Structural Health Monitoring (SHM) techniques are able to monitor the behaviour of critical infrastructure over time, by improving the safety and reliability of the asset. A large amount of data is generated by SHM methods continuously. Therefore, machine learning methods can be developed in order to transform the available data into valuable information for decision makers, by pointing out vulnerabilities of the critical infrastructure. In this paper, a machine learning classifier for condition monitoring and damage detection of bridges is proposed by adopting a Neuro-Fuzzy algorithm. The method allows to assess the health state of the infrastructure automatically, accurately and rapidly, every time when a new measurement of the bridge behaviour is available. The method is validated and tested by monitoring the behaviour of an in-field steel truss bridge, which is subjected to a progressive damage process.
2. Matteo Vagnoli, Rasa Remenyte-Prescott, John
Andrews
A machine learning classifier
for condition monitoring and
damage detection of bridge
infrastructure
3. Outline
• Where we are today
• Bridge condition monitoring and damage
detection using machine learning
• A case study
15. Bridge condition monitoring today
Bridge condition assessment and damage detection strategies are
usually carried out by subjective visual inspection, at intervals of
one to six years.
16. Bridge condition monitoring today
Bridge condition assessment and damage detection strategies are
usually carried out by subjective visual inspection, at intervals of
one to six years.
More than 35% of the over 1 million bridges across Europe are
over 100 years old.
European Commission, EU transport in figures, statistical pocketbook, 2012.
17. Bridge condition monitoring today
Bridge condition assessment and damage detection strategies are
usually carried out by subjective visual inspection, at intervals of
one to six years.
Deterioration processes may lead to a lower safety level and,
potentially, to catastrophic events.
More than 35% of the over 1 million bridges across Europe are
over 100 years old.
European Commission, EU transport in figures, statistical pocketbook, 2012.
18. Bridge condition monitoring today
Bridge condition assessment and damage detection strategies are
usually carried out by subjective visual inspection, at intervals of
one to six years.
Deterioration processes may lead to a lower safety level and,
potentially, to catastrophic events.
How can we change the way
we approach this problem?
More than 35% of the over 1 million bridges across Europe are
over 100 years old.
European Commission, EU transport in figures, statistical pocketbook, 2012.
20. Bridge condition monitoring tomorrow
Real-time
measurement
system
✓Remote structural health
monitoring and damage
detection
✓Overcoming of the visual
inspection limitations
✓Assessment of the health
state of the whole bridge
by analysing the bridge
behaviour
✓Maintenance can be
scheduled based on the
real health state of the
bridge
21. Bridge condition monitoring tomorrow
Real-time
measurement
system
✓Remote structural health
monitoring and damage
detection
✓Overcoming of the visual
inspection limitations
✓Assessment of the health
state of the whole bridge
by analysing the bridge
behaviour
✓Maintenance can be
scheduled based on the
real health state of the
bridge
22. A Method for bridge condition monitoring
Identification of the
bridge free-vibration
Assessment of 22
features over time
Assessment of feature
trend (using EMD)
Identification of the
bridge health state
using a Neuro-Fuzzy
classifier
Raw data from sensors
23. A Method for bridge condition monitoring
Identification of the
bridge free-vibration
Assessment of 22
features over time
Assessment of feature
trend (using EMD)
Identification of the
bridge health state
using a Neuro-Fuzzy
classifier
Raw data from sensors
24. A Method for bridge condition monitoring
Identification of the
bridge free-vibration
Assessment of 22
features over time
Assessment of feature
trend (using EMD)
Identification of the
bridge health state
using a Neuro-Fuzzy
classifier
Raw data from sensors
25. A case study: steel truss bridge
8@7400= 59200 mm
P1 P2
A1 A2 A3 A4 A5
A6 A7 A8
DMG1
DMG2 DMG3
Passing direction
Ai: Accelerometer No. i (Vert.)
DMGi: damage scenario i
Pi: Pier No.i
26. Raw acceleration of the bridge
Identification of the
bridge free-vibration
Assessment of 22
features over time
Assessment of feature
trend (using EMD)
Identification of the
bridge health state
using a Neuro-Fuzzy
classifier
Raw data from sensors
27. Bridge free-vibration identification
Identification of the
bridge free-vibration
Assessment of 22
features over time
Assessment of feature
trend (using EMD)
Identification of the
bridge health state
using a Neuro-Fuzzy
classifier
Raw data from sensors
Free vibration
28. Features assessment
Identification of the
bridge free-vibration
Assessment of 22
features over time
Assessment of feature
trend (using EMD)
Identification of the
bridge health state
using a Neuro-Fuzzy
classifier
Raw data from sensors
29. Trend of the feature using the EMD method
Identification of the
bridge free-vibration
Assessment of 22
features over time
Assessment of feature
trend (using EMD)
Identification of the
bridge health state
using a Neuro-Fuzzy
classifier
Raw data from sensors
30. Automatic classification of the bridge health state
Identification of the
bridge free-vibration
Assessment of 22
features over time
Assessment of feature
trend (using EMD)
Identification of the
bridge health state
using a Neuro-Fuzzy
classifier
Raw data from sensors
31. Automatic classification of the bridge health state
Identification of the
bridge free-vibration
Assessment of 22
features over time
Assessment of feature
trend (using EMD)
Identification of the
bridge health state
using a Neuro-Fuzzy
classifier
Raw data from sensors
The health state of the bridge has been correctly
identified for 6 scenarios out of 6. The nature of the
damage for 3 scenarios out of 4.
The proposed method allows to automatically
monitor and assess the health state of the bridge
32. Automatic classification of the bridge health state
Identification of the
bridge free-vibration
Assessment of 22
features over time
Assessment of feature
trend (using EMD)
Identification of the
bridge health state
using a Neuro-Fuzzy
classifier
Raw data from sensors
The health state of the bridge has been correctly
identified for 6 scenarios out of 6. The nature of the
damage for 3 scenarios out of 4.
The proposed method allows to automatically
monitor and assess the health state of the bridge
The proposed method has been verified on more
challenging in-field bridges and very good results
have been obtained
33. Automatic classification of the bridge health state
Identification of the
bridge free-vibration
Assessment of 22
features over time
Assessment of feature
trend (using EMD)
Identification of the
bridge health state
using a Neuro-Fuzzy
classifier
Raw data from sensors
The health state of the bridge has been correctly
identified for 6 scenarios out of 6. The nature of the
damage for 3 scenarios out of 4.
The proposed method allows to automatically
monitor and assess the health state of the bridge
The Neuro-Fuzzy requires a database of bridge
behaviour for the training process
The proposed method has been verified on more
challenging in-field bridges and very good results
have been obtained
34. Conclusion
Improve safety, reliability and
performance of the transportation
network
Improve the maintenance and
renewals activities of the assets
by optimizing their budget
How can we achieve
these objectives?
Real-time automatic
condition monitoring?
35. The TRUSS ITN project (http://trussitn.eu) has
received funding from the European Union’s
Horizon 2020 research and innovation
programme under the Marie Skłodowska-Curie
grant agreement No. 642453
Thanks for your attention