Seismic retrofit : challenges, opportunities and cost-effective solutions

«Seismic Retrofit» : Challenges,

Opportunities and Cost-Effective

Solutions

Prof. Dr. Ihsan Engin Bal

Keywords for this Presentation

•

Masonry buildings

•

Historical structures

•

Seismic strengthening

Outline

•

Experimental Studies

•

Strengthening of Historical Masonry

•

The test specimen was a frame taken from Mihrimah Mosque, built in the 16th century

by Sinan

•

The specimen was 5.0m long and 5.1m tall. It was 1/5 scale of the main

frame of Mihrimah Mosque

Test Frame

•

Authentic

materials

and

construction

methods are used

•

Metal connectors are placed, covered with

melted lead

•

An outdoor testing facility was built for the experiments

•

The deformations of the structure were monitored with 93 channels

•

Ambient vibration tests were conducted before and after the test

•

Optic measurement system is used to find the strain field at the bottom of the piers

Test Setup

Horizontal Loading

•

An increasing displacement pattern is applied

•

Application of the full pattern required 3 days of testing time

•

The specimen was pushed & pulled up to 2.4% top drift

•

F / W è 22 kN / 130 kN = 17% lateral load coefficient

Experiment –

Deformed Shape @ +2.4% Top Drift

(x5 exaggerated deformed shape)

1

1. Support opening of the main arch

[5.2cm], settlement of the

key-stone [2.6cm], and deformations

at the quarters [-5.7cm @ West

and +3.6cm @ East]

push

East

West

5.7cm

3.6cm

2.6cm

5.2cm

Separation of the main piers

from the small arches

Dislocation of stones

Experiment –

Deformed Shape @ +2.4% Top Drift

Top

Middle Bottom

Damage concentration

Experiment –

Deformed Shape @ +2.4% Top Drift

Push

Slipped vertical

connection

Experiment –

Deformed Shape @ +2.4% Top Drift

Two tie-rods

running parallel

Experiment –

Deformed Shape @ +2.4% Top Drift (x5)

Tension

Compression

Tie rod forces

increase from 10kN

to 73kN in case of

seismic loading

Glossary

Capacity

Demanded by

the Design EQ

Initial Capacity

Reduced Capacity (Small EQs, aging, etc.)

Improved Capacity after Repair

Strengthening

Repair

St

ru

ct

ur

al

Ca

pa

ci

ty

Is

it

ne

ce

ss

ar

y ?

?

Seismic Strengthening

•

60s-90s mostly RC jacketing

shotcrete, bracing

Trend in Seismic Strengthening

RC Jacketing & Shear Walls

Shotcrete on infill walls

Seismic Strengthening

•

Post-90s self-compacting concrete, FRP technologies

seismic isolation, dampers

Trend in Seismic Strengthening

Beyazıt Mosque, İstanbul

• Built between 1501 and 1506, Beyazit is the oldest royal mosque

(funded by Sultan himself) in Istanbul the architect of which, Hayrettin,

had the Byzantine monument Hagia Sophia as an example.

The mosque contains four great brick and cut-stone composed arches,

springing from four stone piers that offer primary support to a central dome

with 16.8m diameter and 36.5m height and to two semi-domes.

Beyazıt Mosque – Structural Properties

S6 Storng Arch (Brick+Stone) Semi Dome Original : Weak Arch (Brick)

Existing : Strong Arch (Brick + Stone)

Central Dome 45.40 m 43.50 m N Column Extension

•Stone elements have been attached

to eachother by led-covered iron ties.

Beyazıt Mosque – Structural Properties

(contd.)

Engineering is

details at the end

•The mosque was completed in 1506 and experienced the most destructive

earthquake in Istanbulʼs recorded history in 1509. Researchers note that the

main dome and semi-domes fell down partly.

•The structure has experienced 6 earthquakes with magnitude larger than 7

occurred in distances between 30 and 130 km, since it was built.

