Professional Documents
Culture Documents
Ebook Resilience of Civil Infrastructure Systems Developments in Testing Sensing Monitoring and Control First Edition Edition Structural Engineer Zhishen Wu Editor Online PDF All Chapter
Ebook Resilience of Civil Infrastructure Systems Developments in Testing Sensing Monitoring and Control First Edition Edition Structural Engineer Zhishen Wu Editor Online PDF All Chapter
https://ebookmeta.com/product/29-single-and-nigerian-incomplete-
first-edition-naijasinglegirl/
https://ebookmeta.com/product/internet-of-things-and-access-
control-sensing-monitoring-and-controlling-access-in-iot-enabled-
healthcare-systems-1st-edition-shantanu-pal/
https://ebookmeta.com/product/security-of-cyber-physical-systems-
state-estimation-and-control-1st-edition-chengwei-wu/
https://ebookmeta.com/product/security-and-resilience-in-cyber-
physical-systems-detection-estimation-and-control-masoud-
abbaszadeh/
https://ebookmeta.com/product/analysis-of-legal-argumentation-
documents-a-computational-argumentation-approach-translational-
systems-sciences-29-hayato-hirata/
https://ebookmeta.com/product/fortschritte-der-physik-progress-
of-physics-band-29-heft-8/
https://ebookmeta.com/product/fortschritte-der-physik-progress-
of-physics-band-29-heft-4/
Experimental Vibration Analysis
for Civil Structures
Testing, Sensing, Monitoring, and Control
Taylor and Francis Series in Resilience and
Sustainability in Civil, Mechanical, Aerospace and
Manufacturing Engineering Systems
Series Editor:
Mohammad Noori
Cal Poly San Luis Obispo
PUBLISHED TITLES
Resilience of Critical Infrastructure Systems
Emerging Developments and Future Challenges
Zhishen Wu, Xilin Lu, and Mohammad Noori
Experimental Vibration Analysis for Civil Structures
Testing, Sensing, Monitoring, and Control
Jian Zhang, Zhishen Wu, Mohammad Noori, and Yong Li
Edited by
Jian Zhang, Zhishen Wu, Mohammad Noori, and Yong Li
MATLAB ® is a trademark of The MathWorks, Inc. and is used with permission. The MathWorks does not warrant the
accuracy of the text or exercises in this book. This book’s use or discussion of MATLAB ® software or related products
does not constitute endorsement or sponsorship by The MathWorks of a particular pedagogical approach or particular
use of the MATLAB ® software
Reasonable efforts have been made to publish reliable data and information, but the author and publisher cannot
assume responsibility for the validity of all materials or the consequences of their use. The authors and publishers
have attempted to trace the copyright holders of all material reproduced in this publication and apologize to
copyright holders if permission to publish in this form has not been obtained. If any copyright material has not been
acknowledged please write and let us know so we may rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or
utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including
photocopying, microfilming, and recording, or in any information storage or retrieval system, without written
permission from the publishers.
For permission to photocopy or use material electronically from this work, access www.copyright.com or contact the
Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. For works that
are not available on CCC please contact mpkbookspermissions@tandf.co.uk
Trademark notice: Product or corporate names may be trademarks or registered trademarks, and are used only for
identification and explanation without intent to infringe.
Typeset in Times
by codeMantra
Contents
Preface............................................................................................................................................... xi
Editors ............................................................................................................................................ xiii
Contributors ..................................................................................................................................... xv
Chapter 4 Combined Actuator-Shake Table Test with Optimized Input Energy ....................... 33
Farhad Behnamfar and Mohammadreza Najar
Chapter 5 Design Spectra for Structures Subjected to Passing Underground Trains ................. 43
F. Behnamfar, A. A. Yazdani, and A. Bakhtiar
Chapter 8 Damage Recognition of Wood Beam Based on Curvature Modal Technology ........ 71
Chunyu Qian, Baozhuang Zhang, Junlong Zhu, and Qifang Xie
v
vi Contents
Chapter 11 The Analysis of the Temperature Effect on Frequencies of a Footbridge ............... 107
De-hui Tang, Jun Teng, and Wei-hua Hu
Chapter 15 Damage Features from Direct Modal Strain Measurements ................................... 145
G. De Roeck, E. Reynders, and D. Anastasopoulos
Chapter 17 Concrete Crack Image Recognition Based on DBSCAN and KPCA ...................... 169
Mingxin Gao and Yang Liu
Chapter 28 Highway Bridge Weigh-in-Motion via Moving Load Identification ........................ 263
Xiangang Lai, A. Emin Aktan, Kirk Grimmelsman, Matteo Mazzotti, and
Ivan Bartoli
Chapter 35 Pseudo Damage Training for Seismic Fracture Detection Machine ........................ 355
Luyao Wang, Ji Dang, and Xin Wang
Chapter 44 Non-Probabilistic Damage Detection Using Classic and Modal Interval Analysis ... 445
J.Y. Huang and S.E. Fang
Chapter 45 A 3D Isolation Device with Vertical Variable Stiffness for Long-Span Spatial
Structures ................................................................................................................. 451
Y. Shi, H. Liu, Z. Chen, and Y. Ding
Chapter 46 Detection of Crack in Euler-Bernoulli Beam Using Bayesian Inference ................. 461
T. Wang, Mohammad Noori, Wael A. Altabey, Z. Ying, and R. Ghiasi
Chapter 49 Wind Speed and Wind Direction Joint Distribution Model Based on
Structural Health Monitoring ................................................................................... 487
X. W. Ye and Y. Ding
Chapter 54 Hybrid DIC-Meshless Method for Evaluating Strain Field around the Crack ......... 541
Zhu Zhihui, Luo Sihui, Wang Fan, Feng Qianshuo, and Jiang Lizhong
Chapter 56 Estimation of Repeated Slip Surface in Cut Slope Stability Analysis ..................... 561
Zulkifl Ahmed, Shuhong Wang, and Dong Furui
Index .............................................................................................................................................. 571
Preface
This book is a collection of selected papers presented at the International Conference on
Experimental Vibration Analysis for Civil Engineering Structures (EVACES), held in Nanjing,
China in September 2019.
EVACES is a premier venue where recent progress in the field of civil infrastructure systems,
especially the latest developments in experimental vibration analysis research in civil engineer-
ing structures, is presented and discussed by experts from the global community. Over the past
decade, due to emergence of new technologies and computational tools, the field of civil engineer-
ing and infrastructure systems has witnessed a significant advancement in the aforementioned
areas. EVACES provides a forum for the global research community to share and exchange the
latest achievements in these areas. After the first seven successful EVACES conferences which
took place in Bordeaux, France (2005), Porto, Portugal (2007), Wroclaw, Poland (2009), Varenna,
Italy (2011), Ouro Preto, Brazil (2013), Dubendorf, Switzerland (2015), and San Diego, United
States of America (2017), EVACES 2019 was organized by the Southeast University (SEU)
at Nanjing, held from September 5–8, 2019, on the main campus of SEU. The main topics of
EVACES 2019 included (1) damage identification and structural health monitoring, (2) testing,
sensing, and modeling, (3) vibration isolation and control, (4) system and model identification,
(5) coupled dynamical systems (including human-structure, vehicle structure, and soil-structure
interaction), and (6) application of big data and artificial intelligence techniques. The key objec-
tive of this meeting was to present and discuss the latest developments in the aforementioned
areas in order to better understand how to assess the state of health of structural, geo-structural,
and complex infrastructure systems and how to predict their remaining useful life using vibration
data collected from these systems when subjected to operational and extreme loads. It is notewor-
thy to indicate this was the first time that the importance of advanced data analysis (e.g., system
and damage identification and artificial intelligence) methods in experimental vibration analysis
of critical infrastructure, as an important and emerging area, was addressed to support decision
making related to maintenance and inspection, retrofit, upgrade, and rehabilitation of these sys-
tems especially in case of emergency response.
