See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/272174647

Finite Element Modelling of Cracks as Acoustic Emission Sources

Article in Journal of Nondestructive Evaluation · February 2015

DOI: 10.1007/s10921-015-0278-8

CITATIONS READS

75 1,469

2 authors:

Markus Günter Ronny Sause Stefan Richler

Universität Augsburg Universität Augsburg
191 PUBLICATIONS 2,866 CITATIONS 3 PUBLICATIONS 178 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Stefan Richler on 13 February 2015.

The user has requested enhancement of the downloaded file.

J Nondestruct Eval (2015) 34:4
DOI 10.1007/s10921-015-0278-8

Finite Element Modelling of Cracks as Acoustic Emission Sources

Markus G. R. Sause · Stefan Richler

Received: 21 August 2014 / Accepted: 16 January 2015

© The Author(s) 2015. This article is published with open access at Springerlink.com

Abstract This paper presents results for a new acoustic 1 Introduction

emission crack source model based on a finite element mod-
elling approach which calculates the dynamic displacement The formation and propagation of cracks in solid media is a
field during crack formation. The specimen modelled is stat- field of research that has been active for decades. Still, the
ically loaded until conditions for crack growth as defined by theoretical description of the physics at the crack tip and the
a failure criterion are fulfilled. Subsequently, crack growth is crack dynamics are active fields of research [1–4]. A phe-
modelled by local degradation of the material stiffness uti- nomenon that is closely related to the crack dynamics is the
lizing a cohesive zone element approach. The displacements generation of acoustic waves due to the crack motion due
due to crack growth generate the acoustic emission signal and to the release of stored elastic energy. These acoustic waves
allow detailed examination of the principles of acoustic emis- propagate within the solid and can be detected at the sur-
sion sources operation. Subsequent to crack growth signal face by suitable sensor systems. This method is known as
propagation is modeled. The signal propagation is modeled acoustic emission analysis and has already proven its signifi-
superimposed on the static displacement field. The presented cance for structural health monitoring as well as its ability to
model comprises a multi-scale and multi-physics approach improve material testing procedures [5]. Despite of the broad
to consider the signal propagation from source to sensor, the range of technical applications, only a small amount of work
piezoelectric conversion of the elastic wave to an electric has been performed recently to advance the understanding of
signal and the interaction to the acquisition electronics. Vali- the physical processes involved in the generation of acoustic
dation of the modeling approach is done by investigating the emission.
acoustic emission signals of micromechanical experiments. In order to interpret the detected acoustic emission signals
Using a specifically developed load stage, carbon fiber fila- in terms of their relevance to material failure it is required
ment failure and matrix cracking can be prepared as model to have concise knowledge of the underlying physics. The
sources. A comparison of the experimental signals to the whole process of the acoustic emission technique can be cat-
modeled signals shows good quantitative agreement in sig- egorized into three subsequent parts. The first part comprises
nal amplitude and frequency content. A comparison between the acoustic emission source, the second part considers the
the present modeling work and analytical theories demon- acoustic emission signal propagation from source to sensor
strates the substantial differences not considered in previous and the third part consists of the acoustic emission signal
modeling work of acoustic emission sources. detection.
In the past, various valuable attempts have been made
Keywords Finite element modeling · Acoustic emission · to provide a theoretical description of acoustic emission
Fracture mechanics · Cohesive zone elements sources. The source model concept used in most of the ana-
lytical approaches was derived from seismology and is most
of the time based on the work of Aki and Richards [6].
M. G. R. Sause (B) · S. Richler
Here, source models are geometrically approximated as point
Institute for Physics, Experimental Physics II, University of Augsburg,
86135 Augsburg, Germany sources, while the dynamic of the source is either approxi-
e-mail: [email protected] mated from iterative refinement of model parameters to fit

123
4 Page 2 of 13 J Nondestruct Eval (2015) 34:4

(a) (b) (c) +F

+F +F

-F -F

z
static geometry static geometry change of geometry
y
x -F during crack growth

Fig. 1 Different types of acoustic emission source model descriptions used in literature employing point sources (a) or extended sources (b) in
conjunction with analytic source functions. New source model description presented herein using dynamic changes of the source geometry based
on fracture mechanics (c)

experimental data or is based on assumptions on the source attempts that have been made to incorporate more accurate
dynamics derived from structural mechanics. Various step- source geometries, while the modeled crack dynamics are
function descriptions exist, which are used to describe the still based on analytical source functions (cf. Fig. 1b). The
3-dimensional spatial displacement of the crack surface dur- third type uses accurate artificial source geometries and does
ing crack formation [7–11]. In particular, the rise-time of the not need an analytical source function to generate acoustic
initial crack surface displacement is an essential parameter emission. Instead, this type of source model is capable to gen-
to model the crack surface motion [12]. However, there are erate the crack dynamics based on experimentally accessible
no reports in literature of successful measurements of rise- parameters and fracture mechanics laws.
times of real acoustic emission sources, e.g. due to crack Currently all source models proposed in literature are of
formation in materials. Instead the rise-time is typically esti- type one or type two, since they all require the definition
mated based on the elastic properties of the bulk material. of an explicit source function. Therefore, no details of the
This type of source modeling has been successfully applied dynamics arising from the crack formation process and the
to many cases, and the basic concept has been used within the subsequent crack surface motion are predicted or considered
generalized theory of acoustic emission by Ono and Ohtsu by those models.
[8,13], the work of Scruby [14] and numerous other analyt- From a mathematical modeling and simulation point of
ical descriptions [7,9,15,16]. view, there are two main challenges in providing a numeri-
In recent years it has become convenient to use numerical cally based acoustic emission source model of the third type.
methods to model acoustic emission sources. In this field, The first challenge consists of the different scales involved
Prosser, Hamstad and Gary applied finite element modeling in the problem (crack length of the order of microns versus
to simulate acoustic emission sources based on body forces signal wavelength of the order of millimeters to centimeters)
acting as a point source in a solid [10,17]. Hora and Cer- and the proper scale bridging. Owing to the vastly different
vena investigated the difference between nodal sources, line observations scales, a full multi-scale approach is thus nec-
sources and cylindrical sources to build geometrically more essary. The second challenge stems from the calculation of
representative acoustic emission sources [18]. At the same temporal and spatial evolution of the surfaces of the crack.
time, we proposed a finite element approach using an acoustic This is a level of detail that is typically not studied in mod-
emission source model taking into account the geometry of a eling approaches used to describe crack formation by means
crack and the inhomogeneous elastic properties in the vicin- of cohesive zone elements, extended finite element methods
ity of the acoustic emission source [11]. or similar implementations.
Based on these investigations we can categorize the dif- In contrast to the source model, the theoretical descrip-
ferent modelling strategies to describe acoustic emission tion and numerical implementation of wave propagation is
sources of crack propagation as shown in Fig. 1. The first already well established [10,11,16,17,19,20,27]. However,
type of source models considers point-like sources explic- it is important to consider the effects of attenuation, disper-
itly defining the source dynamics utilizing analytical source sion and propagation in guiding media to accurately cap-
functions (cf. Fig. 1a). As second type we can interpret those ture the characteristics of the signal (e.g. frequency content).