•Researchers indicate that the structure is constructed over a 48-m deep clay

soil deposit having high plasticity, leading thus high earthquake amplification

even during high amplitude of accelerations.

Comparison among Beyazit – H.Sophia - Süleymaniye

Hagia Sophia Beyazıt Süleymaniye

However, the weakness of the “weak arch” phenomenon was corrected

by Sinan in Süleymaniye !

The most pronounced similarity among these three structures is the structural

truss as four main piers, two semi domes settled on arches and two

perpendicular arches. The similarity of the direction of the framing system

drags the structures to the same destiny during an EQ.

0 1 2 3 4 5 Da m ag e S ta te (EMS sca le , 1 9 9 2 ) 1509 1729 1754 1766 1894 1999 Year of Earthquakes

Damage History of Beyazıt Mosque

It should be noted that the earthquake direction is not the sole determining

parameter of the damage; however, a distinction of the effect of the

direction is easily made

N 1766, Ms=7.2 30 km, Damage : 3 1509, Ms=7.6 47 km, Damage : 4 1894, Ms=7.0 105 km, Damage : 3 1719, Ms=7.6 175 km, Damage : 2 1754, Ms=7.0 147 km, Damage : 2 1999, Ms=7.8 115 km, Damage : 2 Increasing Damage Direction of the weak arch

Retrofitting by Sinan

•The retrofitting is perfectly covered and therefore it was not known till a

restoration in mid-70ʼs. The retrofitting is referred only in a historical source

which was authored by Sinanʼs best friend.

•The only operation conducted during the retrofitting is the strengthening of

the weak arches and associated columns, to authorʼs knowledge.

S6 Storng Arch (Brick+Stone) Semi Dome Original : Weak Arch (Brick)

Existing : Strong Arch (Brick + Stone)

Central Dome 45.40 m 43.50 m N Column Extension

Hagia Sophia was also retrofitted by Sinan, at the same time with Beyazit

Mosque. The concept is the same since the deficiency is identical in both

structures :

“Prevent the weak arches from opening!”

Weak arch

Strong arch

•There are three different arches below the original circular arch

•Key point of this retrofitting is the shape of the additional arch. Because,

the

structural concern which forced the designer to oppose the architectural

compatibility is the reason behind the retrofitting!

•One theory is that the reason behind adding a steep arch is just architectural

Retrofitting by Sinan

(contd.)

•Another theory is that Sinan was concerned about not decerasing the available

distance between two columns and found such a geometrical trick

• Our Theory :

His main concern was to prevent the drum of the dome from

•The answer is hidden among the pages of structural statics books.

•The most known arch forms and their purposes of use are given below;

Kendi ağırlığı veya sabit yayılı yük

Çevresel sabit yayılı yük

Çevresel sabit yayılı yüke eklenmiş uç kuvvet

Retrofitting by Sinan

(contd.)

Formed under its own

self weight Hydrostatic pressure pressure + a tip load at Hydrostatic the crown level

Inverted wire-loading

Retrofitting by Sinan

(contd.)

Crown Lateral Displacement (m)

PG

A

(g

)

Retrofitted (after 1574) Original (before 1571)

0.00 0.10 0.20 0.30 0.40 0.50 0 0.05 0.1 0.15 0.2

Column Top Displacement (m)

PG

A

(g

)

Retrofitted (after 1574) Original (before 1571)

•Simplified

pushover

analysis

results

proive

clearer insight to the fact.

•Force

resistance

and

ductility have been almost

doubled.

•The most impressive result of

the retrofitting is that the

differential settlement of the

dome under combined loading

was decreased

6 times.

Seismic Strengthening

•

With an ever enlarging building inventory and with the decarbonisation

process, the old methods are not enough anymore

•

Seismic strengthening now has to be combined with many other demands

of use of buildings (architectural, energy, circular economy etc.)