This book contains selected papers prepared by the leading researchers that were presented
at EVACES 2019. These contributions can provide strategic roadmaps for research in the area of
experimental vibration analysis for civil engineering structures and for building more resilient and
sustainable civil infrastructure systems.
We express our sincere thanks to the members of the Scientific Committee and, in particular, to
all the authors and participants, for their essential and valuable contributions.
Workshop Co-Chairs and the Organizing Committee:
xi
xii Preface
Professor Zhishen Wu is a distinguished professor at Ibaraki University, Japan and at SEU, China.
His research expertise includes structural/concrete/maintenance engineering and advanced com-
posite materials. He is the author or co-author of over 600 refereed papers including over 200
journal articles and has given 50 keynote or invited papers. He also holds 50 patents. Dr Wu
was awarded the JSCE Research Prize by Japan Society of Civil Engineering in 1990, the JSCM
Technology Award by the Japan Society for Composite Materials in 2005 and 2009, SHM Person
of the Year Award by International Journal of Structural Health Monitoring, and the National
Prize for Progress in Science and Technology (2nd) of China in 2012. He is also a member of Japan
National Academy of Engineering. He chairs China Chemical Fibers Association Committee on
Basalt Fibers and is the president of the ISHMII. He is also an elected follow of American Society of
Civil Engineering (ASCE), Japan Society of Civil Engineers (JSCS), ISHMII, and the International
Institute of Fiber Reinforced Polymers (FRP) in Construction. He also serves as an editor, associ-
ate editor, and editorial board member for more than ten international journals. He is the founding
chief editor of International Journal of Sustainable Materials and Structural Systems. Professor
Wu has supervised a large number of national research projects in China and was the founder of the
International of Institute for Urban Systems Engineering, at SEU.
Professor Mohammad Noori is a professor of mechanical engineering at Cal Poly, San Luis
Obispo, a Fellow of the American Society of Mechanical Engineering, and a recipient of the Japan
Society for Promotion of Science Fellowship. Dr Noori’s work in nonlinear random vibrations, seis-
mic isolation, and the application of artificial intelligence methods for structural health monitoring
is widely cited. He has authored over 250 refereed papers, including over 100 journal articles and
6 scientific books, and has edited 25 technical and special journal volumes. Noori has supervised
over 90 graduate students and post-doc scholars, and has presented over 100 keynote, plenary, and
invited talks. He is the founding executive editor of an international journal and has served on the
editorial boards of over 10 other journals and as a member of numerous scientific and advisory
boards. He has been a Distinguished Visiting Professor at several highly ranked global universities,
and directed the Sensors Program at the National Science Foundation in 2014. He has been a found-
ing director or co-founder of three industry-university research centers and held Chair professor-
ships at two major universities. He served as the dean of engineering at Cal Poly for five years, has
also served as the Chair of the national committee of mechanical engineering department heads,
and was one of 7 co-founders of the National Institute of Aerospace, in partnership with NASA
Langley Research Center. Noori also serves as the Chief Technical Advisor for several scientific
organizations and industries.
xiii
xiv Editors
Dr. Yong Li is an assistant professor in the School of Civil Engineering, SEU, China. He received
his undergraduate degree in 2012 and his PhD in 2017, both degrees in aerospace engineering and
applied mechanics, at Tongji University, in Shanghai, China. His research focused on the coupling
relationship between mechanics and chemical reaction. He has published 1 book and over 20
journal papers in leading journals. He is also a reviewer for several journals in the fields of mechan-
ics and physics. Dr Li is the recipient of an Outstanding Young Investigator Award supported by
the National Natural Science Foundation of China. He is also engaged in several other national
key research and development programs as an investigator. He served as secretary general of the
International Conference on EVACES-8.
Contributors
J. Abell R. Astroza
Faculty of Engineering and Applied Sciences Faculty of Engineering and Applied Sciences
Universidad de los Andes Universidad de los Andes
Santiago, Chile Santiago, Chile
D. Ai Bai Wen
School of Civil Engineering and Mechanics Institute of Engineering Mechanics
Huazhong University of Science and China Earthquake Administration
Technology Beijing, China
Wuhan, China and
Key Laboratory of Earthquake Engineering
and Engineering Vibration of China
J. S. Ali Earthquake Administration
Department of Civil Engineering Harbin, China
Aliah University
Kolkata, India A. Bakhtiar
Department of Engineering
Mohammad Almutairi Sepehr Institute of Higher Education
School of Civil Engineering Isfahan, Iran
University of Leeds
Leeds, United Kingdom D. Bandyopadhyay
and Department of Construction Engineering
Sustainability and Infrastructure Programme, Jadavpur University
Energy and Building Research Center Kolkata, India
Kuwait Institute for Scientific Research
Safat, Kuwait Baozhuang Zhang
State Key Laboratory of Green Building in
Wael A. Altabey Western China
International Institute for Urban Systems Xi’an University of Architecture &
Engineering Technology
Southeast University Xi’an, China
Nanjing, Jiangsu, China.
and N. Barrientos
Faculty of Engineering, Department of Faculty of Engineering and Applied Sciences
Mechanical Engineering Universidad de los Andes
Alexandria University Santiago, Chile
Alexandria, Egypt
Ivan Bartoli
D. Anastasopoulos Department of Civil, Architectural and
Department of Civil Engineering Environmental Engineering
University of Leuven (KU Leuven) Drexel University
Leuven, Belgium Philadelphia, Pennsylvania
xv
xvi Contributors
Dong Furui H. Ge
School of Resource and Civil Engineering Department of Civil Engineering
Northeastern University Meijo University
Shenyang, China Nagoya, Japan
Du Yongfeng M. R. Ghasemi
Institute of Earthquake Protection and Faculty of Engineering, Department of
Disaster Mitigation Civil Engineering
Lanzhou University of Technology University of Sistan and Baluchestan
Lanzhou, China Zahedan, Iran
W. Y He Jun Li
Department of Civil Engineering Centre for Infrastructure Monitoring and
Hefei University of Technology Protection, School of Civil and Mechanical
Hefei, Anhui Province, China Engineering
Curtin University
X. H. He Bentley, Australia
School of Civil Engineering
Central South University Jun Teng
Changsha, China Harbin Institute of Technology (Shenzhen)
and Shenzhen, China
National Engineering Laboratory for High
Speed Railway Construction Junhao Zheng
Changsha, China College of Civil Engineering
and Fuzhou University
Joint International Research Laboratory of Fuzhou, Fujian, China
Key Technology for Rail Traffic Safety
Changsha, China Junlong Zhu
Shanghai Baitong Project Management
Zhiqiang Hou Consulting, Co., Ltd
The Fifth Construction Co., Ltd. of China Shanghai, China
TIESIJU Civil Engineering Group
Jiujiang, China Kang Liu
College of Civil Engineering
J. Y. Huang Fuzhou University
School of Civil Engineering Fuzhou, Fujian, China
Fuzhou University
Fuzhou, Fujian Province, China Muhammad Israr Khan
School of Resources and Civil Engineering
Northeastern University
Shenyang city, Liaoning Province, China
Contributors xix
D. Li H. Luo
School of Civil Engineering School of Civil Engineering and Mechanics
Hefei University of Technology Huazhong University of Science and
Hefei, Anhui, China Technology
Wuhan, China
H. Li
School of Civil Engineering and Luo Sihui
Mechanics School of Civil Engineering
Huazhong University of Science and Central South University
Technology Changsha, China
Wuhan, China
H. Mao
Li Junxing School of Ecological Environment and
Department of Civil Engineering Urban Construction
Harbin Institute of Technology Fujian University of Technology
Harbin, China Fuzhou, China
Li Xu G. C. Marano
College of Civil Engineering Sustainable and Innovative Bridge Engineering
Fuzhou University Research Center (SIBERC)
Fuzhou, Fujian, China College of Civil Engineering, Fuzhou
University
Liao Yuchen Fuzhou, P. R. China
School of Civil Engineering
Southeast University Matteo Mazzotti
Nanjing, China Department of Civil, Architectural and
Environmental Engineering
Lin Dinan Drexel University
Fujian Academy of Bulding Research Philadelphia, Pennsylvania
Fujian Key Laboratory of Green Building
Technology Mingxin Gao
Fuzhou, China Department of Bridge and Tunnel Engineering
Harbin Institute of Technology
Linfeng Hu Harbin, China
School of Architecture and Civil
Engineering T. Nagayama
Xiamen University Department of Civil Engineering
Xiamen, China The University of Tokyo
Tokyo, Japan
H. Liu
School of Civil Engineering Mohammadreza Najar
Tianjin University Department of Civil Engineering
Tianjin, China University of Isfahan
Isfahan, Iran
Liyuan Wang
College of Civil Engineering F. Necati Catbas
Fuzhou University Department of Civil, Environmental, and
Fuzhou, China Construction Engineering
University of Central Florida
Orlando, Florida
xx Contributors
Z. H. Nie G. Z. Qu
College of Science & Engineering China Railway Construction Bridge Design and
Jinan University Research Branch
Guangzhou, Guangdong, China China Railway Siyuan Survey and Design
and Group Co., LTD.
