
Yes! Machine learning only means computing power. And all predictive models are verified.
2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021
Clean failure history dated and associated to network objects
2010 2011 2012 2013 2014 2015 2016 2017 2018
AI learns
AI calculates failure risks in the past future 2019 2020 2021
AI predicts compares its results with past reality 2019 2020 2021 AI model valid
2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021
AI applies
AI predicts failure risks in the real future 2022 2023 2024
14. HpO AI : verifying the AI predictive model
Comparing the efficiency of 4 network replacement strategies in terms of avoided failures
City of Noumea, New Caledonia
Good asset knowledge
12 years of robust failure records
Machine learning failure prediction is
able to identify 44% of future failures in
just 5% of the network length
More efficient than approaches based on
age, past failures or roadwork
coordination
For the same replacement budget
15. Lessons learned from the situation in France
2 billion € invested
every year for
water network replacement
Yet water network efficiency
desperately stagnates:
1 billion m³ lost every year
The most critical pipes are
NOT being replaced…
16. HpO AI : leveraging the insights of the AI/ML predictive model
The predictive model is placed at the core of the decision-making process
Likelihood of failure
You will always choose the most critical pipes
• Direct replacement
• Cross-asset coordination (sewer, gas, road…)
• Leak detection optimisation
• Feeding a more elaborate multicriteria analysis
17. HpO AI : leveraging the insights of the AI/ML predictive model
Likelihood of failure fits into a flexible multicriteria model customized for every utility
Likelihood
of failure
Impact on water distribution
Impact on sensitive customers
Impact on sensitive ressources
Impact on road traffic
Impact on OPEX
Impact on KPIs
Impact on utility’s image
Custom criteria in your context
x
18. HpO AI : you will always choose the most critical pipes
Boosting the efficiency of your asset management strategy with the same budget
Direct critical pipe
replacement
Cross-asset coordination
(sewer, gas, road…)
Leak detection
optimisation
Feeding a more elaborate
multicriteria analysis
Automated pipe clustering and replacement jobcards
19. Projects
1. R&D with the SEDIF (Greater Paris Area)
Largest metropolitan utility in France with 8,500 km of networks
Full prediction on mains, service connections and other buried equipment by
mining huge data sets / study of the impact of climate change
2. Orléans Métropole, France [completed]
3. Tours Métropole, France (ongoing)
4. Limoges Métropole, France [completed]
5. Chartres Métropole, France (recurrent)
6. Le Havre Métropole, France [upcoming]
7. Grand Annecy, France [completed]
8. Syndicat Rivière Woigot, France (ongoing)
9. Nouméa, France [completed]
10. Keetmanshoop, Namibia [completed]
11. Kampala, Uganda (ongoing)
12. San Miguel de Allende, Mexico (ongoing)
22. Essential data: good failure records over at least one year
HpO AI learns from YOUR DATA
there is no universal water network deterioration model
Failure
Date
Location
Type & cause
Mains
Diameter
Material
Laying period
Postion wrt road
Nominal pressure
Int & ext protection
Type of joint
Corrosion
Deposits
Environment
Laying bed
Type of soil
Condition of soil
External data
Minimum pressure
Maximum pressure
Pressure variations
Road traffic
Electric disturbances
Trees/plants
Water properties
Watertable
…
23. HpO Collect®
A free and simple app for the field agent
to collect failure data on pipes, service
connections and fittings
27. Consulting, expertise, software, technical assistance
A modular platform
for a facilitated and reproductible
asset management process
28. A modular platform
for a facilitated and reproductible asset management process
Data manager Data visualization
Map interface
Advanced data processing
Prediagnosis / prediction potential
Likelihood of failure calculation
Library/visualization of CCTV inspections
Mass processing of CCTV inspections
Targeting of future CCTV inspections
Multicriteria analysis
Prioritization
Clustering into operations
Sorting operations by efficiency
Your GIS
Your data
HpO® Collect app
HpO® Citizen app