Trend in Seismic Strengthening

Principles of Seismic Protection

•

Longer peirod of data collection

•

Very well structural documentation of different construction times

•

Longer period of monitoring

•

Much talking vs Less Actual Work

Principles of Seismic Protection

•

60~70% of time à Phase 1: Data Collection

•

15~20% of time à Phase 2: Design & Project Phase

•

15~20% of time à Phase 3: Actual Construction Works

Collection of Data on the Geometry

•

Structural drawings should define the

- Bearing system dimensions

- Materials used

- Connections

- Crack map

- Foundations

Collection of Data on the Material

•

Which part of the structure was built at

which era?

•

What materials were used

•

How are the connections between the

old and the newer?

•

What is the damage history? What

interventions were made?

Collection of Data on the Material

•

Material compositions need to be found, for

mortar, plaster, bricks and stones

•

This is a necessary step for any repair or

strengthening work

•

Material strength is needed for structural

analysis

•

Compression strength, existing stress state

as well as shear strength are possible, but

only with destructive tests

Collection of Monitoring Data

•

Condition monitoring includes all other types of monitoring acitivites

apart from structural monitoring

•

Atmospheric data is in this group of data

•

Measurements on the ground water table are supporting measurements

•

Ground inclinometers are usful for understanding what is happening

with soil layers around the structure

Collection of Monitoring Data

•

Ambient vibration tests are

non-destructive

•

They can be conducted in half a day

•

The results are useful for calibrating

structural models

•

If recorded regularly, information on

detoriation of the structure can also be

collected

Collection of Monitoring Data

•

Accelerometers can be used to detect the period of vibration its change

with atmospheric conditions and aging

•

Displacement sensors can be used for detecting crack development

•

Tiltmeters can be used for detecting soil, retaining wall, or flexible part

movements especially after the earthquakes or in time (without

earthquakes)

Collection of Monitoring Data

Structural Health Monitoring

CAMI INK3A 22/01/2016 22/03/2016 29/04/2016 26/05/2016 25/07/2016 23/09/2016 24/11/2016 De pt h in M et er s 0 5 10 15 20 25 Profile Change in mm -20 -10 0 10 20 CAMI INK3B 22/01/2016 22/03/2016 29/04/2016 26/05/2016 25/07/2016 23/09/2016 24/11/2016 De pt h in M et er s 0 5 10 15 20 25 Profile Change in mm -20 -10 0 10 20

Strengthening

Mortar-binded

AR Glass (open-grid)

Basalt (open-grid)

19/19

Şekil 25. Duvarlarda örnek bazalt uygulaması

Yapıdaki tüm çimento esaslı ekler, sıvalar, duvarlar yapıdan uzaklaştırılacaktır. Yapıda taşıyıcı duvarlarda hidrolik esaslı kireç uygulaması yapılarak, boşluklu olan ve/veya çatlakları bulunan duvarlarda rijitlik ve dayanım artışı sağlanacaktır. Bilgilerinize saygılarımla iletirim. Yrd. Doç. Dr. İhsan Engin BAL İTÜ Deprem Mühendisliği ve Afet Yönetimi Enstitüsü 18/19

Şekil 24. Duvar çatlaklarında dikiş detayı

Bu raporda belirtilen duvarlarda bazalt veya AR glass ile sıva güçlendirmesi yapılacaktır. Bu duvarlarda uygulama şu şekilde yapılacaktır:

• öncelikle mevcut tuğla yüzey temizlenecektir

• bu yüzeye 1-3cm kalınlığında özgününe uygun harç sıva uygulanacaktır • bu sıva halen taze iken, bazalt veya AR glass malzeme bu sıvanın üstüne

veya içerisine yerleştirilecektir (bir örnek için bkz. Şekil 25)

• bazalt veya AR glass malzemenin üzeri de en az 1cm kalınlıkta harç ile kaplanacak ve ardından ince sıva yapılacaktır

Strengthening

Strengthening

Strengthening

Strengthening

Strengthening

14/23

Şekil 17 deki gibi ilave temel u gulamas ger ekle tirilir

Şekil 16 Planda d ar al rnek emel g lama

Şekil 17. Duvar al ip BA emel g lendirme ke i i

2 5 5 2 ap Hidrolik Kireçli Tesviye 10 10 10/20 14 Üst 5 a Ampatman 14 Gövde 2 40 40 40 40 ÖNEML NOT:

1. Ölçüler deği ken olduğu için tüm ölçüler yerinde alınacaktır.