The Key Laboratory of Disaster Forecast and Wuhan, Hubei, China
Control in Engineering
Guangdong, Guangzhou, China Qunjun Huang
College of Civil Engineering
Nikolaos Nikitas Fuzhou University
School of Civil Engineering Fuzhou, Fujian, China
University of Leeds
Leeds, United Kingdom W. X. Ren
Department of Civil Engineering
Mohammad Noori Hefei University of Technology
International Institute for Urban Systems Hefei, Anhui Province, China
Engineering (IIUSE)
Southeast University E. Reynders
Nanjing, Jiangsu, China Department of Civil Engineering
and University of Leuven (KU Leuven)
Department of Mechanical Engineering Leuven, Belgium
California Polytechnic State University and
San Luis Obispo, California Structural Mechanics Division
University of Leuven (KU Leuven)
M. Orchard Leuven, Belgium
Department of Electrical Engineering
University of Chile Z. F. Shen
Santiago, Chile College of Science & Engineering
Jinan University
Pengyu Wang Guangzhou, Guangdong, China
School of Resources and Civil Engineering
Northeastern University S. Shi
Shenyang City, Liaoning Province, China College of Civil Engineering
Fuzhou University
M. C. Qi Fuzhou, China
Key Laboratory of Wind and Bridge
Engineering of Hunan Province Y. Shi
College of Civil Engineering, Hunan Key Laboratory of Coast Civil Structures
University Safety of Ministry of Education
Changsha, China Tianjin University
Tianjin, China
Qifang Xie
State Key Laboratory of Green Building in Shiqi Fu
Western China College of Civil Engineering
Xi’an University of Architecture & Huaqiao University
Technology Xiamen, P. R. China
Xi’an, China
Shuhong Wang
School of Resource and Civil Engineering
Northeastern University
Shenyang, China
Contributors xxi
A. A. Yazdani C. Zhang
Department of civil engineering College of Civil Engineering
University of Isfahan Fuzhou University
Isfahan, Iran Fuzhou, China
K. Ye Hongyu Zhang
Department of Civil Engineering Key Laboratory of Disaster Prevention and
Hefei University of Technology Mitigation in Civil Engineering of Gansu
Hefei, China Province
Lanzhou University of Technology
X. W. Ye Lanzhou, China
Department of Civil Engineering
Zhejiang University J. Zhang
Hangzhou, China Department of Civil Engineering
Hefei University of Technology
P. C. Yin Hefei, China
China Railway Construction Bridge Design and
Research Branch Zhang Kun
China Railway Siyuan Survey and School of Civil Engineering
Design Group Co., LTD. Southeast University
Wuhan, Hubei, China Nanjing, China
Yu Cheng L. Zhou
Department of Bridge Engineering, College of Civil Engineering
School of Transportation Fuzhou University
Southeast University Fuzhou, China
Nanjing, China
H. P. Zhu
X. S. Yu School of Civil Engineering and Mechanics
China Railway Construction Bridge Design and Huazhong University of Science and
Research Branch Technology
China Railway Siyuan Survey and Design Wuhan, P. R. China
Group Co., LTD.
Wuhan, Hubei, China
xxiv Contributors
Mohammad Noori
California Polytechnic State University
W. A. Altabey
Southeast University
Alexandria University
CONTENTS
1.1 Introduction .............................................................................................................................. 1
1.2 Data Preprocessing ................................................................................................................... 2
1.2.1 Feature Detection and Extraction ................................................................................. 2
1.2.2 Statistical Model Formulation ...................................................................................... 4
1.2.3 Scale Selection .............................................................................................................. 4
1.2.4 Feature Damage Index Identification Based Wavelet ................................................... 5
1.3 Numerical Simulation and Results: Discussion ........................................................................ 6
1.4 Conclusions ............................................................................................................................... 6
References ........................................................................................................................................ 10
1.1 INTRODUCTION
Signal-based damage detection techniques have been introduced to extract features to represent
the signal characteristics which are used to answer the questions associated to damage levels [1,2].
Over the last decades, various methods have been developed to improve the feature extraction (FE)
procedure [3–10]. Yoon et al. [11] and Staszewski [12] discussed the signal processing role for FE in
composite materials and stated that the advances in damage detection attribute to the development
of signal processing techniques such as denoising, FE, and optimal sensor location procedures.
FE-based signal processing has been classified into time, frequency, and time-frequency methods
[13]. It aims to transform data into a lower dimension by removing redundant information and pre-
serving the useful elements [14]. In time domain, features are extracted from the signal using linear
and nonlinear functions [15]. In frequency and time-frequency domain wavelet transform (wt) have
been used for FE. Pittner and Kamarthi [16] explored the changes in wavelet coefficients for FE for
pattern recognition tasks. Balafas and Kiremidjian [17] proposed damage features for earthquake
damage estimation based on continuous wavelet transform (CWT), using ground motion and its
structure responses. Patel et al. [18] used complex Gaussian wavelet to extract the discontinuity of
the acceleration response of a reinforced concrete (RC) building. The results showed that wavelet
weights are directly affected by the change in physical properties and can identify damage to a rea-
sonable extent. The modified wavelet energy rate-based damage detection methods are proposed by
1
2 Experimental Vibration Analysis for Civil Structures
several researchers [19–22]. The method based on Approximate Entropy [23] is proposed to assess
the damaged condition from the vibration signals generated by excitation [24–26] on structures.
