A Development Strategy for

Structural Health Monitoring

Applications1
Permanently installed structural health monitoring (SHM) systems are now a viable alter-
native to traditional periodic inspection (nondestructive testing (NDT)). However, their
industrial use is limited, and this article reviews the steps required in developing practical
SHM systems. The transducers used in SHM are fixed in location, whereas in NDT, they are
generally scanned. The aim is to reach similar performance with high temporal frequency,
Peter Cawley low spatial frequency SHM data to that achievable with conventional high spatial fre-

Downloaded from [Link] by guest on 25 May 2022

Imperial College London, quency, and low temporal frequency NDT inspections. It is shown that this can be done
South Kensington, via change tracking algorithms such as the generalized likelihood ratio (GLR), but this
London SW7 2BX, UK depends on the input data being normally distributed, which can only be achieved if
e-mail: [Link]@[Link] signal changes due to variations in the operating conditions are satisfactorily compensated;
there has been much recent progress on this topic, and this is reviewed. Since SHM systems
can generate large volumes of data, it is essential to convert the data to actionable infor-
mation, and this step must be addressed in the SHM system design. It is also essential to
validate the performance of installed SHM systems, and a methodology analogous to the
model-assisted probability of detection (POD) (MAPOD) scheme used in NDT has been
proposed. This uses measurements obtained from the SHM system installed on a typical
undamaged structure to capture signal changes due to environmental and other effects
and to superpose the signal due to damage growth obtained from finite element predictions.
There is a substantial research agenda to support the wider adoption of SHM, and this is
discussed in this study. [DOI: 10.1115/1.4051974]

Keywords: continuous and periodic condition assessment, diagnostic decision support,

on-line diagnostic approaches, ultrasonics

1 Introduction site by a qualified technician; inspection is typically carried out at

widely spaced time intervals, often at plant shutdowns. In contrast,
The replacement of periodic nondestructive testing (NDT) by
scanning is not generally practical in SHM, the transducers being
permanently installed structural health monitoring (SHM) systems
permanently fixed in position. It is not usually economically feasi-
has been discussed for many years, and there is a vast literature
ble to deploy an array of transducers at the scan pitch used in NDT,
on SHM [1]. However, the industrial take-up of SHM technology
so in SHM, the transducers are usually relatively sparsely deployed
has been slow, with very few widely deployed applications. This
over the structure, but frequent measurements can be taken. There-
contrasts with rotating machine condition monitoring [2] that is
fore, in NDT, measurements are taken at high spatial frequency and
very well established in many sectors after initial issues with false
low temporal frequency, whereas in SHM, they are taken at low
calls were overcome. Most machine condition monitoring applica-
spatial frequency and high temporal frequency.
tions use passive measurements of vibration or oil debris, whereas
Unless the probable damage locations are precisely defined and
monitoring of nonrotating structures usually requires active mea-
identified in advance, or the degradation will affect a large area
surements involving both transmitting and receiving transducers.
so a few sample points are sufficient to give a reliable estimate of
The two fields are therefore qualitatively different although they
its severity, then successful SHM requires an area monitoring capa-
are both often termed SHM [3].
bility; it is unlikely to be practical to cover the structure with point
The slow take-up of structural health monitoring technology is
sensors, so a method that gives significant area coverage per sensor
partly a result of organizational and business case issues [1], but
is needed. This makes the use of ultrasonic guided waves very
there are also significant technical problems that make deployment
attractive, and there is a huge literature on their use in SHM [1].
and reliable damage detection and quantification difficult and so
This article mainly focuses on example applications using guided
impact the business case. This article reviews the technical factors
waves, and the application of bulk ultrasonic waves to thick struc-
hindering wider substitution of NDT by SHM and proposes a
tures is also discussed. However, the methodology is equally appli-
system design process, together with a methodology for perfor-
cable to other techniques that may be attractive for some SHM
mance validation.
applications such as potential drop [4–8], current deflection [9],
Most NDT applications involve scanning a transducer or array
and magnetic measurements [10–12].
over the region of structure to be inspected. The scan can be
If the potential of SHM is to be realized widely in industry, a stra-
manual or automated, and the results are usually interpreted on
tegic approach to its development is required. This involves:

1
(1) Analysis of the business case and selection of applications
Paper presented at the ASME 47th Annual Review of Progress in QNDE, Virtual
Conference, August 25–26, 2020.
where the case is strong; development of an SHM system
Manuscript received October 19, 2020; final manuscript received December 23, requires considerable design, development, and validation
2020; published online August 27, 2021. Assoc. Editor: Paul Fromme. effort. This can be justified by a large volume of applications

Journal of Nondestructive Evaluation, Diagnostics NOVEMBER 2021, Vol. 4 / 041012-1

and Prognostics of Engineering Systems
Copyright © 2021 by ASME; reuse license CC-BY 4.0
 such as corrosion monitoring in the oil and gas industry2 or loading, as shown in Figs. 1(a) and 1(b). The corresponding prob-
very high value cases such as in nuclear plant [13]. abilities of cracks occurring at different locations in the two cases
(2) Careful analysis of the design requirements: what defects are shown in Figs. 1(c) and 1(d), respectively, these being
must be detected, under what conditions, with what con- obtained using the Weibull weakest link theory [17] as explained
straints; who will interpret the data; what decisions are to in Ref. [18], where the probabilities are quantified; here, the
be made from it; and what are the costs of incorrect concern is to illustrate the qualitative differences, so the scales
indications. are simply low high. Figure 2 shows two candidate monitoring
(3) Design of the transduction and instrumentation systems to methods, a fixed 25 mm diameter, 45 deg ultrasonic shear wave
survive the operating environment and have appropriate sta- probe in Fig. 2(a), and a potential drop system in Fig. 2(b).
bility, life (including battery if used), and connectivity. The probabilities of detecting a 6 mm × 6 mm vertical crack at dif-
(4) Design of the data processing system including compensa- ferent locations of the beam using the two techniques are shown
tion for environmental changes and conversion to usable in Figs. 2(c) and 2(d), respectively; these were estimated using
information for the recipient; consideration of the scale of finite element analysis with assumptions about the signal-to-noise
deployment is crucial here as a human operator can deal ratio taken from experiments, as described in Ref. [18]. Again,
with reviewing raw data from one or two systems, but auto- the results shown here are qualitative as the intention is to illus-
mation is increasingly essential when the number of deploy- trate the issues involved in technique selection. As would be

