Hybrid Computational Intelligence Approach For Defect Identification In Vietnamese Structures

Hybrid Computational Intelligence Approach for Defect Identification in Vietnamese Structures Viet Long Ho Doctoral dissertation submitted to obtain the academic degree of Doctor of Civi

Hybrid Computational Intelligence Approach for Defect Identification

in Vietnamese Structures

Viet Long Ho

Doctoral dissertation submitted to obtain the academic degree of

Doctor of Civil Engineering

Prof Magd Abdel Wahab, PhD* - Prof Tien Thanh Bui, PhD** - Prof Em Guido De Roeck, PhD***

* Department of Electromechanical, Systems and Metal Engineering

Faculty of Engineering and Architecture, Ghent University

** Department of Bridge Engineering and Underground Infrastructure

Faculty of Civil Engineering, University of Transport and Communications, Vietnam

*** Department of Civil Engineering

Faculty of Engineering Science, KU Leuven

Supervisors

June 2022

Wettelijk depot: D/2022/10.500/34 NUR 956

ISBN 978-94-6355-593-7

Members of the Examination Board

Chair

Prof Patrick De Baets, PhD, Ghent University

Other members entitled to vote

Samir Khatir, PhD, Ghent University

Yong Ling, PhD, Ghent University

Prof Timon Rabczuk, PhD, Bauhaus-Universität Weimar, Germany

Prof Hung Nguyen Xuan, PhD, Ho Chi Minh City University of Technology, Vietnam

Supervisors

Prof Magd Abdel Wahab, PhD, Ghent University

Prof Tien Thanh Bui, PhD, University of Transport and Communications, Vietnam Prof Em Guido De Roeck, PhD, KU Leuven

I would like to dedicate this thesis to my loving family

Declaration

I hereby declare that except where specific reference is made to the work of others, the contents of this dissertation are original and have not been submitted

in whole or in part for consideration for any other degree or qualification in this, or any other university This dissertation is the result of my own work and includes nothing, which is the outcome of work done in collaboration, except where specifically indicated in the text This dissertation contains less than 51,000 words including appendices, bibliography, footnotes, tables and equations and has less than 130 figures

Viet Long HO

2022

Acknowledgements

This thesis has been conducted in the Finite Element Modelling Research Group, Department of Electromechanical, Systems and Metal Engineering, Faculty of Engineering and Architecture, Ghent University, Belgium, in the framework of the VLIR-UOS team project VN2018TEA479A103, “Damage assessment tools for Structural Health Monitoring of Vietnamese infrastructures”

Firstly, I would like to express millions of thanks to my supervisors, Prof Magd Abdel Wahab, A.Prof Thanh Bui-Tien, and Em.Prof Guido De Roeck for their guidance, encouragement, support, valuable advice through my research Without their guidance, this thesis would not have been completed

I would like to acknowledge the financial support of the VLIR-UOS scholarships Moreover, I would like to acknowledge the support of Division

of Bridge Engineering and Underground Infrastructure, Campus in Ho Chi Minh City, and the Department of Bridge Engineering and Underground Infrastructure, University of Transport and Communications, Vietnam They have encouraged me and provided the best conditions to facilitate my study

I would like to thank the research team, friends at Ghent University, Dr Duong Nguyen Huong, Dr Hoa Tran-Ngoc, Dr Samir Khatir, Mr Dhanraj Rajaraman and Miss Ens Discussing and collaborating with them helped me improve my research skills, knowledge, and life experience I also really appreciate my colleagues in Vietnam, Dr Phuc Phung-Van, Dr Tien-Dung Dinh, Dr Truong Vu-Huu, who discuss and share their experience with me, colleagues at UCT company and Hanoi University of Civil Engineering for measurement support, OBSG team for accommodation and helpful supports during my stay in Ghent Living and studying in Ghent is an unforgettable memory for me I would like to send my appreciation to all members of Soete Laboratory There is an excellent atmosphere in our office Especially, I would like to thank Mrs Georgette D’hondt, our laboratory secretary You have offered me a lot of helpful assistance