It is able to capture the deterioration of beams with more sensitive characteristics. Although many
works have been done using wt for damage detection [27,28], however, it cannot always identify the
damage extent or its location. Most of them are not suited when the signal is nonstationary and is
mainly restricted to small structures, simulated models, or laboratory tests. Moreover, some of them
refer to specific load scenarios that are not common in real structures; their applicability is limited
and the ground motion recording is neglected. Further, only one feature was considered to resemble
the dynamic characteristic, and the base wavelet and resolution level are chosen in an arbitrary man-
ner, which results in a negative effect in results. It’s clear that there is a lack of a truly intelligent
and comprehensive approach that can detect the slight damage in a large, complex system and in a
noisy environment that may include a large number of relevant features, and not just a few dynamic
features, to detect the location and the damage extent. Thus, an advanced intelligent algorithm for
sensitive features extraction of large complex data is required. Other crucial issue in damage detec-
tion is how to select the correct features that describe the damage and improve the AI-algorithm
accuracy. In this study, given the modulation characteristics of structural dynamic responses (SDR),
the wavelet is used for damage-sensitive parameter (DSF) extraction which cannot be obtained
either by short Fourier transform or Wigner-Ville distribution. Here, FE is used to describe how to
find the best features which represent various damage conditions, whereas feature selection is used
to define how to select features for damage detection based wavelet DSFs. The proposed algorithm
is able to de-noise and extract the features that define the structure’s true state. The algorithm’s nov-
elty is to identify and extract several key features that can determine the key dynamic characteristics
of the structure and correlate to the damages.
to account for variations in loading and environmental conditions. To distinguish between the useful
and poor features, the relationship between DSFs and the structure’s physical parameters should be
investigated. FE procedures are as follows:
1. Load and visualize the data obtained from sensor i, denoted by xi(t)
2. Data are standardized prior to any subsequent analyses
3. Prepossess the signal to remove the artifacts such as all trends and environmental conditions
4. Choose a wavelet function and a set of scales to analyze.
5. Decompose the measured signal into various frequency bands
6. Study every frequency band component with a resolution matched to its scale.
7. Extract DSFs that define the key structure dynamic characteristics that correlated well
with the damage.
8. Choose the features exposing the best classification accuracy of these features.
The basic analyzed wavelet ϕ(𝑡) is a square integrable function, and it meets the following
relationship:
2
ψ (ω )
CΨ =
∫ R ω
dω < ∞ (1.3)
where s and τ are used to adjust the frequency and time location. Ws,τ shows how closely ψ s,τ (t ) cor-
related with x(t). High scale gives low frequency (LF) and low scale offers HF. HF in a given scale is
an indicator to the signal frequency content at that time [29]. At each scale, there is a corresponding
characteristic frequency Fa that relates to s, time rate Δt, and wavelet center frequency Fc as
Fa = Fc ( s ⋅ Δt ) (1.4)
1. The Fourier transforms for the output and input signals are calculated.
2. The frequencies that correspond to 10% and 90% cumulative Fourier amplitudes are determined.
Damage-Sensitive Features Extraction 5
3. The maximum and minimum frequencies determined above are taken as the outer bound
frequencies.
4. The frequency boundaries are converted to wavelet scales to determine the scale bounds.
5. The scales at which the wavelet transform will be evaluated are linearly spaced between
the scale bounds.
Wavelet Energy at a given scale can indicate damages such as loss of energy and how slowly the
energy decays. Nair and Kiremidjian [15] stated that Ws,τ shift in a given scale is described by the
energy as
∑ Wx ( a,b)
2
Ea = (1.5)
b=1
For a multidegree system with mass (M), damping (C), and stiffness K, subjected to force g(t),
damping ratio in each mode is ξ. The transfer response function of the displacement at the kth
degrees of freedom is derived as
N
φkrφlr
H kl (s) = ∑ (ω
r =1
2
r − s 2 + 2 jξω r s )
(1.6)
N
− s 2φkrφlr Gl (s)
X k (s ) = ∑(
r =1
ω r2 − s 2 + 2 jξω r s )
(1.7)
∞ N
− s 2φkrφlr Gl (s)
∫∑( exp ( jsb ) exp ⎢⎡ − ( − as + ω 0 ) ⎥⎤ ds
a 1 2
WM x k (a,b) = (1.8)
2π
−∞ r =1
ω r − s + 2 jξω r s
2 2
) ⎣ 2 ⎦
Instantons Energy is a feature that provides the distribution energy in each frequency band and is
given as
Ns
Teager Energy is a nonlinear operator used to track the signal energy fluctuations and is defined as
⎛ 1 N s −1 ⎞
TE = log10 ⎜
⎜⎝ N s ∑ (W
s=1
s,τ )2 − (Ws,τ (t − 1) ⋅Ws,τ (t + 1)) ⎟⎟
⎠
(1.10)
Wavelet Entropy: Entropy change rate CR(t) can be used as a damage index to describe the struc-
tural state changes and defined as
where swt (0) is undamaged entropy and swt (t ) for damaged. Approximate Entropy which is a mea-
sure of the irregularity and nonlinear time series complexity can be used as a damage feature. It is
computed as
ApEn = Φ m − Φm + 1 (1.12)
N − m +1
where Φm = ( N − m + 1) −1
∑ log( N ), where N is the number within range points.
i =1
i i
1.4 CONCLUSIONS
In this study a sequence of DSFs for civil structures under arbitrary excitation has been proposed.
DSFs are derived from the acceleration signals by using CWT and both the input ground motion
and the output are taken into account. Here, various algorithms are identified and used in order to
track the change of the structures and to extract the damage features. The procedures for extracting
these features are completely data-driven; there are no assumptions about the structure and there is
minimal user-input besides the input and output signals. These features could be able to distinguish
not only a few dynamic characteristics for detecting the location and the damage extent but also the
damage types in a structure. Further, it can identify the slightest damage in a large, complex system
in a noisy environment subjected to arbitrary dynamic loads.a
Damage-Sensitive Features Extraction
FIGURE 1.2 Single representations: (a) time domain and (b) frequency.
7
8 Experimental Vibration Analysis for Civil Structures
REFERENCES
1. W. A. Altabey, An exact solution for mechanical behavior of BFRP nano-thin films embedded in NEMS,
Advances in Nano Research, vol. 5, no. 4, 2017, pp. 337–357. doi: 10.12989/anr.2017.5.4.337.
2. W. A. Altabey, A study on thermo-mechanical behavior of MCD through bulge test analysis, Advances
in Computational Design, vol. 2, no. 2, 2017, pp. 107–119. doi: 10.12989/acd.2017.2.2.107.
3. Z. Ying, N. Mohammad, A. W. A. Altabey, Z. Wu, Fatigue damage identification for composite pipe-
line systems using electrical capacitance sensors, Smart Materials and Structures, vol. 27, no. 8, 2018,
p. 085023. doi: 10.1088/1361-665x/aacc99.
4. Z. Ying, N. Mohammad, W. A. Altabey, R. Ghiasi, Z. Wu, A fatigue damage model for FRP composite
laminate systems based on stiffness reduction, Structural Durability and Health Monitoring, vol. 13,
no. 1, 2019, pp. 85–103, doi: 10.32604/sdhm.2019.04695.
5. Z. Ying, N. Mohammad, W. A. Altabey, R. Ghiasi, Z. Wu, Deep learning-based damage, load and sup-
port identification for a composite pipeline by extracting modal macro strains from dynamic excitations,
Applied Sciences, vol. 8, no. 12, 2018, p. 2564. doi: 10.3390/app8122564.
6. W. A. Altabey, N. Mohammad, Detection of fatigue crack in basalt FRP laminate composite pipe using
electrical potential change method, Journal of Physics Conference Series, vol. 842, 2017, p. 012079.
doi: 10.1088/1742-6596/842/1/012079.
7. W. A. Altabey, Delamination evaluation on basalt FRP composite pipe by electrical potential
change, Advances in Aircraft and Spacecraft Science, vol. 4, no. 5, 2017, pp. 515–528. doi: 10.12989/
aas.2017.4.5.515.
8. W. A. Altabey, EPC method for delamination assessment of basalt FRP pipe: Electrodes number effect,
Structural Monitoring and Maintenance, vol. 4, no. 1, 2017, pp. 69–84. doi: 10.12989/smm.2017.4.1.069.
9. W. A. Altabey, N. Mohammad, Monitoring the water absorption in GFRE pipes via an electrical
capacitance sensors, Advances in Aircraft and Spacecraft Science, vol. 5, no. 4, 2018, pp. 411–434. doi:
10.12989/aas.2018.5.4.499.