123
J Nondestruct Eval (2015) 34:4 Page 3 of 13 4

While analytical descriptions benefit from the low compu-

tational intensity [16] to describe wave propagation, for the 120
mm
numerical methods a main focus is the improvement of the force F
calculation routines.
Another challenge to obtain acoustic emission signals to
39 mm
compare relative to experimental data is the description of
the detection process. Here the detection process using piezo-
electric sensors can have significant impact on the bandwidth tensile bar
of the detected signals and their relative frequency content.
aluminum pin
While some analytical approaches consider the sensors trans-
z piezoelectric elements
fer function explicitly [16,21], some attempts have also been y
x WD sensor
made to model the response of piezoelectric sensors them- fixed constraint
selves [22]. We recently proposed a finite element approach
Fig. 2 3D-model of the experimental setup including cross-sectional
to directly include the piezoelectric sensor in a modeling
view of acoustic emission sensor used
environment to also account for the interplay between the
sensor and the material it is attached on [23].
In the present work, we first present the source model (see Fig. 3). Subsequently, curing of the liquid resin is carried
description and establish its principle of operation. Subse- out using heating foils attached to the aluminum block and
quently we validate the source description by modeling of an additional heating sleeve wrapped around a small cylinder
acoustic emission sources to correspond to micromechanical covering the aluminum pin and parts of the tensile bar. We use
experiments. The experimental work is based on a micro- a curing cycle comprising a heating rate of 2 ◦ C/min up to the
mechanics test stage, which was constructed to allow the curing temperature of 180 ◦ C. The curing temperature is kept
preparation of fiber breakage and matrix cracking as acoustic constant for 150 minutes with subsequent cooling to room
emission sources. We then apply the new acoustic emission temperature at a rate between 0.5 and 2 ◦ C/min. Due to its
source model concept in combination with in-situ modeling low thermal conductivity the tensile bar made from PEEK
of the signal propagation process and the detection process minimizes the dissipative heat flux and therefore assures a
of a piezoelectric sensor. Comparison is made between the constant temperature of the resin during curing. The tensile
experimental results and the results from the different types bar and the aluminum block are mounted in a universal test
of acoustic emission source simulation. machine so that thermal expansion of the components and
chemical shrinkage of the resin can be compensated by a
closed loop force control. The test machine control adjusts
2 Experimental Setup the tensile bar position to assure zero force acting during
curing, which is necessary to avoid excessive forces acting
In the following, we present the experimental setup and the on the filament causing preliminary failure due to thermal
details of signal analysis as applied to the detected signals. expansion and cure shrinkage of the resin.
As seen in Fig. 2, the core of the micromechanics test stage To prepare fiber breakage, we use a two-component epoxy
consists of an aluminum block with dimensions of 120 mm to bond a HTA carbon fiber to the end of a flat-topped tensile
× 120 mm × 39 mm (length × width × height). At the bar made from PEEK. The fiber is then moved into the resin
center of the top side of the aluminum block a pin with 2 mm droplet using a micrometer stage and an optical microscope.
diameter and 1 mm height was machined out of the block. The free fiber length was chosen to be between 350 μm and
At the opposite side of the block an acoustic emission sensor 450 μm. The embed length was chosen larger than 100 μm
is positioned at the center. In order to ensure reproducible to reach fiber breakage before fiber pull-out occurs. After
mounting conditions, we use a spring system to firmly press embedding, the resin droplet is cured in-situ.
the sensor to the aluminum block. After preparation of the test geometry, the universal test
For all cases investigated the first part of the preparation machine is used to apply a tensile force using a displacement
is to position a small droplet of RTM6 epoxy resin on top of controlled mode with velocities dependent on the selected
the aluminum pin. To facilitate the positioning of the epoxy failure mechanism. We choose a velocity of 20 μm/min for
resin droplet a small depression was machined into the end of fiber breakage and 50 μm/min for matrix cracking. The fail-
the aluminum pin. To prepare the micromechanical stage for ure was monitored by an optical microscope using a mag-
generation of a matrix crack a tensile bar made from polyether nification factor of 100. The images obtained after failure
ether ketone (PEEK) of 2 mm diameter and 80 mm length is are shown in Fig. 3 for the two failure types. The respective
used. The tensile bar is first moved into the liquid epoxy resin acoustic emission signal is detected by a type WD piezoelec-
and then retracted to yield a tapered contour of 380 to 800 μm tric sensor coupled by temperature stable Apiezon-L grease