Finally, yet importantly, I would like to express my biggest gratitude to my family, Miss Trang Without their love, unconditional support, and encouragement, I would not have had this success today This doctoral thesis

is also for you

Viet Long HO Ghent, June 2022

Abstract

Maintenance costs and service life of existing bridges are tough challenges for bridge managers During the operation, damage can emerge anywhere in structures with unknown extents It can lead to overall failure or impair the life

of structures Therefore, early damage detection is an utmost key to this problem Structural health monitoring (SHM), an interdisciplinary engineering field, becomes an effective structural damage identification and condition assessment approach An efficient SHM strategy must combine global and local monitoring techniques Local non-destructive testing methods (NDT), e.g acoustic emission (AE) ultrasonic inspection, guided (lamb) waves (GW),

a visual inspection (a classical onsite assessment), can be a promising solution for damage quantification However, these techniques can be utilized to characterize damages and evaluate the corresponding extents if the early phase

of failure, i.e., damage location, is already identified Vibration-based structural damage detection (VBSDD) is a global technique that can assess the safety and integrity of the monitored structures The presence of damage shifts structural parameters, i.e., stiffness, mass, flexibility, or energy dissipation of the considered structures Subsequently, it changes frequency response functions (FRF), modal parameters, i.e., natural frequencies, mode shapes, and damping ratios VBSDD identifies the modal properties to provide a warning against structural damage Among these characteristics, natural frequencies are easier

to identify with one or few sensors

Nevertheless, natural frequencies are almost insensitive to changes in local stiffness Besides, the effect of environmental factors, e.g., temperature variation, can be larger than that of damage in structures In contrast, being more sensitive to local damage and less sensitive to temperature changes is a typical feature of displacement mode shapes By using an affordable number

of sensors, the obtained mode shapes can be used to detect and localize damages Hence, mode shape-based damage detection indices are employed to localize damage in laboratory structures and large-scale bridges

However, to improve the quality of damage quantification, the use of a hybrid computational intelligence approach consisting of optimization algorithms (OAs) or/and machine learning (ML) for VBSDD applications is investigated

in this thesis Global search-ability of novel optimization algorithms is used to enhance the performance of machine learning algorithms, i.e., Feedforward neural network (FNN), Artificial neural network (ANN) as new approaches in damage identification

The contributions of this thesis are introduced as follows:

- A new improved form of mode shape curvature-based damage detection index, IMCI, is introduced for damage localization in a large-scale cable-stayed bridge The damage detection by the IMCI index is based on the use of a threshold indicator The standard deviation and mean values are employed to calculate the threshold values First, the IMCI index is computed at every measured point allocated along the investigated structures Then these indices are truncated by the threshold value to reveal the damage locations The use of the IMCI provides an obvious representation of the fault locations in the investigated structures compared

to other curvature-based indices e.g curvature damage factor (nCDF) and mode shape curvature square (MCIS) Finally, the effectiveness and accuracy of the proposed index in fault localization are verified and confirmed through single and multi-damage scenarios in a hypothetical structure (e.g a two-span continuous beam-like structure) and the tested cable-stayed bridge

- In addition, two new hybrid computational intelligence approaches are developed for damage quantification The first is a novel optimization algorithm, combining marine predator algorithm (MPA), and FNN, namely MPAFNN The second, so-called ALOANN, or MPAANN coupled antlion optimizer (ALO) or marine predator algorithm (MPA) with ANN The two approaches can be presented in three hybrid models:

 The first hybrid model, MPAFNN, is validated using two FE models

of a simply supported beam, a 2-span continuous beam, and an updated FE model of a laboratory beam The input data are provided by the modal flexibility method The output encompasses two components, i.e damaged elements and the corresponding damage levels Result of a stochastic optimization process is a set of the optimal connection weights and biases

in each layer Noise effects are considered by using additive white Gaussian noise (AWGN) based on signal to noise ratio (SNR) It should note that, before applying to damage detection problems, the proposed model, MPAFNN, demonstrates its efficiency and precision by means of several benchmark classification problems, e.g balloon, Iris, breast cancer, and heart disease The results using MPAFNN are compared with that of several well-known optimization algorithms, e.g particle swarm optimization (PSO), gravitational search algorithm (GSA), hybrid algorithm (PSOGSA), and grey wolf optimizer (GWO) Findings confirm that MPAFNN shows superior performance to the others in light of accuracy and precision Values of average mean squared error (MSE), the best MSE, standard deviation MSE, and classification rate demonstrate this conclusion