2. Güçlendirme kazıları bina dı arısında birer atlamalı yapılacaktır. Dökülen beton pirizini almadan diğer temel açılmayacaktır.

3. Paslanmaz Çelik Temel altı demirleri temel alt kotundan yakla ık 5 cm üstten 40 cm ara ile tek sıra ve Ø60 mm çapında delinecektir.

4. Ø60 mm çapında delinecek olan delikler, Ø30 mm'lik paslanmaz çelik çubuk geçirildikten sonra dı ı tamir harcı ile doldurulacaktır. Bo luk kalmamasına özen gösterilecektir.

5. Kazı sonrası duvar yüzeyleri temizlenecek, derzleri açılacak, eksik kısımlar özgün karı ımlı harç kullanılarak, özgün malzemesiyle tamamlanacak, temel içerisinde ve dolgu içerisinde kalan tüm yüzeyler beton harcının ta yüzeye yapı maması için özgün sıva harcı ile kaba sıva yapılacaktır.

6.Beton temel yüzeylerine 400 Doz ap yapıldıktan sonra su yalıtımı uygulanacaktır. ap

Temel Altı Kırma Ta (10 cm)

Drenaj Borusu (Geotekstil Keçe Sarımlı)

5Ø14 3Ø14 2Ø14 Ø10/20 ±0.00 ±0.00 14 Alt 3 14 Üst 5 14 Gövde 2 14 Alt 3 L=Deği ken L=Deği ken L=Deği ken 5Ø14 3Ø14 2Ø14 Ø10/20 50 50 Ç KISIM DI KISIM 5Ø14 3Ø14 2Ø14 Ø10/20 Ø30/40 5Ø14 3Ø14 2Ø14 Ø10/20 50 50 50 50 Ana Ta ıyıcı Duvar Özgün Harçlı Kaba sıva

Temel Altı Kırma Ta (10 cm)

Çift kat Geotekstil Keçe Çift kat Geotekstil Keçe Kırma ta Özgün Harçlı Kaba sıva Doğal Toprak Dolgu

Doğal Toprak Grovak Dolgu

L:180cm

10/23 ba langıcına kadar m y kseklikte zemin içerisine de m kadar g m l olarak in a edilmi tir İç duvarlar ise zemin içerisine m kadar g m lm t r Mevcut yapı temellerinin tespiti için muayene çukurları açılmı tır Bu çukurlarda yapı duvarlarının ya ampatmansız olarak zemine indiği veya kısmen ya da tamamen ampatman içeren karı ık bir temel sistemine sahip olduğu tespit edilmi tir bkz Şekil 13).

Mermer s tunlar planda yakla ık x x m boyutlara sahip rme ta tekil temeller zerine oturtulmu tur Şekil 14).

Medrese dı duvarlarından açılıp g zlem yapılan G ney Cephe duvarında ise dı tarafta kısmi bir ampatmana rastlanmı olup bu ampatmanın derzlerinin bo aldığı ve ta larının eksik olduğu veya bazılarının niteliğini yitirdiği ve y k ta ıma g revini icra edemedikleri g r lm t r Şekil 15).