123
4 Page 4 of 13 J Nondestruct Eval (2015) 34:4

Fig. 3 Microscopy images of the failure mechanisms investigated: matrix cracking (left) and fiber breakage (right)

at the bottom of the aluminum block. The dimensions of the concentration at the tip of the notch, which will cause crack
aluminum block allow an observation window of the primary initiation at this point.
acoustic emission signal of 18 μs free of reflections from In the second step, the initial conditions for the displace-
the surfaces of the aluminum block. The detected acoustic ment u and stress states σ are chosen to be identical to the
emission signals are digitized by 40 MS/s using a bandpass static values ustatic and σstatic as calculated in the previ-
range between 20 kHz and 1 MHz. Triggering of the signals ous step. Boundary conditions for restricted displacement
was carried out with 10 μs Peak-Definition-Time, 80 μs Hit- components and external loads are kept identical to the pre-
Definition-Time and 300 μs Hit-Lockout-Time at a threshold vious step. In contrast to the previous step, now a transient
level of 45 dB AE . The preamplification factor was chosen as calculation of the displacement field is performed. In addi-
40 dB AE for fiber breakage and as 20 dB AE for matrix crack- tion, boundary conditions at the crack plane are chosen to
ing. allow for crack opening according to a fracture mechanics
law. The duration of this transient calculation t f rac is chosen
to be sufficient until crack propagation has come to a rest.
3 Model Description As seen from Fig. 4, the presence of the static displacement
field causes crack propagation with an accompanying exci-
The model strategy uses the finite element method as tation of an acoustic emission wave. This spatial movement
implemented in the commercial software “COMSOL Mul- is seen best in the velocity field, since the static displacement
tiphysics” and comprises a combination of multi-scale and dominates the displacement scale and therefore inhibits the
multi-physics approaches. All calculations were carried out identification of the very small displacements caused by the
using the “structural mechanics module” as available in acoustic emission wave. A detailed discussion of the crack
COMSOL version 4.4. All descriptions used in the following growth implementation is given in Sect. 3.1.
refer to this version of the software package. For the third step the initial conditions (t = 0) for the
The source model description proposed herein consists of displacement, velocity and stress states of the last time value
three sequential modeling steps as schematically presented of the previous step (t f rac ) are used. Boundary conditions
in Fig. 4. The first step is derived from classical structural for restricted displacement components and external loads
mechanics. Suitable displacement boundary conditions are are kept identical to the first and second step. The boundary
defined for the geometry considered to restrict some of the conditions applied at the crack plane are chosen to allow for
displacement components on one end (cf. Fig. 2). The other independent movement of the new crack surfaces without
end is loaded by a force high enough to initiate fracture at the allowing penetration of each other. This transient calculation
crack plane considered. If this force value is unknown, the is continued for a sufficient duration tend to allow for signal
implementation of a fracture criterion (e.g. fracture tough- propagation in the test geometry as shown in Fig. 4.
ness, max. stress, etc.) to deduce the onset load for crack
initiation is a straight forward procedure using a station- 3.1 Implementation of Crack Propagation
ary solver sequence with incremental loading. If the external
force is known from experiments, the measured force value The present implementation of crack growth requires the def-
can directly be used for the stationary solver. For the exam- inition of a fracture plane, similar to conventional cohesive
ple shown in Fig. 4, the presence of the notch causes stress zone modeling. In the example given in Fig. 4, the frac-

123
J Nondestruct Eval (2015) 34:4 Page 5 of 13 4

t=0 t=0
z-displacement field
= =

t=0 t=0 z-stress field

= =

t=0 z-velocity field

=

z
y
x

t = 0; = 0; =1 t≤ ; = ( , ) <t≤ ; =0

first step second step third step

(loading) (crack propagation) (signal propagation)

Fig. 4 Schematic of source model description using three subsequent modeling steps

ture plane chosen is the horizontal xy-plane located at the To model crack growth, the stiffness vector is multiplied
z-position of the notch tip. The extension of the fracture plane by a degradation function C( r ) evaluated as a function of the
in x-direction is chosen sufficiently large that the crack will position on the fracture surface r.
not grow beyond the end of that plane. The boundary con- One example to define such a degradation function is the
dition “thin elastic layer” as available in COMSOL 4.4 can von Mises equivalent stress σv . For a general stress state this
then be defined for such an internal surface. is written in terms of the normal stresses σi and shear stresses
The stiffness vector k of this thin elastic layer is written τi j as:
in terms of the boundary coordinate system (t1 , t2 , n), the
Young’s modulus E, the shear modulus G and the Poisson’s 
ratio ν as: σv = σx2 +σ y2 +σz2 −σx σ y −σx σz −σ y σz +3(τx2y +τx2z +τ yz
2 )

⎛ ⎞ (4)
kt1
k = ⎝ kt2 ⎠ (1)
kn Degradation of the stiffness vector k occurs if σv exceeds the
materials tensile strength σt .
G For technical reasons, the Comsol environment also
kt1 = kt2 = (2) requires an additional ordinary differential equation to be
th
defined on the fracture surface. This is to track the historic
E(1 − ν) maximum value σmax of σv . Therefore, the current implemen-
kn = (3)
th (1 + ν)(1 − 2ν) tation evaluates, whether the fracture condition is fulfilled in
The parameter th is an effective thickness associated with the the present time step i or was fulfilled in any previous time
thin elastic layer. The thickness value th is chosen sufficiently step.
small (i.e. < 1 nm), so that the value of k has negligible Therefore the degradation function is written in terms of
influence on the overall compliance of the model. the maximum value of either σmax or σv :