Downloaded from [Link] by guest on 25 May 2022

ments reaches 100 s and 1000 s. expected, the ultrasonic system gives very high sensitivity at
(5) Validation of performance: many applications with a strong the middle of the beam, but this drops away severely toward
business case will be on safety critical structures, so evidence the ends of the beam and also at the edges due to the beam
of the efficacy of the system must be provided to the structure being wider (32 mm) than the transducer (25 mm diameter). In
operator and regulators. contrast, the potential drop system gives relatively uniform cover-
age over the beam, but the peak sensitivity is lower than with the
This article concentrates on aforementioned points (2), (4), and
ultrasonic system. The sensitivity and probability of defect occur-
(5); Sec. 2 discusses SHM system design requirements and Sec. 3
rence at different locations maps of Figs. 1 and 2 can be com-
then reviews the possibility of exploiting the high temporal data fre-
bined to give overall probability of detection (POD) values as
quency provided by SHM to overcome the loss of defect sensitivity
presented in Table 1. (The values given are for a very low prob-
implied by its low spatial frequency. The use of receiver operating
ability of false alarm (PFA), but the relative performance is repli-
characteristic (ROC) curves as a performance metric and a proposed
cated at other PFA values.) This shows that the single-point
performance validation methodology are discussed in Sec. 4.
ultrasonic test cannot be used reliably with the four-point bend
Section 5 discusses the remaining research requirements and pre-
loading, but it performs little better than the potential drop
sents the conclusions of the paper.
system for the three-point loading case; this is because of the
reduction in performance at the edges of the beam shown in
Fig. 2(c). This demonstrates that unless the likely defect location
2 System Design Requirements is known very precisely, or it is acceptable to cover the structure
When considering implementation of an SHM system rather than with transducers, wide area and less-sensitive monitoring systems
a traditional periodic inspection approach, it is vital to consider the are likely to be preferable to the very sensitive, highly localized
design of the whole system not just, for example, the transduction. systems that are more often used in NDT. The difference is
Todd and Flynn [14] propose a structured design approach for SHM that in NDT, the transducer can be scanned so that the high sen-
systems, starting with key questions: sitivity is achieved at all scan locations. The question in SHM is
whether obtaining frequent readings in time from a low sensitiv-
(1) What are the failure modes the system is being designed to ity, wide area system can reproduce the performance obtained by
monitor, and, to whatever degree possible, what are their scanning at infrequent intervals in NDT; this is addressed in the
expected probabilities of occurrence? This is vital as not next section.
only it is imperative that the system detect all critical forms An idea of the probability of damage occurrence is important as it
of damage at all likely locations but also it is not overde- will be a key factor in determining an acceptable false call rate. If
signed and so unnecessarily expensive. damage is very unlikely to occur, then a very low false call rate
(2) What specific actions will the SHM system direct in response will be required to ensure that there is a reasonable probability
to the failure mode(s)? This forces consideration of the that a positive damage call correctly indicates that damage is
decision-making process and who will be authorizing present; this is a very well-known issue in medical screening and
actions, e.g., mandating the shutdown of a plant. is related to the base rate fallacy [19]. For example, suppose that
(3) What are the costs associated with the deployment and oper- the probability of false call (PFA) in an inspection is 10% and the
ation of the SHM system and the actions/decisions that the POD of a real defect is 90%, and further suppose that the a priori
SHM system directs. This highlights the need to identify probability of a callable defect being present is 5%. Then, in
the capital and revenue costs of both the SHM system 1000 inspections, we expect 95 false calls, 45 correct defect calls,
itself and the decision-making process. and 5 missed defects; therefore of 140 defect calls, ∼ 70% are
(4) What are the constraints present in the design space? For false. If the a priori probability of a callable defect being present
example, temperature range, intrinsic safety requirements, is reduced to 1%, then more than 90% of defect calls will be
sensor and instrumentation mass, feasibility of cabling false. It is essential that the actions to be taken in the event of a
versus wireless operation, and availability of power. defect call are clearly specified so that their cost is known, and so
Todd and Flynn [14] then go on to use a Bayesian approach to the an acceptable false call rate can be defined. If the cost of investigat-
design of the system, further details on an optimal sensor placement ing each call is high, then a very low false call rate will be required,
study being given in Refs. [15,16]. while maintaining the required POD; this puts severe demands on
The area of structure that must be monitored has significant the SHM system design.
implications for the cost of an SHM system and the choice of Much in-service NDT is performed during plant shutdowns, so
measurement technique. For example, consider simple 20 mm the temperature is close to ambient and the structures are not signif-
deep, 32 mm wide, and 150 mm long beams and their corre- icantly loaded; this means that standard transducers and instrumen-
sponding stress distributions in three-point and four-point tation can be used. In contrast, an SHM system is permanently
attached to the monitored structure so the transducers and instru-
mentation must survive the operational environment; it is some-
2
[Link] times decided only to take readings with the plant off load, but

041012-2 / Vol. 4, NOVEMBER 2021 Transactions of the ASME

 Downloaded from [Link] by guest on 25 May 2022
Fig. 1 Schematic (a) three-point and (b) four-point loading and stress distribution on a beam; corresponding probabilities of
defect occurrence: (c) three point and (d) four point (adapted from Ref. [18])

Fig. 2 Schematic (a) ultrasonic corner-echo and (b) potential drop monitoring systems on beams of Fig. 1; probabilities of detect-
ing defect if present at each location: (c) ultrasonic corner-echo and (d) potential drop monitoring (adapted from Ref. [18])

the system must still survive the full range of conditions, even if it Table 1 Expected probabilities of detection for 6 mm × 6 mm
only has to operate over a narrower range. Design of an SHM crack at 10−6 probability of false alarm in three-point and
system is therefore often much more demanding than in NDT. four-point loading with the monitoring systems of Fig. 2
For example, Fig. 3 shows a high-temperature ultrasonic thickness
monitoring system3 [20] in which the transducers and instrumenta- Expected POD ultrasonic Expected POD potential
sensor (%) drop sensor (%)
tion are outside the insulation over the high-temperature structure,
the ultrasonic waves being sent and received along stainless steel Four-point 32.7 90.5
strip waveguides, the resultant signals being transmitted to the bend
control room via a wireless link. Since the system was designed Three-point 92.5 90.6
to operate in zoned areas of oil and gas plant, it also had to meet bend
intrinsic safety requirements that, inter alia, limit the voltage and
current that can be used. Note: Adapted from Ref. [18].