10. W. A. Altabey, FE and ANN model of ECS to simulate the pipelines suffer from internal corrosion,
Structural Monitoring and Maintenance, vol. 3, no. 3, 2016, pp. 297–314. doi: 10.12989/smm.2016.3.3.297.
11. H. Yoon, K. Yang, C. Shahabi, Feature subset selection and feature ranking for multivariate time series,
IEEE Transactions on Knowledge and Data Engineering, vol. 17, no. 9, 2005, pp. 1186–1198.
12. W. J. Staszewski, Intelligent signal processing for damage detection in composite materials, Composites
Science and Technology, vol. 62, no. 7–8, 2002, pp. 941–950.
13. P. W. P. Man, M. H. Wong, Efficient and robust feature extraction and pattern matching of time series by
a lattice structure. In Proceedings of the Tenth International Conference on Information and Knowledge
Management, Atlanta, GA 2001, pp. 271–278, ACM.
14. K. P. Chan, A. W. C. Fu, Efficient time series matching by wavelets. In Proceedings 15th International
Conference on Data Engineering (Cat. No. 99CB36337), Sydney, Australia, 1999, pp. 126–133, IEEE.
15. K. K. Nair, A. S. Kiremidjian, Time series based structural damage detection algorithm using Gaussian
mixtures modeling, Journal of Dynamic Systems, Measurement, and Control, vol. 129, no. 3, 2007,
pp. 285–293.
16. S. Pittner, S. V. Kamarthi, Feature extraction from wavelet coefficients for pattern recognition tasks,
IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 21, no. 1, 1999, pp. 83–88.
17. K. Balafas, A. S. Kiremidjian, Extraction of a series of novel damage sensitive features derived from
the continuous wavelet transform of input and output acceleration measurements, In Sensors and Smart
Structures Technologies for Civil, Mechanical, and Aerospace Systems, vol. 9061, 2014, p. 90611L.
International Society for Optics and Photonics.
18. S. S. Patel, A. P. Chourasia, S. K. Panigrahi, J. Parashar, N. Parvez, M. Kumar, Damage identification
of RC structures using wavelet transformation, Procedia Engineering, vol. 144, 2016, pp. 336–342.
19. M. Noori, H. Wang, W. A. Altabey, A. I. Silik, A modified wavelet energy rate-based damage identifica-
tion method for steel bridges, Scientia Iranica, vol. 25, 2018, pp. 3210–3230.
20. Z. Ying, N. Mohammad, B. B. Seyed, W. A. Altabey, Mode shape based damage identification for a
reinforced concrete beam using wavelet coefficient differences and multi-resolution analysis, Structural
Control and Health Monitoring, vol. 25, no. 3, 2017, pp. 1–41. doi: 10.1002/stc.2041.
21. Z. Ying, N. Mohammad, W. A. Altabey, Damage detection for a beam under transient excitation via
three different algorithms, Structural Engineering and Mechanics, vol. 63, no. 6, 2017, pp. 803–817. doi:
10.12989/sem.2017.64.6.803.
Damage-Sensitive Features Extraction 11
22. Z. Ying, N. Mohammad, W. A. Altabey, T. Awad, A comparison of three different methods for the iden-
tification of hysterically degrading structures using BWBN model, Frontiers in Built Environment, vol.
4, 2019, p. 80. doi: 10.3389/fbuil.2018.00080.
23. Z. K. Xie, G. H. Liu, Z. H. Zhang, Damage detection of concrete structure based on approximate
entropy, Applied Mechanics and Materials, vol. 226, 2012, pp. 920–925. Trans Tech Publications.
24. W. A. Altabey, Free vibration of basalt fiber reinforced polymer (FRP) laminated variable thick-
ness plates with intermediate elastic support using finite strip transition matrix (FSTM) method,
Vibroengineering, vol. 19, no. 4, 2017, pp. 2873–2885. doi: 10.21595/jve.2017.18154.
25. W. A. Altabey, Prediction of natural frequency of basalt fiber reinforced polymer (FRP) laminated vari-
able thickness plates with intermediate elastic support using artificial neural networks (ANNs) method,
Vibroengineering, vol. 19, no. 5, 2017, pp. 3668–3678. doi: 10.21595/jve.2017.18209.
26. W. A. Altabey, High performance estimations of natural frequency of basalt FRP laminated plates with
intermediate elastic support using response surfaces method, Vibroengineering, vol. 20, no. 2, 2018,
pp. 1099–1107. doi: 10.21595/jve.2017.18456.
27. Z. Hou, A. Hera, M. Noori, Wavelet-Based Techniques for Structural Health Monitoring. In: Health
Assessment of Engineered Structures: Bridges, Buildings and Other Infrastructures, ed., Achintya
Haldar (University of Arizona), World Scientific, Singapore, 2013, pp. 179–202.
28. Z. Hou, M. Noori, R. S. Amand, Wavelet-based approach for structural damage detection, Journal of
Engineering Mechanics, vol. 126, no. 7, 2000, pp. 677–683.
29. D. Hester, A. González, A wavelet-based damage detection algorithm based on bridge acceleration
response to a vehicle, Mechanical Systems and Signal Processing, vol. 28, 2012, pp. 145–166.
2 Deep Learning for Automated
Damage Detection
A Novel Algorithm in CNN
Family for Faster and Accurate
Damage Identification
Wael A. Altabey
Southeast University
Alexandria University
Mohammad Noori
California Polytechnic State University
CONTENTS
2.1 Introduction ............................................................................................................................ 13
2.2 Deep Learning Based Approaches for Image Classification and Object Detection............... 15
2.2.1 Deep Learning for Image Classification..................................................................... 15
2.3 Methodology ........................................................................................................................... 17
2.4 Conclusions ............................................................................................................................. 23
Acknowledgements .......................................................................................................................... 23
References ........................................................................................................................................ 23
2.1 INTRODUCTION
For industrial and civil structures, structural health monitoring is more complicated, especially
monitoring in a real time when the types of structures are complicated as well as when the measure-
ment signal from sensors is corrupted with environmental noise. Therefore, structural health moni-
toring needs other techniques to identify structural damage. Deep Learning can solve this problem
satisfactorily due to its superior adaptive learning of datasets [1–12].
Krizhevsky et al. [13] classify 1.2 million high-resolution images by using a large and deep convo-
lutional neural network (CNN). Bouvrie [14] presented the derivation and implementation of CNNs.
The detection and localization of damage based on real-time vibrations have been proposed using 1D
CNNs [15–19]. A sparse coding algorithm was applied to a large number of untagged examples to
train a feature extractor, and the features were then used to pass to a neural network classifier to dis-
tinguish various damage states from bridges [20]. A type of deep architecture CNN has been used to
detect cracks in concrete without directly extracting the features of the defects, to overcome challenges
such as lighting and shadow changes [21]. The damage in steel structures was indicated by using the
differences in wavelet coefficients to analyze the responses of a healthy and loose connection structure
[22,23]. The wavelets were used to detect strong earthquake response waves of framed structures indi-
cating the occurrence of damage [24]. A new signal processing algorithm was proposed by the deep
learning model, which consists of combined wavelets, neural networks, and Hilbert transform [25,26].
