Structural condition identification for civil infrastructure: an appraisal

based on existing literature reviews

Ying Wang and Marios Chryssanthopoulos

Dept. of Civil & Environmental Engineering, University of Surrey

Guildford, Surrey, GU2 7XH, United Kingdom

+44-1483684090

ying.wang@surrey.ac.uk

Abstract

Infrastructure performance is of great importance for a nation’s economy and its

people’s quality of life. For efficient and effective infrastructure asset management,

structural health monitoring (SHM) has been researched extensively in the past 20-30

years. With an increasing number of SHM systems being installed, the interpretation of

the large volume of monitoring data, i.e. often manifested as condition identification,

becomes essential in asset integrity management. This paper provides an appraisal of

existing literature reviews on SHM, considering both reviews on different types of

structures and those focused on different approaches for data interpretation. It explores

the evolution of research interests in this field and identifies the need for an integrated

physics-based and data-driven structural condition identification approach.

1. Introduction

Civil infrastructure systems are proving more vulnerable than many might have

expected, particularly in the light of technological advances. Recently, two collapsed

bridges, in the United States (1) and Myanmar (2), respectively, have attracted engineers’

and researchers’ attention, not only because both led to casualties but also owing to the

failure occurring at an early stage of the design life. Especially, the former, a pedestrian

bridge in Miami, collapsed just five days after a large segment had been constructed and

placed on fixed supports. Although it was reported to the authority that cracks were

observed (3), the time-varying condition of the structure appears not to have been

assessed properly. Although the exact failure mechanism is still under investigation, it

can be argued that if a structural condition identification approach had been

implemented, the catastrophic consequences of the collapse could have been mitigated.

Indeed, for efficient and effective infrastructure asset management, Structural Health

Monitoring (SHM) has been researched extensively worldwide in the past 20-30 years

(4-6), using different sensors to monitor the performance and behaviour of structures. The

research advancement in this field has been seen tremendous. Not only has a large

amount of papers been published, but also many civil structures have been equipped

with SHM systems (4), including bridges, buildings, wind turbines, offshore structures,

nuclear plants, dams, tunnels, etc. Once sensors are installed, real-time monitoring data,

affected by operational, structural and environmental conditions, can be collected. They

can provide more detailed information regarding the actual condition of a structural

system compared to traditional inspections. Therefore, the interpretation of the large

2

volume of monitoring data becomes increasingly more important in asset integrity

management. This data interpretation process has been extensively referred to in

literature as ‘damage detection/identification’. Indeed, although performance can be

closely related to structural condition, the change in structural condition may not be

necessarily due to damage. For example, in the case of the Miami bridge, an important

change in structural condition took place when the constructed segment was transferred

from temporary supports to its final position supported by piers. Therefore, “structural

condition identification” is the term used in this paper to encompass a data

interpretation process that might lead to an undesirable event being diagnosed or

predicted even before evidence of damage becomes apparent.

Due to the large amount of existing publications in this field, it has become increasingly

difficult to perform a comprehensive literature review based on primary sources, such as

those performed in Refs (5, 6). Moreover, the literature review type papers/books

published after 2000 are becoming specialised and can be categorised, in the main, into

the following groups: 1) review of a particular kind of structure or structural member; 2)

review of a type of structural damage; 3) review of the application of a particular type of

sensor; 4) review of a family of damage/condition identification approaches.

This paper provides an appraisal of (more than thirty) literature reviews, first with a

focus on reviews of different types of structures. Then, a review and discussion on

different structural condition identification approaches will be presented.

2. Reviews of SHM technologies for different structural types

2.1 Bridges

For civil infrastructure, research attention focused initially on bridges. In 2005, Ko and

Ni (7) summarised the technology development of 20 bridges in China which were

instrumented with long-term monitoring systems. Their sensing and data acquisition

systems, environmental effects on the measured data, and the linkage with bridge

maintenance and management were discussed in detail. Stemming from the collapse of

the I-35W Bridge in 2007, a report on bridge health monitoring and inspections was

published in 2009 (8). The focus of this report was to explore existing sensing systems

and their practical applications. A total of 25 systems were discussed, showing their

general characteristics, advantages and disadvantages. Further, 38 monitoring

companies provided their feedback on the application of these systems. These reviews

focus primarily on the availability and potential for implementation of different sensing

systems, as well as the data acquisition challenges in an SHM system.