123
4 Page 6 of 13 J Nondestruct Eval (2015) 34:4


r ) , σv (
1 i f max(σmax ( r )) < σt nal propagation within the test block and piezoelectric signal
r) =
C ( (5)
r ) , σv (
0 i f max(σmax ( r )) ≥ σt conversion for a total duration of 50 μs.
For the epoxy resin, the carbon fiber and the aluminum
For the brittle materials used in the present study, this simple block we use the isotropic material properties as given in
description of material failure was found to be applicable. Table 1. The values for the epoxy resin and carbon fibers were
However, for materials involving larger amounts of plasticity obtained from in-house measurements following established
prior to failure or significantly different interaction between standards. The values of the aluminum alloy were taken from
normal stresses σi and shear stresses τi j then assumed by Eq. the material supplier’s datasheet. For the different materials
(4) other formulations for Eq. (5) have to be used to capture included within the piezoelectric acoustic emission sensor,
the material behavior. the required material properties are listed in Tables 1 and
The advantage of the present description compared to 2 using Voigt notation for the subset indices. The accuracy
other formulations for acoustic emission source models is the of the sensor model geometry and the according material
access to experimental parameters. In the proposed model, properties was validated in detail in [24].
crack growth and acoustic emission is solely defined by the
macroscopic loading condition and the failure criterion used. 3.3 Signal Detection and Post Processing
In particular, no explicit source function comprising internal
forces or rise-times are necessary to initiate an acoustic emis- In order to allow comparability to experimentally obtained
sion signal. signals the detection process by the acoustic emission sen-
sor and the subsequent acquisition chain has to be taken into
account. In the following we use a model of the WD sensor
3.2 Discretization Settings and Material Properties with parameters validated in [24] following the piezoelec-
tric sensor modeling developed in [23]. The model uses the
We conducted convergence studies to set up the discretiza- piezoelectric formulation used within the structural mechan-
tion levels used for the model. As the measure of compari- ics module of COMSOL. To consider the influence of the
son we use the displacement field values at the position of attached cable and the preamplifier we use the P-SPICE cir-
crack initiation (position specific for each model) and the cuit simulation integrated in the ACDC module of COMSOL.
acoustic emission voltage signal as computed by the mod- The circuit outline follows the considerations in [23] com-
eled acoustic emission sensor. We use a refinement strategy prising a low-pass given by a 10  resistor 90 pF capacitor
for mesh and time resolution following [20]. Convergence is combination to model the sensor cable and a parallel series
achieved, when signals of the refined model are within 97 % of a 10 k resistor 15 pF capacitor combination to model
coherence with the selected model. As measure of coherence the preamplifier input properties. The acoustic emission sig-
we focus on the bandwidth between 0 and 10 MHz for the nal is obtained as voltage across the 10 k resistor (see [23]
evaluation of the displacement field at crack initiation and the for detailed description and a circuit diagram). Although this
bandwidth between 0 and 1 MHz for the voltage signal of the explicit sensor modeling considers the multi-resonant behav-
acoustic emission sensor. This way of comparison follows the ior of the WD sensor, the band-pass characteristics of the
routine published in [20]. For the present model configura- preamplifier and the acquisition card are not accounted for
tion seen in Fig. 2 we use a mesh resolution of a maximum in the simulation. Therefore the simulated voltage is subject
edge length of 1.0 mm with several refinement steps when to a subsequent band-pass filtering using a 6th order Butter-
approaching the fracture plane and slight coarsening towards worth high-pass of 100 kHz frequency in combination with
the edges of the aluminum block. The fracture plane itself is a 6th order Butterworth low-pass of 1,000 kHz frequency.
meshed with a maximum edge length of 0.4 μm. All elements
are tetrahedral with quadratic geometry shape order.
For the time step we chose 0.01 ns during the process of 4 Results
crack growth for the carbon fiber breakage and 0.1 ns for the
description of crack growth in the epoxy resin. This differ- In the following we present the results of the source model
ence in convergent time step solutions is due to the vastly computations. The first Sect. 4.1 deals with the temporal and
different sound velocities faced in these two material types. spatial acoustic emission source activity. In Sect. 4.2 we com-
The duration t f rac of this first transient calculation is car- pare the results of the modeling approach to experimental
ried out five times longer than the duration the crack needs results for the setup shown in Fig. 2. Section 4.3 is dedicated
to propagate through the material. This is to allow sufficient to the comparison between the newly obtained results rela-
spreading of the high-frequency components before switch- tive to acoustic emission source model strategies employing
ing to a coarser time step in the subsequent step. This coarse static geometries in conjunction with an analytic rise-time
time step was chosen as 10 ns and is used to compute sig- function (cf. Sect. 1).

123
J Nondestruct Eval (2015) 34:4 Page 7 of 13 4

Table 1 Structural material

properties used in the simulation Material Density ρ (kg/m3 ) Poisson-ratio Elastic moduli (GPa)
process
Epoxy resin (RTM6) 1,140 0.38 E = 2.8
Carbon fiber (HTA) 1,770 0.20 E = 240.0
Aluminum block (6063-T83) 2,700 0.33 E = 69.0
Sensor case (stainless steel) 7,970 0.29 E = 219
Backing material (Al-filled epoxy) 2,700 0.40 E = 30.0
Acoustic insulation (PTFE) 2,200 0.33 E = 0.4
Bonding (Ag-filled epoxy) 1,700 0.45 E = 2.7
Wear plate (Al2 O3 ) 3,965 0.22 E = 400.0
Sensor elements (PZT-5A) 7,750 – C11 = 120.3
C12 = 75.2
C13 = 75.1
C22 = 120.3
C23 = 75.1
C33 = 110.9
C44 = 21.1
C55 = 21.1
C66 = 22.6