3 Exploiting Frequent Data Farrar and Worden [3] identify two broad methodologies for pro-
3.1 Background. As discussed earlier, SHM will become cessing the data: the inverse problem and pattern recognition
much more attractive if wide coverage can be obtained from each approaches. In this context, the inverse problem approach usually
transducer position, but since wide area systems are typically less adopts a physics-based model of the structure and tries to relate
sensitive than the point inspection systems that are scanned in changes in the measurements to changes in the model; this is
NDT, they can be employed only if the lost sensitivity can be recov- strongly related to the more recently termed digital twin concept
ered via processing the frequent, time series data that can be [21,22] that in the civil engineering field has been named structural
obtained. identification [23,24]. It is very challenging to invert a large model
to match measured signal changes to the structural changes (in this
case damage growth) that produce them although there has been
3
See Note 2. significant progress in the full wave inversion of sonic and

Journal of Nondestructive Evaluation, Diagnostics NOVEMBER 2021, Vol. 4 / 041012-3

and Prognostics of Engineering Systems
 minimized usually is the rms residual when the two signals are sub-
tracted; therefore, they process the whole signal and the residual
tends to be dominated by changes in any large reflections that are
present. A blind trial of a guided wave pipe monitoring system
[41] on the pipe shown schematically in Fig. 4 reveals that it was pos-
sible to reliably detect defects about five times smaller than could be
detected by interpreting the signal obtained in a single test.4
The residual signal amplitude as a function of time at a location
where no defect was introduced when processed with the phase
compensation (PSC) method of Ref. [38] is shown in Fig. 5(a),
along with the temperature of the pipe (this result is slightly superior
to that obtained with BSS compensation). It can be seen that the
residual varies significantly and that this variation correlates with
temperature. To limit false calls, the defect “call” level has to be
set above the level of variations in the absence of damage, so
these variations limit the improvement in sensitivity that can be

Downloaded from [Link] by guest on 25 May 2022

obtained. Therefore, it was clear that improved temperature com-
pensation was the key to improving the defect sensitivity obtained
in monitoring.
Fig. 3 High-temperature ultrasonic thickness monitoring Mariani et al. hypothesized that the remnant temperature depen-
system [20] dence was due to interference between different guided wave modes
present in the signal, the interference being a function of tempera-
ture; this is due to the different modes having different velocity
ultrasonic signals, starting in the geophysics field [25]. The pattern coefficients of temperature and also due to the relative amplitudes
recognition approach is therefore more commonly followed in of the excited modes being a function of temperature. They devel-
machine condition monitoring and SHM although once change oped a new temperature compensation method that denoted
has been identified, located, and its magnitude estimated, it is location-specific temperature compensation (LSTC) [42]; unlike
then possible to investigate the size of the structural change that the BSS, OBS, and other previous methods that consider the
has produced it using damage models; this is discussed further in whole signal in a single process, LSTC treats each point on the
Sec. 4. signal, corresponding to a particular location on the test structure
In this context, pattern recognition becomes a change-point separately. The method comprises a calibration phase and a moni-
detection [26,27] or trend identification problem [28], and these toring operation phase. In calibration, a set of baseline signals is
have been widely studied in other fields such as machine condition acquired across the temperature range of interest, which is used to
monitoring [3,2], medical monitoring [29], and finance [30]. Many construct a calibration curve for each signal sample, i.e., each
rotating machines operate in protected environments and under point on the captured waveform. Each curve shows how the
fairly constant conditions that ease the problem [3]. In SHM, it is expected signal amplitude at each location in the absence of
very common for measurements to be taken under varying environ- damage varies with temperature. In the monitoring operation
mental conditions, and this must be accounted for. This was recog- phase, when a new measurement is acquired, at each point, the
nized in early attempts at bridge monitoring using natural frequency expected value at the relevant temperature obtained from the cali-
measurements where it was initially found that the measured natural bration curves is subtracted from the measurement itself. Thus, in
frequency increased with the introduction of damage, whereas it the absence of damage, the expected value of the residual signal
would be expected that damage would decrease the stiffness and is zero. Further details are given in the study by Mariani et al. [42].
so decrease the natural frequencies; this anomaly turned out to be Figure 5(b) shows the residual signals corresponding to those of
caused by temperature changes between the predamage and post- Fig. 5(a) when LSTC compensation is applied following initial PSC
damage measurements [3]. compensation. The temperature dependence of the residuals has
been removed, and Fig. 5(c) shows that the residual signal is now
normally distributed (the data comfortably passed the Lilliefors nor-
3.2 Temperature Compensation. This article concentrates on mality test [43] at a 5% significance level [42]). Figures 6(a) and
the use of ultrasonic-guided wave measurements for SHM, and 6(b) show the results corresponding to those of Figs. 5(a) and
there is a vast literature on signal compensation for environmental 5(b) at the location of defect 2 shown in Fig. 4. The solid curve
changes, particularly temperature. Two widely used methods for labeled defect 2 shows the history of defect introduction that
the temperature compensation of guided wave signals are baseline started after about 330 measurements and grew to a cross section
signal stretch (BSS) and optimal baseline selection (OBS). BSS loss of about 1.8% (note that the amplitude scales on Figs 5 and
requires only one baseline signal that is used as a reference to 6 are different). The residual signal of Fig. 6(a) again correlates
which any current measurement is compressed or stretched in with temperature, making reliable defect calling difficult until the
time to minimize the residuals [31,32], so compensating for velocity defect has grown to above about 1% cross section loss, whereas it
and dimension changes with temperature. In OBS, a set of baseline would be possible to call the defect reliably much earlier from the
signals is stored so that the one deemed most similar to the specific residual signal of Fig. 6(b) that remains low with zero mean
current measurement is used for amplitude subtraction, often after before the defect introduction, and then tracks its growth; the
also applying BSS on the selected baseline [31,33]. Both these tech- final residual signal level expressed as a percentage of the end
niques are in principle applicable to bulk wave and guided wave reflection before damage was grown is slightly above the percent-
signals. age cross section loss since the reflected signal is a function of
Unfortunately, the signal phase often changes with temperature the defect shape and the section loss [44].
with both piezoelectric and electromagnetic acoustic transducer The defect signal amplitude, and hence defect size, that can be
(EMAT) transducer systems [34–37], and Mariani et al. have reliably detected at a given false call rate is determined by the varia-
extended the BSS method to compensate for phase changes [38]. bility in the signal obtained from an undamaged structure. If the
There have also been several attempts to deal with the difficulty of signal from an undamaged structure is normally distributed, as
temperature compensation on multimode guided wave signals
[39,40]. All these methods seek to minimize the difference
4
between the current and baseline signals, and the function that is [Link]