 The second hybrid model, MPAANN, is combined with the new proposed damage detection index, IMCI, for damage localization and quantification in the updated FE model of a cable-stayed bridge In this model, ANN’s performance is improved by optimizing the initial training parameters, e.g size of the hidden layer, data split for training-validation-test, activation function, and training function The mean squared errors between estimated and desired values are the objective function of the stochastic optimization process The tested bridge is assumed to be degraded at separate segments Damaged segments are localized by using the IMCI index Meanwhile, the proposed hybrid model is employed to identify the level of damage In addition, AWGN is added to the modal properties in order to consider noise effects

 The last hybrid model, ALOANN coupled with a mode shape derivative-based damage identification (MSDBDI), is applied to a FE model of a plate-like structure and an updated FE model of a full-size cut-out of a continuous truss bridge deck The highlight of this application is

to use bidirectional displacement mode shapes instead of unidirectional displacement mode shapes as in previous applications However, reduced mode shapes are utilized to satisfy practical applicability in fault diagnosis Besides, unlike the previous model, the last one aims to optimize the initial learnable parameters such as connection weights and biases through an optimization process The idea of this model is that ANN with beneficial starting points will provide a better performance in damage detection compared with the conventional ANN The damage levels in the laboratory structure are quantified much closer to the actual extent using this approach

- Another effective and simple application of optimization algorithms is

to improve the capability in damage identification of a mode shape-based index, the so-called ECOMAC In this approach, the damage in structure

is identified by solving an inverse problem A laboratory steel beam is the main subject of an application example When the iterative procedure meets a stop condition, it terminates and returns a prediction result The findings confirm the feasibility and reliability of the proposed methods, namely the combination of the VBSDD and hybrid intelligence approach for bridge health monitoring

Beknopte samenvatting

Onderhoudskosten en levensduur van bestaande bruggen zijn zware uitdagingen voor brugbeheerders In operationele toestand kan overal schade van onbekende omvang ontstaan Dit kan leiden tot algeheel falen of tot verkorting van de levensduur Om dit te vermijden is vroege detectie van schade van het grootste belang Structurele gezondheidsmonitoring (SHM), een interdisciplinaire experimentele methodologie, is een effectieve methode voor de identificatie van structurele schade en conditie-beoordeling Een efficiënte SHM-strategie moet globale en lokale monitoringtechnieken combineren Lokale niet-destructieve testmethoden (NDT), bijvoorbeeld akoestische emissie (AE) ultrasone inspectie, geleide (lamb) golven (GW), visuele inspectie (een klassieke beoordeling ter plaatse) zijn alle gebruikelijke technieken voor schadekwantificering Zij kunnen echter alleen worden gebruikt om schade en bijbehorende omvang te evalueren indien de vroege fase van het falen, d.w.z de locatie van de schade, al is geïdentificeerd Op trillingen gebaseerde structurele schadedetectie (VBSDD) is een wereldwijde techniek die kan worden gebruikt om de structurele veiligheid en integriteit van

de gemonitorde constructies te beoordelen De aanwezigheid van schade wijzigt structurele parameters, d.w.z stijfheid, massa, flexibiliteit of energiedissipatie van de beschouwde structuren en resulteert vervolgens in veranderingen van frequentieresponsfuncties (FRF), modale parameters (d.w.z natuurlijke frequentie, modevorm en demping) VBSDD identificeert

de modale eigenschappen om te waarschuwen voor structurele schade Van deze kenmerken zijn natuurlijke frequenties gemakkelijker te identificeren met slechts één of enkele sensoren Niettemin zijn natuurlijke frequenties bijna ongevoelig voor veranderingen in lokale stijfheid Bovendien kan het effect van omgevingsfactoren, (bijvoorbeeld een temperatuurvariatie) groter zijn dan die van schade in constructies Daarentegen zijn verplaatsings-modevormen gevoeliger voor lokale schade en minder gevoelig voor temperatuurveranderingen Door voldoende (betaalbare) sensoren te gebruiken, kunnen de verkregen gedetailleerde modevormen worden gebruikt

om schade te detecteren en te lokaliseren Daarom worden op modevormen gebaseerde schadedetectie-indices gebruikt om schade in laboratorium-structuren en grootschalige bruggen te lokaliseren

Om de kwaliteit van schadekwantificering bij VBSDD toepassingen te verbeteren, wordt in dit proefschrift een hybride “rekenkundige intelligentie” methode ontwikkeld, bestaande uit optimalisatie-algoritmen (OA's) en/of machine learning (ML) Om de prestaties van machine learning-algoritmen voor schade-identificatie te verbeteren, worden globale zoekmogelijkheden