Scribd
Upload
82 views18 pages

Machine Learning For Bridge Damage Modelling:: Towards Flood Resilience of Transportation Networks

The document discusses using machine learning to model bridge damage from floods. It notes that floods are increasing and many bridges are threatened. Available damage models rely on limited bridge data. The research aims to use flood depths, bridge attributes, and strains to train machines to estimate damage curves and classify bridge damage without extensive data collection.

Uploaded by

Giuseppe Degan Di Dieco
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
82 views18 pages

Machine Learning For Bridge Damage Modelling:: Towards Flood Resilience of Transportation Networks

The document discusses using machine learning to model bridge damage from floods. It notes that floods are increasing and many bridges are threatened. Available damage models rely on limited bridge data. The research aims to use flood depths, bridge attributes, and strains to train machines to estimate damage curves and classify bridge damage without extensive data collection.

Uploaded by

Giuseppe Degan Di Dieco
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 18

MACHINE LEARNING FOR BRIDGE DAMAGE

MODELLING:
Towards flood resilience of transportation
networks

Turing Community Week


09 September 2021

Giuseppe Degan Di Dieco, University of Bristol


Flood impact to bridges
o Flood events doubled in the last o Globally, hydraulic actions (e.g.
20 years; scour, hydrodynamic load)
o Floods have the highest represent the third cause of bridge
percentage of occurrence (44%) failure, after construction errors
among all the natural hazards. and collision and accidents.
Flood losses to bridges
o 7,5% of EU, UK, US and China o In the UK, the 2015 Storm o In addition, the four-month
riverine bridges (61,000 assets) Desmond caused damages to closure of the A591 road alone
will be threatened by floods in the 557 bridges in the Cumbria caused a loss of £1M per day to
next decades; County Council (CCC; Northwest the regional economy;
England) alone; o The A591 is a North-South
o The reconstruction costs of these Cumbria link, connects the
bridges were estimated in £117 County to the M6 motorway, and
M. crosses the renown touristic area
of the Lake District National Park.
State-of-the-art in damage estimates
Typical
bridges and Flood
failure loading
mechanisms

Sets of
damage
curves

Damage model

o Damage models are used to o Available damage models (e.g., o A damage model for UK riverine
assess the physical and George and Menon, 2021) rely roadway bridges is lacking;
economical impact of natural on data (e.g. piers geometry) that o Remedial actions to reduce
hazards to large number of are not being routinely collected knock-on effects of severe events
assets. by UK infrastructure owners. are impaired;
o Data needed for its development
are currently not available.
Research challenge
o Can computers learn how to estimate damage probability, given flood depths and basic bridge
attributes as input, and bridge strains as output?

Output
Bridge
strains for
different Training
damage
levels Use of the training
dataset for
teaching to various
machines how to
assign damage
states tags to
strains of different
bridge types
Case study
CUMBRIA DEVON LANCASHIRE
o Three winter floods occurred in o Seven winter floods hit the Devon o Two winter floods hit the
Cumbria in the last 12 years; County Council in the last eight Lancashire County Council in the
o 817 bridges damaged between years; last nine years;
2009 and 2015 floods; o 23 bridges were damaged in this o 260 bridges closed after 2015
timeframe; Storm Desmond;
o CCC currently manages 1,703
riverine roadway bridges. o DCC currently manages 2,425 o LCC currently manages 1,241
riverine roadway bridges. riverine roadway bridges.
Machine Learning in Civil Engineering
and Methodology
Reference Take-home message

Ji et al. (2017) Least-Squares Support Vector Machine (LSSVM) for damage curves derivation of slopes

Liao and Hsiung (2017) LSSVM for reliability analysis of seawall when tsunami historic records lack

Kiani et al. (2019) Classification-based tools, e.g., Random Forest, for seismic damage curves of steel buildings

Mangalathu et al. Classification-based tools, e.g., Random Forest, for seismic damage curves of bridge networks
(2019)

Liao et al. (2019) LSSVM for scour-earthquake damage curves of single bridges

Khandel and Deep Learning Networks for river flows and damage curves estimation of single bridges
Soliman (2020)

Sarailidis et al. (2021) Interactive Decision Trees for including experts’ knowledge within machine learning
o Least-Squares Support Vector Machine,
Random Forest, Deep Learning Networks and
Decision Trees;
o Flood depths, bridge geometry, and bridge
strains are needed to train a machine;
o Flood depths data are widely available by state
responsible agencies;
o Bridge strains are either derived via numerical
models or measured via monitoring, or via
Monte Carlo methods;
o Bridge attributes are partially available via asset
owners datasets;
o Current practice groups bridges in classes, thus
only the representative bridge is modelled;
The algorithm of Support Vector Machines (SVM).
Pros and Cons
Pros Cons
Reducing
computational
st
burden What are the mo
es?
reliable techniqu
Encouraging And Why?
bridge response
data sharing Are Machine
te
Learning surroga
Avoiding future models
bridge data appropriate?
collection
Summary
o Floods have the highest frequency, which is Flood depths,
expected to increase; basic bridge Training several
o Around 61,000 riverine bridges are attributes and machines
bridge strains
threatened globally;
o Available damage models rely on bridge data
not yet collected in the UK;
Damage curves
Assessing
o Machine learning could reduce computational and bridge
machines
burden and avoid future data collection; classes
accuracy
derivation
o What Machine Learning techniques better
identifies bridge damages yet to be found;
o A trained machine could help decision Damage models
makers in time-remedial actions. for riverine
roadway bridge
networks
Get in touch!
[email protected]

@pep_4_climate

@3dg90

Research Team

Dr Maria Pregnolato

Dr Andre R. Barbosa

You might also like