Şekil 13 A n oda i erisinde farkl d ar temelleri

Strengthening

Proper foundations

ation Foundat_widthion with _half

Ploa dBe arin g W all w ith ro per Foun datio n

14/23

Şekil 17 deki gibi ilave temel u gulamas ger ekle tirilir

Şekil 16 Planda d ar al rnek emel g lama

Şekil 17. Duvar al ip BA emel g lendirme ke i i

2

5

5

2

ap

Hidrolik Kireçli

Tesviye

10

10

10/20

14 Üst

5

a

Ampatman

14 Gövde

2

40

40

40

40

ÖNEML NOT:

1. Ölçüler deği ken olduğu için tüm ölçüler yerinde alınacaktır.

2. Güçlendirme kazıları bina dı arısında birer atlamalı yapılacaktır. Dökülen beton pirizini almadan diğer

temel açılmayacaktır.

3. Paslanmaz Çelik Temel altı demirleri temel alt kotundan yakla ık 5 cm üstten 40 cm ara ile tek sıra ve

Ø60 mm çapında delinecektir.

4. Ø60 mm çapında delinecek olan delikler, Ø30 mm'lik paslanmaz çelik çubuk geçirildikten sonra dı ı

tamir harcı ile doldurulacaktır. Bo luk kalmamasına özen gösterilecektir.

5. Kazı sonrası duvar yüzeyleri temizlenecek, derzleri açılacak, eksik kısımlar özgün karı ımlı harç

kullanılarak, özgün malzemesiyle tamamlanacak, temel içerisinde ve dolgu içerisinde kalan tüm yüzeyler

beton harcının ta yüzeye yapı maması için özgün sıva harcı ile kaba sıva yapılacaktır.

6.Beton temel yüzeylerine 400 Doz ap yapıldıktan sonra su yalıtımı uygulanacaktır.

ap

Temel Altı Kırma

Ta (10 cm)

Drenaj Borusu (Geotekstil Keçe Sarımlı)

5Ø14 3Ø14 2Ø14 Ø10/20

±0.00

±0.00

14 Alt

3

14 Üst

5

14 Gövde

2

14 Alt

3

L=Deği ken

L=Deği ken

L=Deği ken

50

50

Ç KISIM

DI KISIM

5Ø14

3Ø14

2Ø14

Ø10/20

50

50

50

50

Ana

Ta ıyıcı

Duvar

Özgün Harçlı

Kaba sıva

Temel Altı Kırma

Ta (10 cm)

Çift kat Geotekstil

Keçe

Çift kat

Geotekstil Keçe

Kırma ta

Özgün Harçlı

Kaba sıva

Doğal Toprak Dolgu

Doğal Toprak

Grovak Dolgu

L:180cm

Strengthening

Proper foundations

Natural Soil

Load Bearing

Wall

16/23

Şekil 19 S n al la nda BA emel g lendi me g lama BA emel ke i le i ahada ine Şekil deki gibi ka e a labili

Betonarme temel takviyeleri g n yap birimlerine yap mayacak ekilde araya

koruyucu

g ne uygun ve geri kart labilir tabakalar

g ne uygun s va

tabakas uyguland ktan sonra in a edilecektir S va tabakas horasan olup serpme

ve p r l bir ekilde uygulanacakt r

Temel betonunda yerle menin iyi olmas gerekti inden betona ak kanla t r c

madde kat lmas ve kendinden yerle en beton kullan lmas gerekir.

Temel betonunun mr n u atmak i in beton poro itesini a altan katk lar veya

buna uygun imento tipi kullan lmas nerilir

7. Yapısal İyileştirme Önerileri

Yap da temel iyile tirilip duvar ve s tun yatmalar kontrol alt na al nd ktan sonra

farkl taraflara yatan duvarlar birbirine ba lamak ve yap da b t nl

sa lamak

i in Şekil 20 ve Şekil 21 de g sterilen B model

erinde yap lan anali ler

sonucunda Şekil 22 de verilen gergiler nerilmi tir Bu gergiler AISI316

kalitesinde, 30 mm apl ve iki taraf yivli olarak tertip edilecektir. Bu gergiler

h cre i duvarlar na iki y den paralel gidip i ve d ta UPN160 profil veya

e de eri bir k l olu turacakt r bk Şekil

ve

. Gergiler i in en a mm

apl delik delinecek deli in i i gergi yerle tirildikten sonra hidrolik kire

enjeksiyonu ile doldurulacakt r

50 50 50 Ta Tekil Temel (Mevcut) Ta y c S tun Temel Alt K rma Ta 10 cm Temel Alt Do al Toprak T em el K az Al an c d e 30 paslanmaz çelik çubuk f g 14 5 14 2 14 3 10/20 10/20 50