041012-4 / Vol. 4, NOVEMBER 2021 Transactions of the ASME

Fig. 4 Geometry of the 8 in. schedule 40 test pipe used in blind
trial [41]. The 7 and 2 m pipe sections are connected by a 90 deg
elbow with the bend radius being 1.5 times the outer diameter of
the pipe. The defects introduced in the “forward” direction are
highlighted and numbered. The distances given of the defects
and weld in the forward direction are measured from the

Downloaded from [Link] by guest on 25 May 2022

middle of the sensor produced by Guided Ultrasonics Ltd.

shown in Fig. 5(b), then the probabilities of detection and false

alarm are a function of the ratio of the defect signal amplitude to
the standard deviation of the signal from an undamaged structure;
this is discussed in more detail in Ref. [45].

3.3 Data Analysis. In most conventional inspections, the

signals are analyzed on site by the inspection technician who inter-
prets the data and will often carry out any necessary follow-up
inspection to confirm a defect call or to size the indication more
accurately. Therefore, those responsible for the integrity manage-
ment of the structure are only alerted when there is an abnormality.
In contrast, SHM data are generated automatically, and it is typi-
cally much more frequent than NDT inspections (e.g., daily rather
than annual) and is transmitted direct to the structure operator. As
the number of monitoring locations on a structure or plant increases,
this data stream can become unmanageable unless some automatic
preprocessing is applied. When the thickness monitoring system5
[21] was first deployed at 100 s of locations on a plant, operators
described the experience of frequent, multipoint data as being like
“drinking from a hosepipe.” Goulet and Smith [46] note that with
increasing availability of communication systems and decreasing Fig. 5 Residual signal and temperature as a function of mea-
cost of sensors, more and more structures are being measured, but surement number (time) at 1.45 m location on pipe of Fig. 4,
where no defect was grown when residuals obtained using
our capacity to analyze large amounts of data is only marginally method of (a) Ref. [38] and (b) Ref. [38] followed by LSTC
increasing. method [42]. (c) Distribution of residual signals of (b) with
The Cambridge Centre for Smart Infrastructure and Construction normal curve fit shown.
(CSIC) stresses the importance of data handling and interpretation
and present a pyramid model [47], a simplified version of which
is shown in Fig. 7. Here, data are processed to a digestible form where rates are highest and also to see commonality between differ-
before being passed to the decision-making level, different ent locations that is also valuable information not readily seen from
degrees of automation being possible at each level. individual plots.
While the number of sensors on a plant remains modest, it may be
sufficient simply to compress the information displayed to the oper-
3.4 Automated Analysis and Change Detection. As data
ator in time and space. For example, the display used in the thick-
volumes increase, it is essential to increase the automation of pro-
ness monitoring system6 [21] was initially a single-point A-scan as
cessing, a reasonable target to keep the absolute volume of data
shown in Fig. 8(a) and then graduated to a thickness versus time
requiring manual processing constant with time. Therefore, the frac-
plot as shown in the uncompensated graph of Fig. 8(b). This
tion of data requiring manual intervention decreases in proportion to
showed sudden jumps in thickness that caused confusion for oper-
the number of sensors installed. Gradually increasing automation in
ators until it was understood that they were due to temperature
this way ensures that the development team doing the manual pro-
excursions changing the speed of sound; this led to a thermocouple
cessing sees a large number of cases with their associated difficul-
being incorporated in the transducer of Fig. 3, which enabled the
ties, so informing the development of automated processing. This
temperature compensated thickness versus time plot of Fig. 8(b)
also gives a significant database of cases that can be used for super-
to be produced. In turn, this enabled reliable corrosion rates to be
vised learning if machine learning algorithms are to be considered;
computed, and the different rates shown in Fig. 8(b) proved to cor-
this is discussed further below.
relate with feedstock changes in the plant. Figure 8(c) shows the
Since the a priori probability of the structure being defective is
next level of data compression with corrosion rates being displayed
usually low, the most effective and valuable way to reduce the
at all locations in the plant, enabling the operator to see at a glance
volume of data to be processed manually is to automatically identify
those cases where there is high confidence that no significant defect
5
See Note 2. is present. This usually corresponds to the cases with the cleanest
6
See Note 2. signals, so this is relatively simple to automate. In the thickness

Journal of Nondestructive Evaluation, Diagnostics NOVEMBER 2021, Vol. 4 / 041012-5

and Prognostics of Engineering Systems
 Downloaded from [Link] by guest on 25 May 2022
Fig. 6 Residual signal and temperature as a function of measurement number (time) at
location of defect 2 on pipe of Fig. 4 when residuals obtained using method of
(a) Ref. [38] and (b) Ref. [38] followed by the LSTC method [42]

monitoring example of Fig. 8, this might involve identifying those Figs. 5(a) and 6(a) makes the automated reliable calling of small
cases where there is high confidence that the corrosion rate is less defects very difficult. However, if the environmental effects are
than, for example, 0.1 mm/year. compensated satisfactorily, then automated processing is relatively
The presence of uncompensated environmental effects such as straightforward as shown by the corrosion rate calculations of
the excursions with temperature shown in the uncompensated thick- Fig. 8(b). Likewise, the normal distribution of the residual guided
ness plot of Fig. 8(b) or the residual guided wave signals of wave signals after LSTC compensation of Figs. 5(b) and 5(c)
makes it possible to apply standard change tracking algorithms
that have been developed in the statistical process control (SPC)
field.
Over the past century, researchers in SPC have produced a
number of methods for the quality control of manufacturing pro-
cesses that involve the analysis of monitored parameters (e.g., tem-
perature, pressure, humidity) that are typically assumed to follow
normal distributions [48,49]. Increasing computer power has
resulted in increased attention from SPC researchers to the general-
ized likelihood ratio (GLR) approach, which was first proposed by
Lorden in 1971 [50]. Although the GLR is rather computationally
intensive, it is particularly attractive as it does not require the
extent or time of change to be specified in advance. Mariani and
Cawley [45] tested it on guided wave pipe monitoring data of the
type shown in Figs. 5 and 6 and showed that it can enable the reli-
able detection of defects equivalent to as little as 0.1% cross section
loss in cases similar to those of Fig. 6 without a significant number
of false calls. This assumes that the sensor is stable with time and
that the residual signal in a given structure state remains normally
distributed. Nonstationarity of the data coming from an unchanged
system is a recognized problem in change-point detection that is still
the subject of current research [27]. Therefore, as signal processing
Fig. 7 Pyramid model of decreasing data volume and increas- is used to drive down the minimum detectable defect size, the
ing data value (adapted from Ref. [47]) demands on sensor stability with time increase so it is necessary