After 2010, more specialised reviews, targeting different types of damage or using

various sensing technologies, can be seen. For example, Nair and Cai (9) gave a review

of the acoustic emission (AE) method and its application to bridge health monitoring.

They suggested that a statistical quantitative analysis technique, intensity analysis, is

promising in interpreting AE monitoring data. Yi et al. (10) reviewed the research and

development activities in the field of bridge health monitoring using Global Positioning

System (GPS). They suggested the integration of GPS receivers with other sensors to

improve the accuracy and reliability of the SHM system and to develop data processing

3

software packages for filtering, smoothing and synchronizing. Indirect bridge

monitoring methods using instrumented vehicles were reviewed by Malekjafarian et al. (11). The theoretical background related to live load monitoring and the link between

monitoring and condition identification methods were discussed. The advantages and

disadvantages of both modal-based and non-modal-based methods were summarised.

The monitoring of a special type of damage, i.e. bridge scour, was reviewed by

Prendergast and Gavin (12). Different equipment types were discussed, with focus placed

on how to use changes in structural dynamic properties for scour monitoring.

Overall, bridges appear to be the most mature structural type for the installation of SHM

systems. They have the potential not only to provide data streams for algorithm

development and validation but to serve as “live labs” in which important interactions

between operation and environment can be studied under real conditions. Significant

research efforts have already been placed on how to interpret data from specific sensing

systems in order to understand structural condition or damage location and severity.

2.2 Wind turbines

More recently, increasing research attention has been focused on the monitoring of wind

turbines. Ten damage detection techniques, including acoustic emission (AE), thermal

imaging, ultrasonic methods, and vibration-based methods, were discussed in detail by

Ciang et al. (13). Their conclusions suggested the use of fibre optic sensors to monitor

global load conditions and the use of AE methods for early warning of the onset of

damage or for registering an impact event. Hameed et al. (14) further divided the

monitoring systems for wind turbines into condition monitoring and fault detection.

Different techniques and algorithms were highlighted, though not systematically

grouped. This broad categorisation has been followed by others (15-17). They showed that

research and development is still at the level of selecting the most appropriate sensing

systems. With regard to condition identification methods, in spite of their large number,

their functions are still basic and further development is needed. This finding was

echoed by Wymore et al. (18), who identified that most SHM research in this field

focuses on instrumentation issues, even though the current barrier for more significant

industry uptake is data interpretation. It is suggested that statistical pattern recognition is

a promising structural condition identification scheme, a tool originally proposed by

Sohn et al. (19) and comprehensively covered by Farrar and Worden (20). It has also been

used to structure a review on the monitoring of offshore wind turbines (21). Five

supervised learning methods and three unsupervised learning methods were reviewed,

with the former preferred due to their damage classification and quantification

capabilities.

Reviews in this field have become more specialised since 2015. For example, the

monitoring of wind turbine blades was reviewed by Li et al. (22). Four promising damage

detection methods were discussed, namely AE, wave propagation, impedance, and

vibration-based methods. The conclusion was that damage detection for turbine blades

is far from mature and there are still many challenging issues to be resolved. The wind

turbine bearing condition monitoring was reviewed recently (23). Available techniques

were listed, however there was no conclusion on how to select between alternatives. A

wind farm-level health management was also reviewed (24), with the suggestion that data

4

from multiple sensing systems need to be integrated, for which data-based methods can

be promising for diagnosis and prognosis. Among various data sources, the use of data

from the turbine supervisory control and data acquisition (SCADA) is appealing (25),

since it has the potential to be used in different applications together with data-driven

algorithms for anomaly detection, damage modelling, etc.

The number of literature reviews on wind turbines is growing rapidly. However, as

many have indicated, neither the monitoring nor condition identification technologies

are mature. Further investigation on how to interpret data from different sources and

how to define/achieve structural condition identification is the current research hotspot.