Table 2 Piezoelectric properties of PZT-5A ns, which approaches the crack tip velocity limit given by
Material Coupling constants Relative the Rayleigh velocity of 6652 m/s calculated based on the
(C/m2 ) permittivity approximation given by [25]. As consequence of the crack
process a dynamic displacement field is generated. As pre-
Sensor elements (PZT-5A) S31 = −5.4 χ11 = 919.1
viously noted in Sect. 3, this displacement field is hard to
S32 = −5.4 χ22 = 919.1
visualize due to the superimposed static displacement field.
S33 = 15.8 χ33 = 826.6 Therefore discussion is made with reference to the veloc-
S24 = 12.3 ity field instead. Shown in Fig. 5 is the z-component of the
S15 = 12.3 velocity field. The formation of the velocity field follows
the progress of the crack tip and the wave also propagates
into the adjacent materials. The sound velocities of the car-
4.1 Temporal and Spatial Acoustic Emission bon fiber and the resin part differ by one order of magnitude
Source Activity which causes substantially different distances of the wave
front after t = 1.4 ns in the two materials.
For modeling of fiber breakage we use the experimentally For the model of matrix cracking we use a notch with
obtained force value of 189 mN as load in the first (stationary) 1.0 μm radius to initiate crack propagation on one edge of
step. As additional geometric modification to the description the tapered area of the resin similar to the fiber breakage
in Sect. 2 we use a small notch with 0.1 μm radius applied model. As seen in Fig. 6, this also causes crack initiation
at one end of the fiber to produce a stress concentration at at the designated site. Based on microscopic observations,
one edge of the specimen. This modification is motivated by the fracture plane is selected in the tapered area of the resin.
fracture mechanics, which assumes the presence of a flaw Sometimes an inclination of the fracture planes and a natural
in the material causing subsequent fracture. In reality it is roughness of the fracture surface were observed. However,
likely, that failure will occur due to internal flaws at the posi- to demonstrate some fundamental relationships it is advan-
tion of highest stress concentration. The latter was observed tageous to use this simple morphology of the fracture plane.
to be at the position, where the carbon fiber is embedded In the stationary step, we use a force of 54 N as measured
in the resin. As seen by the images of the z-velocity field experimentally for the cross-section used. The resulting crack
in Fig. 5, the crack initiates at the notch position and prop- propagation in the resin takes 1.1 μs for a crack length of 800
agates through the carbon fiber at the designated fracture μm. This again approaches the crack tip velocity defined by
plane. The total duration for the crack length of 7 μm is 1.2 the Rayleigh velocity of 889 m/s of the epoxy resin calcu-

123
4 Page 8 of 13 J Nondestruct Eval (2015) 34:4

z z z
fracture
y x y x y x
plane

z-velocity [m/s]

z-velocity [m/s]

z-velocity [m/s]
initiation
site crack
growth
wave
epoxy carbon propagation
resin fiber

t = 0.1 ns t = 0.8 ns t = 1.4 ns

Fig. 5 Image stills of the acoustic emission source operation showing the z-velocity field at t = 0.1 ns, t = 0.8 ns and t = 1.4 ns for fiber breakage

epoxy
z resin z z
initiation
z-velocity [m/s]

z-velocity [m/s]

z-velocity [m/s]
crack
y site y growth y
x x x

wave
propagation
aluminum
pin

t = 100 ns t = 450 ns t = 900 ns

Fig. 6 Image stills of the acoustic emission source operation showing the z-velocity field at t = 100 ns, t = 450 ns and t = 900 ns for matrix cracking

lated based on the approximation given by [25]. As conse- the fracture surface was also recently validated by means of
quence of the crack propagation, an acoustic wave propagates electromagnetic emission measurements [28]. However, the
into the adjacent materials, which is subject to immediate present case considers fracture of a free fiber on the upper
interaction with the nearby boundaries, the newly formed half, so no further boundary constraints act on the surface of
crack surface and the different velocities of the adjacent the fiber. This situation will be slightly different in a fiber rein-
materials. forced composite, where the bond of the fiber to the surround-
In order to discuss this matrix source model relative to pre- ing matrix will reduce the amount of vibration of the fiber
vious modeling concepts, a cross-section evaluation is per- surface.
formed using the cut-plane function in Comsol. Therefore The predictions of the maximum z-displacement z theor y
we evaluate the average z-displacement of the lower half of calculated according to the theory of Green [26] is marked as
the fracture surface using a position slightly offset (shifted a dashed line in Fig. 7b. This is evaluated using the Young’s
by z = −10−22 m) to the initial position. modulus E and Poisson ratio ν, as well as the accumulated
The respective evaluation of the z-displacement of the stress σ and the crack propagation length a:
fracture surface of the fiber breakage model is shown in
Fig. 7a, b. During the short duration of crack propagation 4 · (1 − ν)
t f rac , the z-displacement increases continuously, but does z theor y = ·σ ·a (6)
π·E
not settle at the moment of crack-through. Instead, the z-
displacement increases further until a maximum value is As can be seen from the comparison, for the case of fiber
reached. Subsequently, the fracture surface starts to vibrate breakage, the analytical value is significantly lower than the
and settles at a new equilibrium position. This vibration of result of the numerical model. This is attributed to various
the fracture surface has been reported in previous model- reasons. For a simple case following the assumptions of sta-
ing attempts [11,24,27] and points out one important differ- tic extension of a flaw due to an external load as made by
ence to the classically assumed source functions including Green [26] the results between the numerical model and
only a step-function like behavior. The presence of these the analytical computation were found to be in 99 % agree-
surface vibrations superimposed to a step-function rise of ment. But as soon as dynamic displacements are taken into

123
J Nondestruct Eval (2015) 34:4 Page 9 of 13 4

(a) 1.0 (b)

7
0.9
negative z-displacement [µm]

negative z-displacement [µm]