041012-6 / Vol. 4, NOVEMBER 2021 Transactions of the ASME

 Downloaded from [Link] by guest on 25 May 2022

Fig. 8 Increasing compression of data presentation: (a) single-point A-scan, thickness being inferred (automatically) from time
between successive peaks and speed of sound, (b) thickness versus time plots uncompensated and compensated for temperature
variations, and (c) corrosion rate at all sensor locations on three units on plant (darker color—higher wall loss rate). Sensors on unit
C and some on unit A installed later than on unit B.

Journal of Nondestructive Evaluation, Diagnostics NOVEMBER 2021, Vol. 4 / 041012-7

and Prognostics of Engineering Systems
to pay increased attention to the design of transducers and their corresponding to unity probability of detection and zero probability
attachment to the structure; it is also essential to check the statistics of false alarm, but of course this is not practically attainable. The
of the signals as part of the automated analysis process. ROC curve enables the following questions to be answered:
In some systems, it is possible to employ signal processing
methods that eliminate the effect of drift, one example being a • What is the POD at an allowable false call rate of 1% for a
bulk wave ultrasound test in which the amplitude of the echo from given damage size, type, and location and under a specified
the back face of the test piece is monitored. In this case, Mariani range of operating conditions?
et al. [51] used a method first proposed by Achenbach et al. [52] in • What is the smallest damage size that can be detected at 95%
which instead of using the amplitude of the back wall echo as the POD and 1% false call rate given a specified measurement fre-
measure of structural health, the ratio between successive back quency and range of operating conditions?
wall echoes is employed; the ratio is much less sensitive to drift.
However, this technique is only applicable in a limited number of For example, Fig. 9 shows ROC curves for a guided wave pipe
cases, so more research is needed on the problem of sensor drift. monitoring system using location-specific temperature compensa-
An alternative to the aforementioned traditional approaches is to tion (LSTC) for three different levels of cross section loss. They
employ supervised machine learning. There has been substantial are plotted on a semi-log scale, so that low levels of probability
research interest in applying supervised machine learning, usually of false alarm can be seen, recognizing that the PFA must be very
low for the system to be acceptable, as discussed earlier. The perfor-

Downloaded from [Link] by guest on 25 May 2022

in the form of multilayer perceptron (MLP) networks, to NDT
signals, see, e.g., [53]. Due to the limited hardware capabilities in mance improves dramatically as the defect size (cross section loss)
early work, shallow MLP networks consisting of a limited increases, being almost perfect at unit POD and < 0.01% PFA for
number of hidden layers were the preferred choice, and the 0.36% cross section loss. The dashed line corresponds to the perfor-
signals were typically preprocessed to reduce their dimensionality mance expected from a random guess strategy; this would be a
by extracting features deemed to be sensitive to defect reflections, straight, 45 deg line on a linear plot
see, e.g., Ref. [54]. However, this work has not found widespread Unfortunately, it is typically even more costly to define ROC
application in NDT, partly due to the need for skillful operators curves experimentally in SHM applications than it is to conduct
to manually determine defect-sensitive features effective for each POD trials in NDT. This is because the SHM system must be
specific application. installed on many structures of the required design that are cycled
It would be attractive to feed full time-domain waveforms to the through the full range of operational conditions, damage being
machine learning algorithm, hence using the network itself to learn introduced in some of the structures at a representative range of
the defect detection features, rather than relying on operator input; locations. In most cases, this is totally impractical and its cost
this requires more complex architectures able to model the com- would preclude the application of the SHM system. In NDT,
plexities of ultrasonic wave propagation, especially when dealing there is an increasing move toward model-assisted POD
with multimodal propagation of guided waves. (MAPOD) evaluation [58], and Liu et al. [59] have proposed a
In the last decade, with the advent of GPU-based machines, there related approach for ROC curve generation in SHM.
has been a proliferation of deep learning architectures involving the Modern computational resources mean that the signature pro-
use of multiple hidden layers. Mariani et al. [55] have reviewed duced by damage in bulk wave ultrasound, guided wave ultrasound,
the field and shown that in two guided wave monitoring examples, and other techniques can be reliably predicted, even when it has
the WaveNet convolutional neural network originally developed for complex shape [44]. In contrast, the reliable prediction of signal
audio signals [56] was able to learn features and/or patterns related changes due to environmental and other variability is not possible
to the presence of waves scattered from damage, thus eliminating so it is not feasible to predict the signal changes seen in the
the need for any feature input from human operators. The absence of damage. However, it is straightforward to obtain
network outperformed the optimal baseline selection and baseline
signal stretch compensation methods discussed earlier, and it was
especially encouraging that the improvements over the conven-
tional approach were particularly marked when the “current”
signals were taken at temperatures well outside the temperature
range available in the set of baseline signals. This suggests that
this class of network can complement or replace existing
methods, especially when testing occurs under new environmental
and operational conditions. This is also a possible approach for
dealing with the effects of sensor drift, but this requires much
more investigation.
The need for very large training data sets makes the application of
supervised machine learning in SHM difficult although it is possible
to supplement experimental data on damaged structures with data
from undamaged structures on which predicted damage signatures
are superposed; this is discussed further in the next section. A
further concern is whether the training data set covers a wide
enough range of cases, and it is also more difficult to qualify a
supervised machine learning system for use in safety critical appli-
cations than one based on predefined signal processing operations.

4 Performance Validation
Receiver operating characteristic curves [57] are routinely used in Fig. 9 Example ROC curves plotted on semi-log scale. Case of
guided wave pipe monitoring system using LSTC temperature
the evaluation of diagnostic tests in medicine and engineering; they compensation with defects corresponding to 0.12% cross
plot the sensitivity (POD in NDT/SHM terminology) for the test section loss (black – lowest solid curve), 0.24% loss (blue –
against its specificity (PFA in SHM) for a given change in condition middle solid curve), and 0.36% loss (red – top solid curve).
(defect size and type in SHM), both axes being on a (0,1) scale. The Data are used to compute Fig. 13(a) of Ref. [45]. Dashed line is
ideal operating point is at the top left-hand corner of the plot, random guess that is 45 deg line on a linear plot.