2.3 Other structures

There are only few reviews of SHM research on other types of structures. In 2000, a

review of vibration-based health monitoring for composite material/structures was

published (26). It identified the limitations of modal-based methods, i.e. the difficulty in

identifying minor damage and in differentiating the type of damage. A later review

suggested the use of nonlinear features and/or statistical parameters for this kind of

structures (27).

In the railway industry, the wayside SHM technologies were reviewed according to

different sensing systems and/or targeted structural components (28), though the

interpretation of data was not discussed. More recently, the emerging economies related

to renewable energy have received attention, as demonstrated in reviews of wind

turbines. For example, the inspection and SHM techniques for concentrated solar power

plants were reviewed by Papaelias et al. (29). It focused on alternative non-destructive

evaluation (NDE) techniques for the inspection of solar receivers and insulated pipes,

though data interpretation methods were, once again, not discussed. Similarly, a review

on field monitoring of offshore structures focused on the sensing technologies,

including the design of a monitoring system, but did not address condition identification

problems (30). A survey on the condition-based maintenance in the nuclear power

industry systematically looked at monitoring, diagnostics and prognostics issues in this

highly specialised industry (31). A hybrid method, i.e. physics-based and data-driven,

was recommended as a future direction.

Based on above appraisal of reviews, it can be concluded that: 1) SHM research efforts

in civil infrastructure have been shifting from bridges to wind turbines, and other

renewable energy structures/devices; 2) for emerging structural types, research focuses

first on sensing technologies before tackling the next step, namely data interpretation, in

turn leading towards structural condition identification. Therefore, future research

efforts should be concentrated in this field, especially on various possibilities for

structural condition identification methods that can interpret monitoring data from

typical sensing systems implemented in specific structural types.

3. Reviews of structural condition identification approaches

Due to the large number and variation of structural condition/damage identification

approaches, comprehensive insights can be found in relevant textbooks (e.g. Refs (20,

5

32)), whereas literature reviews are relatively limited, compared to other types of review

papers. Generally, these approaches can be broadly classified into two categories:

physics-based (or model-based in many references) and data-driven (or data-based).

Although some overlapping does exist, this distinction is helpful in enabling a succinct

discussion of alternative techniques/methods and in allowing a perspective to be

presented on an integrated approach.

3.1 Physics-based methods

3.1.1 Physics-based damage indicators

Due to their sound theoretical background, computational efficiency, and easy

implementation, physics-based damage indicators received significant attention as soon

as SHM was proposed in the early 1990s. For vibration-based SHM systems, the

majority of those indices/indicators/features are derived from modal parameters

obtained from modal testing results. Theoretically, natural frequencies of a structure are

positively related to its stiffness. Thus, changes of natural frequencies can indicate an

alteration in structural stiffness from a prior value, which in turn can be linked to a

structural health condition. When damage severity increases, natural frequencies usually

decrease, while the degree of the reduction is dependent on the position of the damage

relative to the mode shape for each particular vibration mode. Therefore, information on

natural frequencies of different modes can be used to derive/quantify damage existence,

location and severity (33).

Similarly, other damage indices based on mode shapes have also been used to indicate a

change in structural stiffness. Mode shape curvature (MSC), modal strain energy, and

other indicators derived from basic modal parameters have been found to be more

sensitive to damage than some other indicators. However, they are also sensitive to the

environmental noise (usually high frequency), which needs careful consideration (20). To

exploit information from both natural frequencies and mode shapes, flexibility-based

methods have been developed. In this group of methods, natural frequency information

is intrinsically added in the form of weighting factors to mode shapes. By doing this, the

information on lower modes plays a more important role than that on higher modes,

based on the observation that physical test results for the former are usually more

reliable than for the latter. A comparative study (34), together with a review of different

methods, showed that MSC is more robust under different scenarios, e.g. when MSC is

derived from displacement mode shapes. Li (35) reviewed different types of strain-based

damage indicators and indicated that the strain frequency response function index is

good at data detection and localisation, while the strain energy-based index performs

well with noisy signals.