BA Betonu Alt na Uygun Harçl Kaba S va

11/23

Şekil 14 S emelle i

Dış duvarlarda dışa doğru dönme ve kubbelerden açılmalar gözlemlenmektedir Kolonların bazılarında, özellikle kesişim yerlerindeki kolonlarda düşey aksından eğilmeler gözlemlenmiştir Bu dönmelerin sebepleri deprem ya da farklı yükler kaynaklı yatay yüklerin etkimesi ve dış duvarların bu yüklerden kaynaklanacak dönme momentlerini karşılayacak bir temel sisteminin bulunmaması olarak düşünülebilir

Strengthening

16/23

Şekil 19 S

n al la nda BA emel g lendi me g lama BA emel ke i le i ahada ine Şekil

deki gibi

ka e a labili

Betonarme temel takviyeleri g n yap birimlerine yap mayacak ekilde araya

koruyucu

g ne uygun ve geri kart labilir tabakalar

g ne uygun s va

tabakas uyguland ktan sonra in a edilecektir S va tabakas horasan olup serpme

ve p r l bir ekilde uygulanacakt r

Temel betonunda yerle menin iyi olmas gerekti inden betona ak kanla t r c

madde kat lmas ve kendinden yerle en beton kullan lmas gerekir.

Temel betonunun mr n u atmak i in beton poro itesini a altan katk lar veya

buna uygun imento tipi kullan lmas nerilir

7. Yapısal İyileştirme Önerileri

Yap da temel iyile tirilip duvar ve s tun yatmalar kontrol alt na al nd ktan sonra

farkl taraflara yatan duvarlar birbirine ba lamak ve yap da b t nl

sa lamak

i in Şekil 20 ve Şekil 21 de g sterilen B model

erinde yap lan anali ler

sonucunda Şekil 22 de verilen gergiler nerilmi tir Bu gergiler AISI316

kalitesinde, 30 mm apl ve iki taraf yivli olarak tertip edilecektir. Bu gergiler

h cre i duvarlar na iki y den paralel gidip i ve d ta UPN160 profil veya

e de eri bir k l olu turacakt r bk Şekil

ve

. Gergiler i in en a mm

apl delik delinecek deli in i i gergi yerle tirildikten sonra hidrolik kire

enjeksiyonu ile doldurulacakt r

50

50

50

Ta Tekil

Temel

(Mevcut)

Ta y c

S tun

Temel Alt

K rma Ta 10 cm

Temel Alt

Do al Toprak

T

em

el

K

az

Al

an

c

d

e

30 paslanmaz

çelik çubuk

f

g

50

BA Betonu Alt na Uygun

Harçl Kaba S va

Natural Soil Under

Foundation

Marble

Column

Fi30 stainless

steel bar

Protective

authentic-like cement-free mortar

below concrete

Existing

stone

masonry

foundation

Before that - A close look in the current sensor technology

Accelerometers

Pieso-electric

Force-balanced

MEMS

Q-MEMS (digital output)

Translational Displacement Sensors

LVDTs

Potentiometers

Other Sensors

Tilt-meters

Velocity-meters (vibrometers)

What we measure in structures ?