0.8 6

0.7 5
t frac = 1.2 ns
0.6
4
0.5
3
0.4

0.3 2

0.2 second step third step 1

0.31 µm
0.1
0
0.0
0.000 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009 0.010 0 10 20 30 40 50

time [µs] time [µs]

Fig. 7 Average z-displacement of the fracture surface for fiber breakage in the initial part (a) and for the full duration of the computation (b). For
comparison the analytic solution of Greens theory is given as dashed line

(a) 14 (b) 14
tfrac = 1.1 µs
12.2 µm
12 12
negative z-displacement [µm]
negative z-displacement [µm]

10 10

8 8

6 6

4 4
second step third step
2 2

0 0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 0 10 20 30 40 50

time [µs] time [µs]

Fig. 8 Average z-displacement of the fracture surface for matrix cracking in the initial part (a) and for the full duration of the computation (b).
For comparison the analytic solution of Greens theory is given as dashed line

account, the analytical prediction underestimates the strength over, the oscillation frequency is different and does not decay
of acoustic emission sources. For the present geometry it is as fast as for fiber breakage. This is also attributed to the dif-
hard to approximate the situation as a notched beam with infi- ference in Rayleigh wave velocity and the different geometry.
nite extension in one axis. Therefore, additional geometrical Also, the maximum source displacement occurs before final
effects are likely, which also cause deviation of the source crack-through. This is due to the averaging process of the z-
energy release. However, as will be demonstrated in Sect. 4.2, displacement of the full fracture surface. In the beginning the
the computed source displacements turn into acoustic emis- newly forming fracture surface close to the initiation moves
sion signals, which are in good agreement with the experi- in the negative z-direction. After a certain time the move-
mental signals. ment of this part of the fracture surface settles and starts
In Fig. 8a, b the average z-displacement of the fracture sur- to move in the positive z-direction. The latter movement is
face of the matrix crack is shown. This was evaluated using already present before the final crack-through and therefore
a cut-plane analogous to the procedure for fiber breakage. contributes to the averaging process. The prediction given by
Compared to the fiber breakage case there are several obvi- the theory of Green [26] is still lower than the peak value,
ous differences. The initial rise of the signal is slower than for but is systematically above the average z-displacement levels
fiber breakage. This is owed to the different Rayleigh veloci- after t = 3 μs. This difference is again owed to the geometric
ties limiting the speed of crack propagation and the different differences between the assumptions made in [26] and the
length of crack propagation faced in the two setups. More- present situation seen in Fig. 6.

123
4 Page 10 of 13 J Nondestruct Eval (2015) 34:4

(a) 1200 0.0

(b) 1200 0.0
2.6E+05 2.6E+05
1000 1000

frequency [kHz]
frequency [kHz]

5.2E+05 5.2E+05
7.8E+05 7.8E+05
800 1.0E+06 800 1.0E+06
1.3E+06 1.3E+06
600 1.6E+06 600 1.6E+06
1.8E+06 1.8E+06
400 2.1E+06 400 2.1E+06
2.3E+06 2.3E+06
200 200

0 0
5 10 15 20 25 30 35 5 10 15 20 25 30 35
amplitude [V]

amplitude [V]
4 4
2 2
0 0
-2 -2
-4 -4
5 10 15 20 25 30 35 5 10 15 20 25 30 35
time [µs] time [µs]

(c) 1200 0.0

(d) 1200 0.0
2.6E+05 2.6E+05

frequency [kHz]
1000 1000
frequency [kHz]

5.2E+05 5.2E+05
7.8E+05 7.8E+05
800 1.0E+06 800 1.0E+06
1.3E+06 1.3E+06
600 1.6E+06 600 1.6E+06
1.8E+06 1.8E+06
400 2.1E+06 400 2.1E+06
2.3E+06 2.3E+06
200 200

0 0
5 10 15 20 25 30 35 5 10 15 20 25 30 35
amplitude [V]

amplitude [V]

4 4
2 2
0 0
-2 -2
-4 -4
5 10 15 20 25 30 35 5 10 15 20 25 30 35
time [µs] time [µs]

Fig. 9 Comparison between simulated (a, c) and experimental (b, d) results of fiber breakage and matrix cracking, respectively

4.2 Comparison to Experimental Results the fiber breakage case. Moreover, this also indicates, that
the von Mises failure criterion is applicable for the present
In this section comparison is made between the results case.
obtained from the numerical modeling and the correspond- Also for the matrix cracking case shown in Fig. 9c, d there
ing experimental signals. As noted in Sect. 3, the modeled is very good agreement in the voltage scales and the time-
signals are the result from a full 3D computation including frequency signature shown in the Choi-Williams distribution.
piezoelectric conversion within the sensor model application Slight differences arise in the modeled signal after t = 10 μs.
of a P-Spice circuit simulation and subsequent band-pass This is due to the repetitive approach-retract cycles of the
filtering. After amplification of the modeled signals by 20 newly formed fracture surface. In the modeling part, those
dB AE or 40 dB AE (see Sect. 2) this allows for direct compar- fracture planes are partially restricted in their relative motion
ison of the acoustic emission signal amplitudes in the voltage due to the selected symmetry plane. Therefore their dominant
scale. movement direction is in the z-direction. In the experimental
For the case of fiber breakage, the comparison is found part, the fracture surface might experience additional slid-
in Fig. 9a, b. As seen from the voltage scale and the time- ing and torsional motions as well as additional interlocking
frequency signature given in the Choi-Williams distribution of rough surface parts. This may account for the smoother
there is very good agreement between the modeled and the appearance of the spikes in the experimental signals. How-
experimental signal. In particular, the signal amplitudes show ever, the signal amplitudes are in good agreement, which
almost identical peak values and the echoes of the initial pulse also indicates the validity of the source movement reported
are adequately captured. Therefore, we assume the source in Sect. 4.1 and the applicability of the von Mises failure
function and intensity as reported in Sect. 4.1 to be valid for criterion for the present modeling work.