041012-8 / Vol. 4, NOVEMBER 2021 Transactions of the ASME

experimental data with environmental variation from an SHM aerospace sector, and many prototype systems have been developed
system installed on a typical undamaged structure. Therefore, mea- and flown on a trial basis; however, few systems are in routine use
sured data can be obtained over multiple environmental cycles on apart from a few local hotspot monitoring systems, particularly on
an undamaged structure, and predicted damage signatures can be military aircraft.
added at different locations with different growth patterns. The There is a substantial research agenda to support the wider adop-
effect of different degrees of environmental variation on the ROC tion of SHM, key topics being:
curves is then straightforward to simulate by appropriate selection
• Transducers and instrumentation to withstand harsh environ-
of signal sets, and the effect of varying damage severity, multiple
ments and remain stable
damage sites, frequency of readings, and so on is easy to assess.
• Calibration methods to account for any drift in sensitivity,
The approach has been validated on data obtained in the blind
phase, and so on with time
trial of a guided wave pipe monitoring system [41] by Heinlein
• Methods to increase area coverage per sensor at the required
et al. [60].
sensitivity and low probability of false call
Crucially, this approach can also be used to validate performance
• Further techniques for performance validation including both
of an installed SHM system. Suppose that the regulator requires
probability of defect detection and false call rate
assurance that the SHM system installed on a nuclear power plant
• Development of use cases including both the technology and
is correct in indicating that no defect larger than a specified size
business case

Downloaded from [Link] by guest on 25 May 2022

has grown at a particular location. Signals from defect growth at
• Efficient data handling to give operators information on which
the specified location can be predicted and added to the sequence
decisions should be taken, rather than raw data
of raw signals measured from the installed SHM system; the result-
• Fusion of data from multiple sources to provide better prog-
ing synthetic signal sequence can then be processed by the algo-
nostic information
rithms used to determine the health of the system, so verifying
whether the required defect would be detected if present. The statis-
tics of the measured signals can also be checked to ensure that any
assumptions about, for example, normality of residuals as discussed Acknowledgment
earlier are valid. The author thanks Dr. Stefano Mariani for multiple helpful
discussions, for his review of the paper prior to submission and
for generating the plot of Fig. 9. The author also thanks
5 Conclusions and Future Research Needs Dr. Joseph Corcoran for reviewing the paper and for several
helpful suggestions.
Permanently installed SHM systems are now a viable alternative
to traditional periodic inspection. The SHM system must have high
reliability over the envelope of operating conditions, a particular
concern being the avoidance of false calls due to, for example, tem- Conflict of Interest
perature or load changes. Since scanning a transducer over the There are no conflicts of interest.
structure, as is routinely done in NDT, is not possible with the
fixed transducers used in SHM, the data obtained in SHM typically
have a much lower spatial frequency than that in NDT, and it is
desirable to choose methods that enable high volume coverage References
from each transducer. On the other hand, it is possible to acquire [1] Cawley, P., 2018, “Structural Health Monitoring: Closing the Gap Between
Research and Industrial Deployment,” Struct. Health. Monit., 17(5), pp. 1225–
data much more frequently in SHM than in NDT; hence, the aim 1244.
is to achieve similar performance with low spatial frequency, high [2] Randall, R. B., 2011, Vibration-Based Condition Monitoring, Wiley, Chichester,
temporal frequency SHM data as is obtained in NDT with high UK.
spatial frequency, and low temporal frequency information. This [3] Farrar, C. R., and Worden, K., 2013, Structural Health Monitoring: A Machine
Learning Perspective, Wiley, UK.
can be done via change tracking algorithms such as the GLR, but [4] Farrell, D. M., Robbins, B. J., Sikka, P., and Seaman, M., 2004, “On-Line
this depends on the input data being normally distributed, which Monitoring and Control of Furnace Wall Corrosion in PF-Fired Boilers,”
can be achieved only if signal changes due to variation in the oper- Power Plant: Oper. [Link]. Issues, 3, pp. 16–21.
ating conditions are satisfactorily compensated. There has been [5] Farrell, D. M., Robbins, B. J., Stallings, J., Cardoso, S., and Bakker, W., 2008,
“Crack Growth Monitoring on Industrial Plant Using Established Electrical
much progress in this area, and the recently developed LSTC Resistance ‘Scanner’ Technology,” Insight, 50(12), pp. 690–694.
method [42] has shown excellent results on both ultrasonic-guided [6] Strommen, R. D., Horn, H., and Wold, K. R., 1993, “New Technique Monitors
wave and bulk wave data. There is also increasing interest in the Pipeline Corrosion, Cracking,” Oil Gas J., 91, pp. 88–93.
application of supervised deep learning methods to SHM signals. [7] Corbin, D., and Willson, E., 2007, “New Technology for Real-Time Corrosion
Detection,” Tri-Service Corrosion Conference, Denver, CO.
The transducers and instrumentation used in SHM must survive [8] Kuroiwa, T., Homada, Y., Ogawa, Y., Iida, H., Kuroki, M., and Hayakawa, M.,
the operating environment and remain stable over extended 2005, “Application of Potential Drop Technique to the Inspection of Welded
periods; this is much more demanding than in NDT where inspec- Boiler High Temperature and Pressure Parts,” 11th International Conference on
tion is generally carried out offline. Since SHM systems can gener- Fracture, Turin, Italy.
[9] Jarvis, R., Cawley, P., and Nagy, P. B., 2018, “Permanently Installed Corrosion
ate large volumes of data and often transmit it direct to the engineers Monitoring Using Magnetic Measurement of Current Deflection,” Struct. Health.
responsible for the plant, it is essential to convert the data to action- Monit., 17(2), pp. 227–239.
able information, and this step must be addressed in the SHM [10] Draper Inc., 2017, “A New Tool for Detecting Oil and Gas Pipeline Corrosion,”
system design. It is also essential to validate the performance of [Link]
corrosion, Accessed January 15, 2021.
installed SHM systems, and a methodology analogous to the [11] Yokagawa Inc., “Pervasive Corrosion Monitoring on Piping,” [Link]
scheme used in NDT has been proposed. This involves using mea- [Link]/uk/about/research-development/inv_center/pcm/, Accessed
surements obtained from the SHM system installed on a good struc- January 15, 2021.
ture to capture signal changes due to environmental and other [12] Li, Z., Jarvis, R., Nagy, P. B., Dixon, S., and Cawley, P., 2017, “Experimental and
Simulation Methods to Study the Magnetic Tomography Method (MTM) for Pipe
effects and to superpose the signal due to damage growth obtained Defect Detection,” NDT&E Int., 92, pp. 59–66.
from finite element or other predictions. [13] Plant Integrity Ltd, “Materials Testing and Inspection Services,” [Link]
The routine industrial use of SHM is in its infancy [1], with high [Link]/case-studies/teletest-guided-wave-technology-supports-uk-nuclear-
value oil and gas industry applications to the fore. There are many industry/, Accessed January 15, 2021.
[14] Todd, M. D., and Flynn, E. B., 2011, “A Bayesian Experimental Design
sensors deployed on civil engineering structures, but data process- Approach for Structural Health Monitoring,” Proc XIV International
ing to provide useful information to operators is far from standar- Symposium on Dynamic Problems of Mechanics (DINAME 2011), Sao
dized [46]. There has been a great deal of interest in SHM in the Sebastiao, Brazil.