Although many of the above methods were satisfactory for numerical cases and thus

popular in early studies, some practical issues limit their application to damage

identification of real structures. In particular, less controlable factors, such as

temperature, excitation methods and ambient noise, may significantly affect the test

results of modal parameters. Therefore, the damage indicator methods suffer from the

difficulty of discerning the changes caused by other factors from those caused by

6

structural damage. Further, the computational accuracy of damage indicators becomes

increasingly unreliable for complex structures.

3.1.2 Model updating

Due to the limitations of the above methods, additional research efforts were placed on

model updating methods (or inverse methods) (36) from the late 1990s onwards. This

type of approach requires the integration of numerical models and experimental results.

Simply speaking, model updating is to find the best match of numerically derived

parameters to those obtained from the monitoring data, through iteratively changing

physical parameters in the numerical model using an optimisation algorithm.

First, a numerical model needs to be constructed for a real structure. Secondly, based on

numerical simulations, damage-sensitive features can be defined and quantified, usually

based on modal parameters. Thirdly, monitoring or vibration tests of the real structure

needs to be performed. Fourthly, damage-sensitive features can be derived based on the

monitoring/test results. The results from numerical simulation are almost always

different from those from tests in terms of the attributes of the damage-sensitive

features. Hence, the discrepancies are minimised through iteratively changing particular

parameters of the numerical model using optimisation algorithms. Selection of

appropriate parameters is non-trivial in many real applications. Finally, when the

discrepancies are small enough, the numerical model is regarded to be able to simulate

the behaviours of the real structure, i.e. a digital twin.

(a) (b)

Figure 1. Scheme of model updating. a) Model updating process; b) Condition

identification using two-step model updating

When this process has been completed, the parameters in the numerical model can

reliably represent those of the real structure. From then on, if there is any damage to the

structure, the monitoring/test results will change and, in turn, so will the features. By

repeating the model updating process, the corresponding parameters in the numerical

model can be re-updated and the new results will be correlated with damage existence,

location, and severity. This process is illustrated in Figure 1, where (a) and (b) represent

the model updating process and condition indeitification process, respectively. In Figure

Numerical model

Optimisation

Structures

Vibration test/monitoring

Updated numerical model

Features Features

Difference <

threshold?

Yes

No

Numerical model

Calibration

Existing test data

New test data

Condition identification

Updating

7

1(b), the box with dashed line includes the process to calibrate the numerical model (the

first step), while the box with solid line shows the process of condition identification

(the second step). Both steps need the model updating process as shown in Figure 1(a).

These methods are theoretically sound and have the potential to analyze complex

structures. However, the numerical models, typically finite element models, normally

have many variables. When the number of variables is large, it is very hard to find the

optimum set of updating variables. Furthermore, this kind of inverse problems is usually

underdetermined, which leads to a number of possible solutions. To differentiate the

solutions may need experts’ judgment or multiple broadly consistent events which

might allow elimination of some possible solutions. Last but not least, as the structures

become more complex, the computational costs become higher, and often unacceptable

for real-time monitoring.

3.2 Data-driven methods

3.2.1 Data-driven damage indicators (and unsupervised learning)

Damage identification can be achieved through non-modal-based data-driven damage

indicators. Since 2000, this family of methods/techniques has been receiving increasing

attention. First, wavelet transforms were researched extensively. Compared with

methods based on the Fourier transform, any particular local feature of the time domain

signal can be analysed based on the scale and translation characteristics of wavelets. For

example, Peng and Chu (37) reviewed the application of wavelet analysis to time-

frequency analysis of signals, fault feature extraction, singularity detection, etc. A

recently published book (32) compared the performances of different transforms

inlcuding Fourier, Wavelet, and Hilbert-Huang transform.

Depending on whether there is a training process, machine learning methods can be

classified into either supervised or unsupervised learning. Unsupervised learning

methods are usually based on statistical parameters of the monitoring/test data, without

training. Therefore, this method is recommended for novelty detection only (20). For

example, Sohn et al (19) presented two statistical pattern recognition techniques based on

time-series analysis successfully applied to fibre-optic strain gauge data by

distinguishing data sets from different structural conditions.