123
J Nondestruct Eval (2015) 34:4 Page 11 of 13 4

(a) 0.15 point source model (b) 100 point source model
extended source model extended source model
new source model new source model
0.10

FFT-magnitude [1/V]
10
0.05
amplitude [V]

0.00 1

-0.05
0.1
-0.10

-0.15 0.01
0 5 10 15 20 25 30 35 0 200 400 600 800 1000 1200 1400 1600 1800 2000
time [µs] frequency [kHz]

(c) 100 point source model (d) 100000 point source model
extended source model extended source model
80 new source model new source model
60 10000

FFT-magnitude [1/V]
40
amplitude [V]

20 1000

-20 100

-40

-60 10

-80

-100 1
0 5 10 15 20 25 30 35 0 200 400 600 800 1000 1200 1400 1600 1800 2000
time [µs] frequency [kHz]

Fig. 10 Comparison of computation results for three different source model descriptions for fiber breakage case (a,b) and matrix cracking (c,d)

4.3 Comparison Between Different Source Modeling along the z-axis is chosen. The dipole is positioned at the
Strategies center of the fiber and the tapered area of the resin, respec-
tively.
After comparing the results of the newly proposed source As second model concept, the same prescribed force-
model to experimental results, the aim of this section is to time source functions are used. However, the active area is
discuss the relevance relative to previous source modeling extended to the full fracture surface. Therefore the full fiber
concepts. cross-section and the full resin cross-section are subject to
Therefore, the model setup shown in Fig. 2 is taken as an the cosine bell step function. This model will be referred to
example and signals are calculated in this identical geometry as extended model. The third model concept uses the newly
using the same sensor model, but using three different source proposed implementation described in the sections above and
model concepts. will therefore be referred to as new source model.
The first source model concept will be referred to as point In Fig. 9a, b a comparison is made between the unfiltered
source model. This model uses the implementation of an results of the three source types for modeling fiber break-
acoustic emission source as internal point couples applying a age. This was chosen to discuss the differences of the three
cosine bell force-time function with the experimentally mea- descriptions in the highest possible bandwidth (i.e. not lim-
sured force values of 189 mN and 54 N for fiber breakage and ited by the experimentally used range). As seen in Fig. 9a for
matrix cracking, respectively. As rise-time 0.1 μs is selected the fiber breakage case, all three models yield comparable
for the fiber breakage model and 1.0 μs is used for the matrix source amplitudes. Also, the signals frequency content and
cracking model since these were calculated to be the dura- shape are still in reasonable agreement as seen in Fig. 9b.
tion until the maximum z-displacement was reached for fiber Considering the geometrical arrangement as seen in Fig. 5,
breakage and matrix cracking, respectively (cf. Figs. 7b, 8b). this is not unexpected. Although the fiber has a certain geo-
A simple dipole representation of 1 μm axis length directed metrical extension, the excited wavelengths are of the same

123
4 Page 12 of 13 J Nondestruct Eval (2015) 34:4

or larger dimension. Therefore, despite of the geometrical was demonstrated that previous source model descriptions,
complexity of the source it seems sufficient to be described such as point couples or prescribed static geometries cannot
by a single dipole source operating at this position. account for the dynamic processes around the source once the
For the case of matrix cracking shown in Fig. 9c this seems geometrical dimensions of the source approaches the wave-
to be different. Here, the point source models result in ampli- length of the generated signals. However, careful revision
tudes, which are of the same order of magnitude as the new is required for the applied failure criterion and constitutive
source model. In contrast, the extended source model over- equations, if large plastic deformation is expected prior to
shoots this range by a factor of four. But the signal arrivals, failure or other interaction between normal and shear stress
the amplitudes and the frequency content (cf. Fig. 9d) do components occurs.
not show a close match for the three cases. This is readily As with all cohesive zone modeling approaches, the
explained by the dimension of the source relative to the wave explicit definition of a fracture surface also requires some
lengths involved. As seen by the spreading of the wave field assumptions. However, for simple load cases, the position
in Fig. 6, the size of the source is two orders of magnitude of the fracture surface is straightforward or readily deduced
larger than for the fiber breakage case. Hence, the wave length from microscopic observations after fracture. Also, inclu-
of the initially emitted wave starts to interfere with the sur- sion of more complex fracture surfaces to account for further
rounding boundaries and causes interference with the wave details of experimental fracture morphologies is straightfor-
still being emitted by the source. Therefore the spatial posi- ward in the approach presented.
tion and sequence of excitation does play an important role,
which is not adequately captured by a point source model or Acknowledgments I would like to thank Marvin Hamstad and Malte
Peter for the fruitful discussions on acoustic emission source modeling.
the extend source model. Hence, the newly proposed source
model is expected to yield a more realistic description of the Open Access This article is distributed under the terms of the Creative
displacement field caused by the crack propagation (Fig. 10). Commons Attribution License which permits any use, distribution, and
reproduction in any medium, provided the original author(s) and the
source are credited.