13
14 Experimental Vibration Analysis for Civil Structures
In addition, three types of deep neural network models, such as deep Boltzmann machines, deep belief
networks, and stacked auto-encoders, were investigated to identify rolling bearing failure conditions
[27]. Compared to a traditional support vector machine (SVM) and backpropagation neural network,
planetary gearbox conditions could be effectively detected with the best diagnostic accuracy using
deep CNNs [28]. The benefits of generalization and memorization ability for recommending systems
were combined by using the wide and deep common learning by applying wide linear models and
deep neural networks [29]. A multi-scale structural health monitoring system has been constructed
to monitor health conditions and assess the usefulness of large-scale bridges using the Hadoop eco-
system (MS-SHM-Hadoop). A Bayesian network has been studied to assess the reliability of specific
components as a function of ease of maintenance and inter-component correlations [30] and a hybrid
response surface method [31]. Unlike buildings or bridges, for pipeline networks system, damage usu-
ally begins on the internal surface, especially when the internal temperature is higher than the external
temperature. Micro damage, cracking of the matrix [32], delamination [33,34], Webepage [35], and
fiber failure are the damage cases that result in ultimate failure of fiber reinforced polymers (FRP)
composite structure, and damages may probably initiate and concentrate inside the pipes rather than
outside the surface. Previous studies have shown that by acquiring images from structural damage
areas (mostly outside surfaces of structures), it may be good to distinguish different health states of
structures. Images captured from the external surface of pipelines may not be an appropriate way to
identify damage. The specific purpose of current work is to identify and locate closed-circuit television
(CCTV) image pipeline failures through deep learning to overcome the limitations of conventional
computer vision techniques. Automatic detection of image faults and binary segmentation (mapping)
with very high precision were established; the current disk format can have a high degree of accuracy
in measuring the type and location of image errors, especially in mini image errors. The current
approach provides pointers for others to apply in-depth learning techniques that address similar topics,
such as identifying and locating civic infrastructure failures in building and property management.
Practical implementation of the proposed approach in industry is expected to significantly reduce
inspection time and resources and improve the efficiency of pipeline status assessment.
The present work generally relates to a new algorithm in Region-based CNN (R-CNN) family
architecture and is called Faster Dual/Multi R-CNN (Faster D/M-R-CNN) since it provides a fast
and accurate damage detection and classification for various structures and especially pipelines at real
time. This method has the potential to be used as the fastest system for damage identification in various
structures such as different infrastructures (bridges, high buildings, dams, pipelines, tanks, etc.), for
traffic control system, and transportation systems. It can also be used for image analysis and processes
occurring in smart cities and traffic control and transportation systems. Figure 2.1 shows the block dia-
gram depicting the new algorithm (Faster D/M-R-CNN) for damage detection and image classification.
FIGURE 2.1 A block diagram depicting the new algorithm (Faster D/M-R-CNN) for damage detection and
classification.
Deep Learning for Automated Damage Detection 15
can be predicted. One limitation of R-CNN is that the multi-staged training process which is time
consuming requires a high computation cost. In addition, the detection speed is quite slow for each
image as convolution, classification, and regression need to be implemented for each region’s proposal.
Among the many methodologies, deep learning based damage detection is actively explored in
recent years. However, how to automatically extract damage quickly and accurately at a pixel level, i.e.
damage delineation (including both detection and segmentation) at real time, is a challenging issue.
This work proposes a new algorithm in R-CNN family-based architecture that is Faster D/M-R-
CNN for real-time object (damage) detection and classification with high accuracy.
Let’s quickly summarize the different algorithms in the R-CNN family (R-CNN, Fast R-CNN,
and Faster R-CNN).
R-CNN extracts a bunch of regions from the given image using selective search, and then checks
if any of these boxes contains an object. We first extract these regions, and for each region, CNN is
used to extract specific features. Finally, these features are then used to detect objects. Unfortunately,
R-CNN becomes rather slow due to the multiple steps involved in the process.
Fast R-CNN, on the other hand, passes the entire image to ConvNet which generates regions of
interest (instead of passing the extracted regions from the image). Also, instead of using three differ-
ent models (as we saw in R-CNN), it uses a single model which extracts features from the regions,
classifies them into different classes, and returns the bboxs.
All these steps are done simultaneously, thus making it execute faster as compared to R-CNN.
Fast R-CNN is, however, not fast enough when applied on a large dataset as it also uses selective
search for extracting the regions.
Faster R-CNN fixes the problem of selective search by replacing it with Region Proposal Network
(RPN). We first extract feature maps from the input image using ConvNet and then pass those maps
through an RPN which returns object proposals. Finally, these maps are classified and the bboxs
are predicted.
A block diagram depicting the new algorithm (Faster D/M-R-CNN) for object (damage) detec-
tion and classification is shown in Figure 2.1. As shown in the figure, apply Dual/Multi RPN (D/M-
RPN) {3} on these feature maps {2}, get Dual/Multi object (damage) proposals for each candidate
object (damage) in the image, and compare between these proposals to create a confidence score
representing a confidence that the each candidate object (damage) detected in the bboxs {3C} is the
desired object (damage) meticulously. This algorithm increases the detection accuracy in short time,
without following the traditional method followed in other networks in the R-CNN family by sup-
plementing more images into the database to reduce overfitting and to improve detection accuracy.
Faster D/M-R-CNN has high precision and recall value and very high speed for extracting all
object (damage) features from images at real time, which is important for accurately detecting dam-
ages from images collected that increased previous damage detection systems ability to achieve
real-time detection.
Therefore, the Faster D/M-R-CNN will be investigated and applied in this invented product
for automated structure damage and object detection from image. We expected that the proposed
approach will be demonstrated to be applicable for detecting structure damages and object detection
with high accuracy and speed for extracting all damage features from images.
We expected this innovative product will increase the speed and accuracy of object (damage)
detection and improves dataset size, training mode, and network hyper-parameters, which have
influence on model performance.
Specifically, when an increase of dataset size and convolutional layers can improve the model
speed and accuracy, it can achieve a mean average precision (mAP) of up to 98%–99% approximately.
The new designed network will lay the foundation for applying a new generation of deep learning
techniques in structural damage detection systems as well as addressing deficiencies in the previous
structural damage detection systems based on deep learning.
The comparison between the algorithms in R-CNN family and the new algorithm described
herein is shown in Table 2.1.
Deep Learning for Automated Damage Detection 17
TABLE 2.1
Comparison between the Algorithms in R-CNN Family and the New Algorithm
(Faster D/M-R-CNN) Described Herein Using Damage Detection and Classification
Prediction
Algorithm Features Time/Image Limitations
CNN Divides the image into multiple regions and then – Needs a lot of regions to predict
classifies each region into various classes accurately and hence high
computation time
R-CNN Uses selective search to generate regions. Extracts 40–50 High computation time as each
around 2,000 regions from each image seconds region is passed to the CNN
separately. Also, it uses three
different models for making
predictions
Faster Each image is passed only once to the CNN and feature 2 seconds Selective search is slow and hence
R-CNN maps are extracted. Selective search is used on these computation time is still high
maps to generate predictions. Combines all the three
models used in R-CNN together
Faster Replaces the selective search method with RPN, which 0.2 seconds Object proposal takes time and as
R-CNN makes the algorithm much faster there are different systems
working one after the other, the
performance of systems depends
on how the previous system has
performed
Faster Applied D/M-RPN to get Dual/Multi object (damage) 0.1–0.2 Needs to obtain an optimum
D/M-R- proposals for each candidate object in the same image seconds number of RPN to investigate
CNN and compare between these proposals to result in the the target
desired object, which makes the algorithm with the
highest accuracy and the fastest
2.3 METHODOLOGY
As shown in Figure 2.1, the steps followed by a Faster D/M-R-CNN algorithm to detect objects
(damage) in various structure image are summarized below:
1. Take an input image {1} and pass it to the Deep-CNN {2A} which returns feature maps {2}
for the image.
2. Apply D/M-RPN {3} on these feature maps {2} and get Dual/Multi object (damage) pro-
posals for each candidate object (damage) in the image.
3. Compare between the Dual/Multi object (damage) proposals to create a confidence score
representing a confidence that the object (damage) detected in the bboxs {3C} is the desired
object (damage).
4. Pass these proposals to an FC {3B} layer in order to classify and predict the bboxs {3C} for
the image.