•

_{The structural monitoring may have several purposes, such as:}

- Dynamic characterization (OMA)

- Long-term structural health monitoring (SHM)

Dynamic Characterization (OMA)

•

_{All structures vibrate with amplitudes outside of the range of human}

senses

•

These vibration are in extremely small amplitudes, meaning that they

are in elastic range (no damage zone) and can get confused with

ambient noise

•

Structural vibrations are low-frequency, typically in the range of 0.1 to

3-4 seconds fundemental periods, thus filtering the data helps in

processing

Dynamic Characterization - Challenges

•

_{Good sensor producers shift track to mobile technologies}

•

Wireless technologies, surprisingly, are still not problem-free

•

There is need for plug&play technologies

•

_{There is need for smart data processing}

Sensor

Digitizer

Computer

Computer

Max 50-60m

Max. 200m with

ethernet cables

Few meters

Sensor

Digitizer

Long-term Structural Health Monitoring

•

_{Structures are organisms that move, but very very slowly}

•

If we monitor them with sensitive enough sensors for a long enough

time, we can detect this movement

•

_{Structural Health Monitoring is a long-term investment with very}

sensible and useful results, especially for old structures where we

should not be in a hurry anyhow

Example of Long-Term Monitoring

Future of seismic and vibration data collection

5

Figure 2 Active fault map of MTA Turkey [29] and the location of the Eurasia Tunnel Operational Vibrations

The speed limit in the tunnel is 70kph (43mph) and only cars and small vans are allowed in the tunnel Most vehicles travel close to that speed creating peaks at the sensor measurements as they travel.

The upper deck of the tunnel (Figure 3a) is a slender member of the structure generating amplified vibrations from the traffic. The background noise during operation, without a car passage, is within the band of +-1µg. The daily traffic causes acceleration peaks approximately up to +-0.0007g. An example time series from the daily traffic can be seen in Figure 4.

There are two fire trucks (Figure 3b) that are approximately 11ton in weight, heavier than the usual vehicles using the tunnel. These trucks have a patrol duty at night, travelling from one side of the tunnel to the other. They also cause vibrations, which due to their weight are much higher than those of the daily traffic due to their weight. The data show that the fire truck passage can cause vibration levels up to 0.0015g on the installed sensors (Figure 4).

Earthquake

Decision

Support

Example

Application:

Eurasia Tunnel

Future of seismic and vibration data collection

6

Figure 3 Section of the tunnel with upper and the lower decks (a), view from inside the tunnel, above the

upper deck (b) and one of the twin the fire trucks that patrol every night (c)

Figure 4 Time histories of the records from each label with the largest PHA (note that most of the records

in different labels have PHA values close to each other, only the ones with the largest PHA are presented

here for demonstration purposes)

Earthquake

Decision

Support

Example

Application:

Eurasia Tunnel

Future of seismic and vibration data collection

6

Figure 3 Section of the tunnel with upper and the lower decks (a), view from inside the tunnel, above the upper deck (b) and one of the twin the fire trucks that patrol every night (c)

Figure 4 Time histories of the records from each label with the largest PHA (note that most of the records in different labels have PHA values close to each other, only the ones with the largest PHA are presented

here for demonstration purposes)

11

Figure 6 Probability density function of each of the numeric features used in training Training and Validation

Supervised ML methods of SVM, kNN and Ensemble Decision Trees are used for analyses. There are several sub-methods and options within these sub-methods, some of which will be discussed and compared here. In general, more complex options have the tendency to require a higher computational time and performance although for the size of the problem presented here, the overall training and validation time did not exceed some minutes per model. Note that the entire ML framework, from data cleaning and structuring to implementing the best model, is given in Figure 7. It can be seen that there is a considerable amount of work for preparing the final array of values that consists of the extracted features.