Journal of Nondestructive Evaluation, Diagnostics NOVEMBER 2021, Vol. 4 / 041012-9

and Prognostics of Engineering Systems
[15] Todd, M. D., and Flynn, E. B., 2010, “A Bayesian Approach to Optimal Sensor Baseline Subtraction for Structural Health Monitoring,” Struct. Health. Monit.,
Placement for Structural Health Monitoring With Application to Active Sensing,” 19(1), pp. 26–47.
Mech. Syst. Signal Process, 24(4), pp. 891–903. [39] Fendzi, C., Rébillat, M., Mechbal, M., Guskov, M., and Coffigna, G.,
[16] Flynn, E. B., and Todd, M. D., 2010, “Optimal Placement of Piezoelectric 2016, “A Data-Driven Temperature Compensation Approach for Structural
Actuators and Sensors for Detecting Damage in Plate Structures,” J. Intell. Health Monitoring Using Lamb Waves,” Struct. Health. Monit., 15(5),
Mater. Syst. Struct., 21(3), pp. 265–274. pp. 525–540.
[17] Weibull, W., 1939, A Statistical Theory of the Strength of Materials, [40] Zoubi, A. B., Kim, S., Adams, D. O., and Mathews, V. J., 2019, “Lamb Wave
Generalstabens Litografiska Anstalts Förlag, Stockholm. Mode Decomposition Based on Cross-Wigner-Ville Distribution and Its
[18] Leung, M. S., 2020, “Probabilistic Approach to Structural Integrity Assessment of Application to Anomaly Imaging for Structural Health Monitoring,” IEEE
Fatigue Damage Using Permanently Installed Monitoring Systems,” Ph.D. thesis, Trans. Ultrason. Ferroelectr. Freq. Control, 66(5), pp. 984–997.
Imperial College, London. [41] Heinlein, S., Cawley, P., Vogt, T., and Burch, S., 2018, “Blind Trial Validation of
[19] Bar-Hillel, M., 1980, “The Base-Rate Fallacy in Probability Judgements,” Acta a Guided Wave Structural Health Monitoring System for Pipework,” Mater. Eval.,
Psychol., 44(3), pp. 211–233. 76, pp. 1118–1126.
[20] Cegla, F. B., Cawley, P., Allin, J., and Davies, J., 2011, “High-Temperature [42] Mariani, S., Heinlein, S., and Cawley, P., 2020, “Location Specific Temperature
(> 500 °C) Wall Thickness Monitoring Using Dry-Coupled Ultrasonic Compensation of Guided Wave Signals in Structural Health Monitoring,” IEEE
Waveguide Transducers,” IEEE Trans. Ultrason. Ferroelectr. Freq. Control, Trans. Ultrason. Ferroelectr. Freq. Control, 67(1), pp. 146–157.
58(1), pp. 156–167. [43] Lilliefors, H. W., 1967, “On the Kolmogorov–Smirnov Test for Normality With
[21] Tuegel, E. J., Ingraffea, A. R., Eason, T. G., and Spottswood, S. M., 2011, Mean and Variance Unknown,” J. Am. Stat. Assoc., 62(318), pp. 399–402.
“Reengineering Aircraft Structural Life Prediction Using a Digital Twin,” [44] Carandente, R., and Cawley, P., 2012, “The Effect of Complex Defect Profiles on
Int. J. Aerosp. Eng., 2011, pp. 1–4 . the Reflection of the Fundamental Torsional Mode in Pipes,” NDT&E Int., 46,
[22] Tao, F., Cheng, J., Qi, Q., Zhang, M., Zhang, H., and Sui, F., 2018, “Digital