5 Conclusion

We presented a new approach to model acoustic emission

References
sources. In contrast to previous source model descriptions,
the proposed model does not require definition of a force-time 1. Livne, A., Bouchbinder, E., Fineberg, J.: Breakdown of linear elas-
curve as source function. Moreover, no assumptions regard- tic fracture mechanics near the tip of a rapid crack. Phys. Rev. Lett.
ing directivity of source couples, their relative intensities or 101, 264–301 (2008)
positioning is required. The present source model operates 2. Livne, A., Bouchbinder, E., Svetlizky, I., Fineberg, J.: The near-tip
fields of fast cracks. Science 327, 1359–1363 (2010)
solely based on accessible experimental parameters such as 3. Buehler, M., Gao, H.: Dynamical fracture instabilities due to local
external loads, a constitutive equation for the material and a hyperelasticity at crack tips. Nature 439, 307–310 (2006)
fracture mechanics based failure criterion. The latter initiates 4. Fineberg, J., Gross, S.P., Marder, M., Swinney, H.L.: Instability in
and stops crack propagation using a cohesive zone element dynamic fracture. Phys. Rev. Lett. 67, 457–460 (1991)
5. Große, C., Ohtsu, M.: Acoustic Emission Testing in Engineering:
modeling approach. This enables an improved representation Basics and Applications. Springer Verlag, Berlin (2008)
of acoustic emission sources and avoids additional assump- 6. Aki, K., Richards, P.G.: Quantitative Seismology, Theory and
tions on source strengths or source rise-times not accessible Methods. University Science Books, Sausalito (1980)
by experimental means. 7. Giordano, M., Calabro, A., Esposito, C., D’Amore, A., Nicolais,
L.: An acoustic-emission characterization of the failure modes in
The acoustic emission signals generated by the present polymer-composite materials. Compos. Sci. Technol. 58, 1923–
source model description have been validated against exper- 1928 (1998)
imental signals obtained from micro-mechanical experi- 8. Ohtsu, M., Ono, K.: The generalized theory and source represen-
ments. In these experiments the failure mode is easily observ- tation of acoustic emission. J. Acoust. Emiss. 5, 124–133 (1986)
9. Lysak, M.: Development of the theory of acoustic emission by prop-
able using microscopy imaging and the external load is agating cracks in terms of fracture mechanics. Eng. Fract. Mech.
straight-forward to measure. The model also comprises a 55(3), 443–452 (1996)
full 3D representation of the according propagation medium 10. Hamstad, M., O’Gallagher, A., Gary, J.: Modeling of buried
(large aluminum block) and a model of a commercially used acoustic emission monopole and dipole sources with a finite ele-
ment technique. J. Acoust. Emiss. 17(3–4), 97–110 (1999)
acoustic emission sensor. The latter implies a piezoelec- 11. Sause, M., Horn, S.: Simulation of acoustic emission in planar
tric conversion process in conjunction with a subsequent carbon fiber reinforced plastic specimens. J. Nondestruct. Eval.
P-SPICE circuit simulation to account for the impact of 29(2), 123–142 (2010)
the preamplifier. Simulated and experimental acoustic emis- 12. Hamstad, M.: On lamb modes as a function of acoustic emission
source rise time. J. Acoust. Emiss. 27, 114–136 (2009)
sion signals were found to show very good agreement. It

123
J Nondestruct Eval (2015) 34:4 Page 13 of 13 4

13. Ohtsu, M., Ono, K.: A generalized theory of acoustic emission and 23. Cervena, O., Hora, P.: Analysis of the conical piezoelectric acoustic
Green’s function in a half space. J. Acoust. Emiss. 3, 27–40 (1984) emission transducer. Appl. Comput. Mech. 2, 13–24 (2008)
14. Scruby, C.: Quantitative acoustic emission techniques. Nonde- 24. Sause, M., Hamstad, M., Horn, S.: Finite element modeling of
struct. Test. 8, 141–210 (1985) conical acoustic emission sensors and corresponding experiments.
15. Schubert, F.: Numerical Modeling of Acoustic Emission Sources Sensors Actuators A 184, 64–71 (2012)
and Wave Propagation in Concrete NDT.net 7, 9 (2002) 25. Sause, M.: Identification of Failure Mechanisms in Hybrid Materi-
16. Wilcox, P., Lee, C., Scholey, J., Friswell, M., Wisnom, M., als Utilizing Pattern Recognition Techniques Applied to Acoustic
Drinkwater, B.: Progress towards a forward model of the complete Emission Signals. mbv-Verlag, Berlin (2010)
acoustic emission process. Adv. Mater. Res. 13–14, 69–76 (2006) 26. Bergmann, L.: Der Ultraschall und seine Anwendung in Wis-
17. Prosser, W., Hamstad, M., Gary, J., O’Gallagher, A.: Comparison senschaft und Technik, p. 560. Hirzel-Verlag, Stuttgart (1954)
of finite element and plate theory methods for modeling acoustic 27. Green, A., Zerna, W.: Theoretical Elasticity. Oxford University
emission waveforms. J. Nondestruct. Eval. 18(3), 83–90 (1999) Press, London, New York (1954)
18. Hamstad, M.A., O’Gallagher, A., Gary, J.: Modeling of buried 28. Burks, B., Kumosa, M.: A modal acoustic emission signal clas-
acoustic emission monopole and dipole sources with a finite ele- sification scheme derived from finite element simulation. Int. J.
ment technique. J. Acoust. Emiss. 17(3–4), 97–110 (1999) Damage Mech. 23(1), 43–62 (2013)
19. Hora, P., Cervena, O.: Acoustic emission source modeling. Appl. 29. Gade, S., Weiss, U., Peter, M., Sause, M.: Relation of electromag-
Comput. Mech. 4, 25–36 (2010) netic emission and crack dynamics in epoxy resin materials. J.
20. Rose, J.: Ultrasonic Waves in Solid Media. Cambridge University Nondestruct. Eval. 33, 711–723 (2014)
Press, Cambridge (1999)
21. Sause, M., Hamstad, M., Horn, S.: Finite element modeling of lamb
wave propagation in anisotropic hybrid materials. Compos. B 53,
249–257 (2013)
22. McLaskey, G., Glaser, S.: Acoustic emission sensor calibration for
absolute source measurements. J. Nondestruct. Eval. 31(2), 157–
168 (2012)

123

View publication stats