Since data-driven damage indicators exploit information from time-domain data directly

and flexibly, they are usually more sensitive to damage and may be able to discern

damage from features related to other factors. However, these methods may not be able

to deal with complex structures, thus exhibiting a similar disadvantage as the physics-

based damage indicators.

3.2.2 Supervised learning

With the development of machine learning algorithms, researchers have also started to

exploit supervised learning methods for structural damage identification. The

fundamental research hypothesis is that 1) the monitoring data embody various patterns

under different structural conditions; and 2) given a particular monitoring data set, the

8

structural condition can be identified through pattern recognition. Specifically, the

problem is to find the pattern 𝑦𝑝 (output) for a new monitoring data set 𝐱𝑝 (input), given

a system with different conditions. In this framework, the system is a machine learning

model based on 𝑁 training data sets {(𝐱1, 𝑦1), … , (𝐱𝑁 , 𝑦𝑀)} with the monitoring time-

domain data 𝐱𝑖 and the corresponding pattern label 𝑦𝑖 ∈ {1, … , 𝑀}. Here 𝑀 is the total

number of existing patterns and 𝑀 ≤ 𝑁, since more than one dataset may be associated

with one pattern.

(a) (b)

Figure 2. Scheme of supervised learning. (a) training based on calibrated

numerical simulation; (b) training based on physical test data

These methods can achieve damage identification for complex problems such as a pipe-

soil system (38, 39). The most important challenge is the training, i.e. how to reliably

acquire data representing a nearly unlimited number of conditions for a structure. There

are two possible training routes, numerical simulation or physical tests (20). Figure 2

presents the condition identification process for both routes based on supervised

learning. Although training based on test data is more direct and can, in principle, better

represent actual conditions, tests are usually expensive, time consuming and can be

affected by noise and lack of repeatability. For civil infrastructure, it is even more

difficult because: 1) to encounter real environmental conditions is rare (for example,

hurricanes and earthquakes are not predictable); 2) usually a scaled model is involved

and unpredictable size effect may affect the results; 3) some damage processes evolve

over very long periods (e.g. corrosion). Thus, training based on numerical models and

simulation appears more promising for civil structures, aided in the first phase of

training by the creation of physical test results obtained in relatively controlled

laboratory environments involving scaled prototypes.

3.3 Perspective

Despite their popularity, physics-based methods face two main challenges: first, it is

often difficult to find a feature that is sensitive to structural conditions while insensitive

to environmental noise. Second, model updating methods suffer from relatively low

computational efficiency due to their reliance on complex simulation models. To

address these challenges, efforts have switched to data-driven approaches in recent

years, where condition identification can be achieved through machine learning and

Training

Machine learning algorithm

Numerical model

Calibration

Simulation

Data “pool”

Existing test data

Condition identification

New test data

Training

Machine learning algorithm

Existing test data

Condition identification

New test data

9

artificial intelligence algorithms. Using such methods, the features can be generated

automatically and may achieve better structural condition identification results than

traditional methods, while computational costs can be significantly reduced. However,

existing data-driven methods face a main challenge, i.e. they often lack the complexity

embedded in numerous and diverse scenarios in real structures, considering different

possible conditions, environmental factors, and loading histories. Therefore, an

integrated approach using both physics-based and data-driven methods should, in

principle, be explored for structural condition identification to exploit the merits of the

two currently pursued approaches, while compensating their shortcomings. One

integration aspect is to use the results from physics-based methods as the training set for

the data-driven algorithm, as demonstrated in recent studies (38, 39). More innovative and

penetrating integration methods/techniques are needed to enhance the cost-effectiveness

of the training process. The fast pace of development in machine learning and artificial

intelligence areas in other sectors should pave the way for this to happen, though civil

infrastructure presents specific challenges, as indicated in the previous sections, for

which bespoke solutions will be needed.

It should be noted that many above methods have only been verified by numerical

examples and/or experimental studies in the laboratory. For the on-site monitoring of

structures, practical issues exist and there is still work in order to establish the

framework that will deliver consensus on how to select the most suitable structural

condition identification approach in the light of cost and reliability constraints.