SVM is among the first examined methods with the dataset. There are various options of applying the SVMs, but the best result was produced by the cubic SVM. One interesting observation is that, when all 14 features are used, the SVM had a slightly smaller overall success of predictions as compared to the other methods tested (89.3%), but the prediction of the earthquake vibrations alone was even higher than that of the other methods. If the problem is reduced to a binary problem as “earthquake” vs “non-earthquake”, then the SVM method provides the highest accuracy in prediction by 97.0% (see

). In order to compare the performance of the SVM methods with the most successful method, EBS - Ensemble Bagged Trees, the same 9 features are used for another trial. The overall success of SVM decreased to 87.8% while the success in predicting earthquakes was at 95.0%. It should be noted that, although it made only a difference of some minutes in the example predicted here, in relative terms, Cubic SVM is a significantly more computationally demanding method than the EBS, a crucial difference if the dataset and the number of features used are substantially larger.

Future of seismic and vibration data collection

If

properly

trained,

machines can differentiate

earthquakes from other

vibrations

14

Weighted kNN is used as an alternative method of classification. It exhibited the lowest success ratio among the five most successful methods. In implementing the weighted kNN, squared inverse is used as distance weight, while the distance metric was Euclidian. It should be noted the accuracy of the kNN also depends on the user selection of the number of the neighbors, k, that is examined at a time by the algorithm to see in which class the data point falls into. A sensitivity check on this parameter revealed that the selection of k had a minimal effect on the accuracy because k values of 3, 10, 30 and 50 resulted in 84.3%, 84.0% 74.3%, 84.4% and 84.0% overall accuracy, respectively. The accuracy in earthquake prediction alone was 89%, 88%, 53%, 81% and 79%, respectively. It is thus clear that if kNN is to be used, a sensitivity study would be needed to find the optimum k value as the accuracy is not linearly correlated with the k value.

The Ensemble method is another computationally efficient approach that implements decision trees. Both bagging and boosting methods, explained above in detail, are tested here with varying sub-options. The accuracy range of the Ensemble method with different sub-options was between 90% to 95.2% for overall accuracy, and between 94% and 98% for earthquake prediction alone. Ensemble methods are proved to be the most suitable and at the same time the most efficient methods for the problem presented here.

Accuracy of a ML model on different labels is presented with a confusion matrix. The confusion matrix exhibits the class predictions per label versus the true class of each label. A heavily diagonal confusion matrix indicates success in predictions, while off-diagonal numbers represent the prediction errors. A perfect model with 100% accuracy would thus present a confusion matrix with 100% on the diagonal and 0% on the off-diagonal elements. The results of this study are presented also in a confusion matrix, as shown in Figure 8. When calculating the accuracy, cross-validation is used for validating the presented model. Cross-validation is an efficient tool to prevent over-fitting, a modelling defect that renders the model too much dependent on the seen data and not successful for the unseen data. It can be seen in Figure 8 that the prediction of earthquakes alone is 98%, while the prediction of fire truck passage and daily traffic are as accurate as 96% and 94%, respectively. The model is deemed to be reasonably successful for automatically detecting earthquake in the tunnel, by using 9 features only (see

for the selected 9 features).

Figure 8 Confusion matrix of the most accurate model

The features are also evaluated one by one by using the best method, to define what effect they have on the overall and earthquake-alone accuracy. The results are presented in

. When earthquake-alone predictions are considered, the most effective features, in the order of effectiveness, are Difference in Arrival Time (tAD) and the 5% damped Spectral Accelerations at 0.2sec period (Sa02). These two

parameters are able to best differentiate the earthquake vibrations from the rest. In the overall accuracy, the most influential features, in order of effectiveness, are Difference in Arrival Time (tAD), 5% damped Spectral

Accelerations at 0.2sec period (Sa02), Total Arias Intensity (TAI) and the Effective Duration between 5% and 75% of the Arias Intensity plot (t5-75). The last two parameters clearly indicate a pattern difference between the

May be in the future they

also can

- locate and characterize

earthquakes

- estimate damages

- estimate human needs

- and finally, take desicions

Ihsan Engin Bal

Professor in Earthquake Resistant Structures

i.e.bal@pl.hanze.nl

@

EQ

Research

Hanze

/

EQ

Research

Hanze

www.

EQ

Research

.nl