Downloaded from [Link] by guest on 25 May 2022

pp. 41–47.
Twin-Driven Product Design, Manufacturing and Service With big Data,” [45] Mariani, S., and Cawley, P., 2020, “Change Detection Using the Generalized
Int. J. Adv. Manuf. Technol., 94(9–12), pp. 3563–3576. Likelihood Ratio Method to Improve the Sensitivity of Guided Wave Structural
[23] Catbas, F. N., and Kijewski-Correa, T., 2013, “Structural Identification of Health Monitoring Systems,” Struct. Health. Monit., p. 147592172098183 (in
Constructed Systems: Collective Effort Toward an Integrated Approach That press).
Reduces Barriers to Adoption,” ASCE J. Struct. Eng., 139(10), pp. 1648–1652. [46] Goulet, J. A., and Smith, I. F., 2013, “Predicting the Usefulness of Monitoring
[24] Moaveni, B., Hurlebaus, S., and Moon, F., 2013, “Special Issue on Real-World for Identifying the Behavior of Structures,” ASCE J. Struct. Eng., 139(3),
Applications of Structural Identification and Health Monitoring pp. 479–479.
Methodologies,” ASCE J. Struct. Eng., 139(10), pp. 1637–1638. [47] CSIC, “The Smart Infrastructure Paper,” [Link]
[25] Warner, M., Ratcliffe, A., Nangoo, T., Morgan, J., Umpleby, A., Shah, N., Vinje, [Link]/files/the-smart-infrastructure-paper, Accessed January 15, 2021.
V., Stekl, I., Guasch, L., Win, C., Conroy, G., and Bertrand, A., 2013, [48] Montgomery, D. C., 2012, Introduction to Statistical Quality Control, 7th ed.,
“Anisotropic 3D Full-Waveform Inversion,” Geophysics, 78(2), pp. R59–R80. Wiley, New York.
[26] Niu, Y. S., Hao, N., and Zhang, H., 2016, “Multiple Change-Point Detection: A [49] Basseville, M., and Nikiforov, I. V., 1993, Detection of Abrupt Changes: Theory
Selective Overview,” Stat. Sci., 31(4), pp. 611–623. and Application, Vol. 104, Prentice Hall, Englewodd Cliffs.
[27] Aminikhanghahi, S., and Cook, D. J., 2016, “A Survey of Methods for Time [50] Lorden, G., 1971, “Procedures for Reacting to a Change in Distribution,” Ann.
Series Change; Point Detection,” Knowl. Inf. Syst., 51, pp. 339–367. Math. Stat., 42(6), pp. 1897–1908.
[28] Maurya, M. R., Paritosh, P. K., Rengaswamy, R., and Venkatasubramanian, V., [51] Mariani, S., Liu, Y., and Cawley, P., 2020, “Improving Sensitivity and Coverage
2010, “A Framework for On-Line Trend Extraction and Fault Diagnosis,” Eng. of SHM Using Bulk Ultrasonic Waves,” Struct. Health. Monit.,
Appl. Artif. Intell., 23(6), pp. 950–960. p. 147592172096512. (in press).
[29] Staudacher, M., Telser, S., Amann, A., Hinterhuber, H., and Ritsch-Marte, M., 2005, [52] Achenbach, J., Komsky, I., Lee, Y. C., and Angel, Y. C., 1992, “Self-Calibrating
“A New Method for Change-Point Detection Developed for On-Line Analysis of the Ultrasonic Technique for Crack Depth Measurement,” J. Nondestruct. Eval.,
Heart Beat Variability During Sleep,” Phys. A, 349(3–4), pp. 582–596. 11(2), pp. 103–108 .
[30] Albert, A. A., Lopez, L. F., and Blas, N. G., 2019, “Multilinear Weighted [53] Legendre, S., Massicotte, D., Goyette, J., and Bose, T. K., 2001, “Neural
Regression (MWE) With Neural Networks for,” Appl. Soft Comput., 82, Classification of Lamb Wave Ultrasonic Weld Testing Signals Using Wavelet
p. 105555. Coefficients,” IEEE Trans. Instrum. Meas., 50(3), pp. 672–678.
[31] Lu, Y., and Michaels, J., 2005, “A Methodology for Structural Health Monitoring [54] Sbarufatti, C., Manson, G., and Worden, K., 2014, “A Numerically-Enhanced
With Diffuse Ultrasonic Waves,” Ultrason., 43(9), pp. 717–731. Machine Learning Approach to Damage Diagnosis Using a Lamb Wave
[32] Harley, J. B., and Moura, J. M., 2012, “Scale Transform Signal Processing for Sensing Network,” J. Sound Vib., 333(19), pp. 4499–4525.
Optimal Ultrasonic Temperature Compensation,” IEEE Trans. Ultrason. [55] Mariani, S., Rendu, Q., Urbani, M., and Sbarufatti, C., 2020, “Causal Dilated
Ferroelectr. Freq. Control, 59(10), pp. 2226–2236. Convolutional Neural Networks for Automatic Inspection of Ultrasonic Signals
[33] Konstantinidis, G., Wilcox, P. D., and Drinkwater, B. W., 2007, “An in Non-Destructive Evaluation and Structural Health Monitoring,” Mech. Syst.
Investigation Into the Temperature Stability of a Guided Wave Structural Signal Process.
Health Monitoring System Using Permanently Attached Sensors,” IEEE Sens. [56] van den Oord, A., Dieleman, S., Zen, H., Simonyan, K., Vinyals, O., Graves, A.,
J., 7(5), pp. 905–912. Kalchbrenner, N., Senior, A., and Kavukcuoglu, K., 2016, “WaveNet: A
[34] Clarke, T., Simonetti, F., and Cawley, P., 2010, “Guided Wave Health Monitoring Generative Model for Raw Audio,” [Link] Accessed
of Complex Structures by Sparse Array Systems: Influence of Temperature October 14, 2020.
Changes on Performance,” J. Sound Vib., 329(12), pp. 2306–2322. [57] Krzanowski, W. J., and Hands, D. J., 2009, ROC Curves for Continuous Data,
[35] Ha, S., Lonkar, K., Mittal, A., and Chang, F.-K., 2010, “Adhesive Layer Effects Chapman and Hall/CRC, New York.
on PZT-Induced Lamb Waves at Elevated Temperatures,” Struct. Health. Monit., [58] Rentala, V. K., Mylavarapu, P., and Gautam, J. P., 2018, “Issues in Estimating
9(3), pp. 247–256. Probability of Detection of NDT Techniques—A Model Assisted Approach,”
[36] Herdovics, B., and Cegla, F. B., 2018, “Structural Health Monitoring Using Ultrason., 87, pp. 59–70.
Torsional Guided Wave Electromagnetic Acoustic Transducers,” Struct. Health. [59] Liu, C., Dobson, J., and Cawley, P., 2017, “Efficient Generation of Receiver
Monit., 17(1), pp. 24–38. Operating Characteristics for the Evaluation of Damage Detection in Practical
[37] Herdovics, B., and Cegla, F. B., 2018, “Compensation of Phase Response Structural Health Monitoring Applications,” Proc. R. Soc. A, 473(2199),
Changes in Ultrasonic Transducers Caused by Temperature Variations,” Struct. p. 20160736.
Health. Monit., 18(2), pp. 508–523. [60] Heinlein, S., Cawley, P., and Vogt, T., 2019, “Validation of a Procedure for the
[38] Mariani, S., Heinlein, S., and Cawley, P., 2020, “Compensation for Evaluation of the Performance of an Installed Structural Health Monitoring
Temperature-Dependent Phase and Velocity of Guided Wave Signals in System,” Struct. Health. Monit., 18(5–6), pp. 1557–1568.

041012-10 / Vol. 4, NOVEMBER 2021 Transactions of the ASME