5. Finally, receive the final image results {4} and compute a confidence score associated with
it. The computing device can then provide a requestor with an output including object clas-
sification and/or confidence score.
The flowchart of the new algorithm process steps of the Faster D/M-R-CNN is introduced in Figure 2.4.
To ensure the accuracy of the object (damage) detection and classification network, the D/M-
RPN {3} can be trained with training images. A discussion of the object (damage) detection and
classification network training will be discussed in greater detail with regard to Figure 2.5.
18 Experimental Vibration Analysis for Civil Structures
FIGURE 2.4 A flow diagram that illustrates a process flow for damage detection and classification network.
The Faster D/M-R-CNN training process consists of four steps. The first step is training a D/M-
RPN {3} initialized from pre-trained Deep-CNN {2A}, in this case generating the Dual/Multi object
(damage) proposals using the trained D/M-RPN {3}. The second step is training a Faster D/M-R-
CNN initialized from the pre-trained Deep-CNN on the Dual/Multi object (damage) obtained by
the first step. In the third step, the D/M-RPN {3} is trained again using the weights from step two as
shown in Figure 2.4, without changing the Deep-CNN {2A} layers.
After generating new Dual/Multi object (damage) proposals from the trained D/M-RPN {3},
Faster D/M-R-CNN is trained again in step four with the parameters trained in the previous step.
The classification process is presented in Figure 2.6. A series of images from the image of a
single target (i.e., as a sequence of temporally consecutive frames of the structures damage types)
are fed to the D/M-CNN that is applied to feature extraction.
The two-step learning method is applied, i.e., the D/M-CNN is trained with the first N − 1 layers
viewed as feature maps, and these maps are used to train a Dual/Multi SVM (D/M-SVM) classi-
fier. The output of the SVM from each CNN is compared with them and all damage features from
images with high accuracy are collected and presented as P-Tensors as follows:
FIGURE 2.5 A flow diagram that illustrates a process flow of training one or more parameters of an object
(damage) detection network.
20 Experimental Vibration Analysis for Civil Structures
FIGURE 2.6 Data flow diagram depicting an example architecture of the process of the damage
classification network described in Figure 2.1.
where ci , j is the probability for class ( i, j ), nc is the number of classes, and n is the number of images
of training examples; thus there will be one P-Tensor for each image in any given image series.
The composition of the tensor is the following: calculate the average of the size estimates of the
image series, check from the size-look-up table all the classes that contain the average size, e, and
turn those elements to one and set the others to zero, yielding
With elements:
⎧⎪ 1 if e fitsclass i, j,
e(i , j ) = ⎨ (2.3)
⎩⎪ 0 otherwise.
The velocity of the target damage type is composed in a similar way as the E-Tensor in Size Estimate
(2.2), i.e., check from the velocity-look-up table all the classes that contain the provided velocity, v,
and turn those elements to one and the others to zero.
Phot.: Andrae.
Der Tigris bei Assur.
Neun Jahre verwandten Andrae und seine Mitarbeiter auf die
Freilegung der Festungswerke Assurs (vgl. „Die Festungswerke von
Assur“ von Walter Andrae. 2 Bde., 1913), denn ihre Bestimmung war
nicht nur wichtig für die Erkenntnis der ganzen Stadtanlage, für
Ermittlung ihrer Zugänge und zugleich des Verlaufs der wichtigsten
Handelsstraßen, sondern auch für die Geschichte der
Befestigungskunde überhaupt, da man assyrische Festungen bis
dahin nur ungenügend kannte. Die Arbeit war um so schwerer, als
die Mauern am Rande des Stadthügels naturgemäß am meisten der
Vernichtung ausgesetzt waren, und obendrein der Tigris den größten
Teil der Ostfront zerstört hatte.
Assur liegt auf der Spitze eines Ausläufers der Chanukekette,
und der Platz war für eine Festung wie geschaffen. Im Osten
bespülte ihn der schnellfließende, das ganze Jahr über tiefe Tigris,
ein Angriff von dort war also unmöglich. Im Norden fiel der Fels
(weicher Sandstein und Kieselkonglomerat) jäh nach einem
Stromarm ab, der trefflich als Festungsgraben diente. Am Rande
dieses noch erkennbaren Flußbettes hatten wir unser Lager
aufgeschlagen. Vor der Westfront erleichterten zwei kleine Täler die
Anlage von Gräben, die nur da zugeschüttet waren, wo Straßen zu
den Toren führten. Im Süden war eine Geländesenkung. Der einzige
Nachteil war, daß man von dem Hügelplateau im Westen aus in die
Stadt hineinsehen konnte. Deshalb baute man die Westmauer am
höchsten.
Das Alter der Festungsbauten Assurs ist sehr verschieden.
Andrae unterscheidet die archaische Zeit bis zur Mitte des 2.
vorchristlichen Jahrtausends, die altassyrische bis Ende des 2.
Jahrtausends, die jungassyrische vom Anfang des 1. Jahrtausends
bis Sargon, die spätassyrische unter Sargon und den Sargoniten bis
zur Zerstörung des assyrischen Reiches im Jahre 606 v. Chr., die
nachassyrische der Wiedereinwanderung unter den Neubabyloniern
und Cyrus (6. Jahrhundert) und die parthische Zeit, die ersten zwei
Jahrhunderte vor und nach Christus.
Phot.: Andrae.
Der Strand von Assur.
Aus vorgeschichtlicher Zeit haben sich keine Befestigungen
gefunden, nur Grundmauern von Häusern, Feuerstätten und
Kanälen. An der Ostseite führte man schon zu Anfang des 2.
Jahrtausends Mauern auf, um die Stromfahrt zu beherrschen und
das Ufer gegen die Erosion zu schützen. Diese Mauern befestigte
Adadnirari I. in altassyrischer Zeit. Davon ist noch vieles erhalten.
Auch legte man Landeplätze und Treppen am Ufer an. An der
Nordostecke der Stadt lag der Assurtempel mit der Front nach
Norden, und an der Nordwestecke der Palast Tukulti-Ninibs I. auf
seiner ungeheuren Plattform.
In jungassyrischer Zeit baute Salmanassar III. im Westen und
Südwesten eine äußere und eine innere senkrechte Mauer, auf
denen je eine Fahrstraße hinlief. Blaugelbe und schwarzweiß
glasierte Ziegel schmückten die Zinnen. Nach seiner Regierung,
aber vor Sargon, verfiel die innere Mauer; an ihrer Stelle entstanden
Wohnhäuser, und davor legte man eine niedrigere Mauer an.
In spätassyrischer Zeit führten Sargon und Sanherib noch
mancherlei Verbesserungen aus. Die Achämeniden dagegen ließen
die Befestigungen unverändert, und auch in der parthischen Zeit
wurde nichts daran getan.
Aus der Zeit Salmanassars III. grub man sieben Tore aus. Jedes
Tor flankierten zwei Türme; nach innen waren Wachtstuben,
Rampen und Treppen, die zur Mauerzinne hinaufführten. Eines der
Tore hieß Abul gurgurri, das Stadttor der Metallarbeiter; die übrigen
sind bisher namenlos. Die Zapfen der gewaltigen Flügel des
Gurgurritores, zylinderförmige Basaltblöcke, sind noch vorhanden.
Verkohlte Zedernholzbalken lassen auf eine Feuersbrunst schließen.
Zwei sargonitische Kalksteinblöcke an diesem Tor tragen Sanheribs
Namen. Eine Bildsäule Salmanassars III., die im Gurgurritor stand,
besitzt jetzt das Ottomanische Museum in Konstantinopel, eine
andere, die den König auf seinem Throne sitzend darstellt, das
Britische Museum. Beide sind in Lebensgröße.
Der ebenfalls ausgegrabene, offenere Zugang von Norden her,
den man nicht als Tor bezeichnen kann, hieß Muschlal und wird
schon Ende des 3. Jahrtausends auf Ziegelinschriften erwähnt.