4. Conclusions

In this paper, a succinct appraisal of literature review type papers/books in the field of

SHM of civil infrastructure is presented. First, focus is placed on the reviews of

different types of structures. Based on this, a transition of research interests from

bridges to renewable energy infrastructure can be clearly identified. More importantly,

the evolutionary nature of research objectives can be traced to reveal a similar sequence:

initially sensing technologies are trialled/evaluated, then data acquisition and basic

processing from different types of sensors is addressed, and lastly condition

identification for different damage types or different structural

component/assembly/system is undertaken. The paper then goes on to review different

structural condition identification methods, grouped as physics-based and data-driven.

Since both have their own advantages and disadvantages, innovative integration

methods using advanced data analytics algorithms are identified as one of the most

promising future research directions in this field.

Acknowledgements

The review work described in this paper was undertaken in the context of the EPSRC

First Grant: EP/R021090/1, “An integrated physics-based and data-driven approach to

structural condition identification”.

References

1. https://edition.cnn.com/2018/03/15/us/miami-bridge-collapse/index.html

10

2. http://www.elevenmyanmar.com/local/13668

3. http://abcnews.go.com/US/engineer-reported-cracks-florida-bridge-days-collapse-

officials/story?id=53811847

4. Brownjohn JM. (2007) Structural health monitoring of civil infrastructure.

Philosophical Transactions of the Royal Society of London A: Mathematical,

Physical and Engineering Sciences. 365(1851):589-622.

5. Doebling SW, Farrar CR, Prime MB and Shevitz DW, (1996). Damage

identification and health monitoring of structural and mechanical systems from

changes in their vibration characteristics: a literature review. Los Alamos

National Laboratory report LA-13070-MS

6. Sohn H, Farrar CR, Hemez FM, Shunk DD, Stinemates DW, Nadler BR and

Czarnecki JJ, (2004) A review of structural health monitoring literature: 1996-

2001. Los Alamos National Laboratory report LA-13976-MS.

7. Ko JM and Ni YQ, (2005). Technology developments in structural health

monitoring of large-scale bridges. Engineering Structures, 27(12), 1715-1725.

8. Gastineau A, Johnson T and Schultz A, (2009). Bridge Health Monitoring and

Inspections - A Survey of Methods. Minnesota Department of Transportation,

MN/RC 2009-29

9. Nair A and Cai CS, (2010). Acoustic emission monitoring of bridges: Review and

case studies. Engineering Structures, 32(6), pp.1704-1714.

10. Yi T, Li H and Gu M, (2010). Recent research and applications of GPS based

technology for bridge health monitoring. Science China Technological Sciences,

53(10), 2597-2610.

11. Malekjafarian A, McGetrick PJ and OBrien EJ, (2015) A review of indirect bridge

monitoring using passing vehicles. Shock and vibration, 2015.

12. Prendergast LJ and Gavin K, (2014). A review of bridge scour monitoring

techniques. Journal of Rock Mechanics and Geotechnical Engineering, 6(2), 138-

149.

13. Ciang CC, Lee JR and Bang HJ, (2008). Structural health monitoring for a wind

turbine system: a review of damage detection methods. Measurement Science and

Technology, 19(12), p.122001.

14. Hameed Z, Hong YS, Cho YM, Ahn SH and Song CK, (2009). Condition

monitoring and fault detection of wind turbines and related algorithms: A review.

Renewable and Sustainable Energy Reviews, 13(1), 1-39.

15. Liu WY, Tang BP, Han JG, Lu XN, Hu NN and He ZZ, (2015). The structure

healthy condition monitoring and fault diagnosis methods in wind turbines: A

review. Renewable and Sustainable Energy Reviews, 44, 466-472.

16. Qiao W and Lu D, (2015). A survey on wind turbine condition monitoring and

fault diagnosis - Part I: Components and subsystems. IEEE Transactions on

Industrial Electronics, 62(10), 6536-6545.

17. Qiao W and Lu D, (2015). A survey on wind turbine condition monitoring and

fault diagnosis - Part II: Signals and signal processing methods. IEEE

Transactions on Industrial Electronics, 62(10), 6546-6557.

18. Wymore ML, Van Dam JE, Ceylan H and Qiao D, (2015). A survey of health

monitoring systems for wind turbines. Renewable and Sustainable Energy

Reviews, 52, 976-990.