Adadnirari I. in altassyrischer Zeit ließ ihn erneuern; auch die
jungassyrische Zeit unter Salmanassar III. kennt ihn. Unter Sanherib
heißt es: „Der Palast Muschlal in der Stadt Assur“, und bei
Assarhaddon: „Bît muslalu, das am Palast der Stadt Assur liegt, ließ
ich aufs neue erbauen als Ein- und Ausgang“.
Straßen an der inneren Mauer stammen aus spätassyrischer
Zeit; sie erinnern an die Straßen Pompejis und der heutigen Städte
des Orients.
Professor Andrae beschreibt ausführlich alle Einzelfunde, die an
den Mauern gemacht wurden, Ziegelkanäle, Straßen und Häuser,
Abflußtrommeln, Poternen, Wehrgänge, Turmtreppen, Bastionen,
emaillierte Terrakottareliefs, Kupferbecken, Konsolen, Haken und
Pfeilspitzen aus Bronze, Gräber und Ziegel mit Inschriften, von
denen folgende aus der Zeit Salmanassars III. als Probe angeführt
sei: „Salmanassar, der König des Alls, König des Landes Assur, der
Sohn des Assurnasirpal, des Königs des Landes Assur. Erobernd
herrschte ich vom großen Meer beim Lande Amurru gegen
Sonnenuntergang bis zum Meer beim Lande Kaldu, genannt Marratu
(d. h. der Salzstrom). Da brach ich die Ruinen der früheren
Festungsmauer meiner Stadt Assur nieder, die Tukulti-Ninib,
Salmanassars Sohn, ehedem gebaut hatte; ich erreichte ihren
Grund; von ihrem Fundament bis zu ihrer Brustwehr fügte und
vollendete ich sie; prächtiger und gewaltiger als zuvor machte ich
sie. Meine Tafeln und Urkunden brachte ich an. Ein zukünftiger Fürst
soll ihre Ruinen wieder aufrichten und meinem Namen wieder seinen
Platz einräumen, dann wird Assur seine Gebete erhören.“
Salmanassar gedachte also der kommenden Jahrtausende, die
seinen Namen vergessen könnten. Dann sollten die Steine für ihn
reden!
Andraes Grabungsmethode in den Ruinen Assurs war eine
andere als die Koldeweys in Babylon. Er zog 5 Meter breite
„Suchgräben“ quer über das ganze Stadtgebiet; sie laufen je 100
Meter voneinander entfernt parallel von der Westmauer bis nach
dem Tigrisufer im Osten. Stieß solch ein Graben auf Reste von
Palästen, Mauern, Toren, Häusern, Kanälen usw., so grub man
seitwärts weiter, bis der ganze Fund bloßgelegt war. Manchmal
zwangen Bodengestaltung oder neuere mohammedanische
Grabstellen zur Aufgabe des 100-Meter-Zwischenraums. Solch ein
unregelmäßiger Graben führte in den Jahren 1909–1911 zur
Entdeckung der merkwürdigen Königspfeiler im Winkel zwischen
dem breiten Nordteil und dem schmalen Südteil der Stadt. (Vgl. „Die
Stelenreihen in Assur“ von Walter Andrae, Leipzig, 1913).
Diese Pfeiler stammen aus der Zeit zwischen dem 14. und 7.
Jahrhundert v. Chr. Sie sind flach, oben abgerundet und tragen eine
Inschrift oder ein Reliefbild der Personen, zu deren Gedächtnis sie
errichtet wurden. Die größten sind aus Basalt und nennen Tukulti-
Ninib I., Semiramis und Assurnasirpal III.; kleinere sind mit den
Namen anderer Könige und hoher Beamten bezeichnet. Einer aus
körnigem, gelbgrauem Kalkstein zeigt das Bild einer Palastdame
Sardanapals. Sie sitzt, nach rechts gewendet, auf einem Thron, ist
mit Armbändern und Ohrringen geschmückt, trägt Rosetten auf den
Schultern und auf ihren üppigen, den Rücken herabwallenden
Locken eine Königskrone; in der Linken hält sie eine Blume, die
Rechte streckt sie nach oben. Das Gesicht entspricht dem
Schönheitsideal des Orients: volle runde Wangen, kräftiges Kinn,
gerade, scharf gezeichnete Nase, schön geschwungene, breite
Augenbrauen und lachende Lippen.
Phot.: Schölvinck.
Ein Suchgraben in Assur.
Phot.: Schölvinck.
Das deutsche Expeditionshaus in Assur.
Einundzwanzigstes Kapitel.
Erlebnisse auf einer
Etappenstraße.
Giara.
Das Bahnhofsgebäude von Giara hatte nur einen bewohnbaren
Raum, eine ungewöhnlich kühle, gewölbte Kammer, in der der
Stationsvorsteher unter einem von Fliegen umschwirrten
Mückennetz an Ruhr erkrankt darniederlag und aus einem primitiven
Filtrierapparat, einem großen Lehmkrug mit porösem Boden, Wasser
tropfen ließ. Hier mußten wir die heißesten Tagesstunden abwarten,
denn die Temperatur draußen war allmählich unerträglich geworden.
Schon morgens um 7 Uhr hatte sie 31 Grad betragen, um 1 Uhr
stieg sie auf 41,2 und anderthalb Stunden später auf 42,6 Grad.
Konsul Schünemanns persischer Schimmel hatte einen Hitzschlag
und Kolik und außerdem Blutegel in Gaumen und Hals. Noch am
Morgen war das Tier ganz frisch gewesen; jetzt legte es sich im
Schatten des Stalles nieder und verendete. Auch im Schlund der
andern Pferde hatten sich beim Trinken Blutegel festgebissen, und
unsere Kutscher befreiten sie mit vieler Mühe von diesen
Plagegeistern.
In der Kranken- und Fliegenstube von Giara zu übernachten, war
unmöglich. Am Spätnachmittag machten wir uns daher zur nächsten
Station Schura auf, die fünf Stunden entfernt sein sollte. Nahe bei
Giara hatten wir ein ziemlich tief und steil eingeschnittenes Tal zu
passieren, auf dessen nackter Sohle Salzkristalle schimmerten und
Erdpechquellen zutage traten. Der Herzog und Busse ritten voraus;
Schölvinck und ich folgten in der Droschke und fuhren in einer
Morastrinne fest. Die Pferde mußten ausgespannt, der Wagen
zurückgeschoben und ein anderer Weg versucht werden. Nicht
besser erging es dem vorausfahrenden Automobil, das weiter vorn in
einem Graben saß und nicht weiter konnte. Wir luden das Gepäck
ab, aber der Wagen rührte sich nicht vom Fleck, und wir mußten
warten, bis die ganze übrige Kolonne nachgekommen war. Darüber
wurde es dunkel, und im Westen erhob sich drohend eine
Wolkenwand, die den Mond verdeckte. Nach langem Warten kamen
die andern, und mit vereinten Kräften machten wir erst das Auto
wieder flott, das nunmehr jeden einzelnen Wagen über die
schwierige Stelle hinüberziehen mußte; die müden Tiere allein
hätten das nicht fertiggebracht. Drei Stunden kostete uns dieser
Graben — eine schöne Etappenstraße!
Dann ging es weiter, Stunde auf Stunde in stockfinsterer Nacht;
die Lampen des Autos wiesen den Weg. Endlich leuchtete vor uns
der Schein eines Feuers auf: es war Schura, aber noch in weiter
Ferne. Ein neuer Graben hielt die Wagen auf; unsere Droschke kam
glücklich hinüber, und endlich tauchte die hohe Mauer des
Stationsgebäudes aus dem Dunkel hervor. Hastig aßen wir auf dem