11

19. Sohn H, Farrar CR, Hunter NF and Worden K, (2001). Structural health

monitoring using statistical pattern recognition techniques. Journal of dynamic

systems, measurement, and control, 123(4), 706-711.

20. Farrar CR and Worden K (2012) Structural health monitoring: a machine

learning perspective. John Wiley & Sons.

21. Martinez-Luengo M, Kolios A and Wang L, (2016). Structural health monitoring

of offshore wind turbines: A review through the Statistical Pattern Recognition

Paradigm. Renewable and Sustainable Energy Reviews, 64, 91-105.

22. Li D, Ho SCM, Song G, Ren L and Li H, (2015). A review of damage detection

methods for wind turbine blades. Smart Materials and Structures, 24(3),

p.033001.

23. de Azevedo HDM, Araújo AM and Bouchonneau N, (2016). A review of wind

turbine bearing condition monitoring: State of the art and challenges. Renewable

and Sustainable Energy Reviews, 56, 368-379.

24. Kandukuri ST, Klausen A, Karimi HR and Robbersmyr KG, (2016). A review of

diagnostics and prognostics of low-speed machinery towards wind turbine farm-

level health management. Renewable and Sustainable Energy Reviews, 53, 697-

708.

25. Tautz-Weinert J and Watson SJ, (2016). Using SCADA data for wind turbine

condition monitoring–a review. IET Renewable Power Generation, 11(4), 382-

394.

26. Zou Y, Tong L and Steven GP, (2000). Vibration-based model-dependent damage

(delamination) identification and health monitoring for composite structures - a

review. Journal of Sound and vibration, 230(2), 357-378.

27. Montalvao D, Maia NMM and Ribeiro AMR, (2006). A review of vibration-based

structural health monitoring with special emphasis on composite materials. Shock

and Vibration Digest, 38(4), 295-324.

28. Barke D and Chiu WK, (2005). Structural health monitoring in the railway

industry: a review. Structural Health Monitoring, 4(1), 81-93.

29. Papaelias M, Cheng L, Kogia M, Mohimi A, Kappatos V, Selcuk C, Constantinou

L, Muñoz CQG, Marquez FPG and Gan TH, (2016). Inspection and structural

health monitoring techniques for concentrated solar power plants. Renewable

Energy, 85, 1178-1191.

30. Wang P, Tian X, Peng T and Luo Y, (2018). A review of the state-of-the-art

developments in the field monitoring of offshore structures. Ocean Engineering,

147, 148-164.

31. Ayo-Imoru RM and Cilliers AC, (2018). A survey of the state of condition-based

maintenance (CBM) in the nuclear power industry. Annals of Nuclear Energy,

112, 177-188.

32. Chen HP, (2018). Structural Health Monitoring of Large Civil Engineering

Structures. John Wiley & Sons.

33. Salawu OS (1997) Detection of structural damage through changes in frequency: a

review. Engineering Structures, 19(9), pp.718-723.

34. Fan W and Qiao P, (2011). Vibration-based damage identification methods: a

review and comparative study. Structural Health Monitoring, 10(1), 83-111.

35. Li YY, (2010). Hypersensitivity of strain-based indicators for structural damage

identification: A review. Mechanical Systems and Signal Processing, 24(3), 653-

664.

12

36. Friswell MI (2007) Damage identification using inverse methods. Philosophical

Transactions of the Royal Society of London A: Mathematical, Physical and

Engineering Sciences, 365(1851), 393-410.

37. Peng ZK and Chu FL, (2004). Application of the wavelet transform in machine

condition monitoring and fault diagnostics: a review with bibliography.

Mechanical Systems and Signal Processing, 18(2), 199-221.

38. Wang Y and Hao H, (2013). Damage identification scheme based on compressive

sensing. ASCE Journal of Computing in Civil Engineering, 29(2), p.04014037.

39. Wang Y (2015) An innovative approach to structural damage identification via

dictionary learning technologies. 7th International Conference on Structural

Health Monitoring of Intelligent Infrastructure, Turin, Italy