Dimensionality of Elastic Wave Scattering

3 On the Dimensionality of Elastic Wave

6 Anton Van Pamel1

7 Department of Mechanical Engineering

8 Imperial College London

9 London, SW7 2AZ, UK

10

11 Peter B. Nagy

12 Department of Aerospace Engineering and Engineering Mechanics

13 University of Cincinnati

14 OH 45221-0070, USA

15

16 Michael J. S. Lowe

17 Department of Mechanical Engineering

18 Imperial College London

19 London, SW7 2AZ, UK

1
a.van-pamel11@imperial.ac.uk
Manuscript accepted for publication in J. Acoust. Soc. Am. 140(6), December 2016.

1 /18
 Dimensionality of Elastic Wave Scattering

20 Abstract - Elastic waves scatter when their wavelength becomes comparable to random spatial

21 fluctuations in the elastic properties of the propagation medium. It is postulated that within the long-

22 wavelength Rayleigh regime, the scattering induced attenuation obeys a D=1,2,3 dimensional

23 dependence on wavenumber, kD+1, whilst within the shorter-wavelength stochastic regime, it becomes

24 independent of the dimensions and thus varies as k2. These predictions are verified numerically with a

25 recently developed Finite Element method in both 3D, 2D, and 1D, for the example of ultrasonic waves

26 propagating within polycrystalline materials. These findings are thought to be practically useful given

27 the increasing uptake of numerical methods to study highly scattering environments which exhibit

28 multiple scattering, but often remain limited to 2D given computational constraints. It is hoped that

29 these results lay the groundwork for eventually producing computationally efficient 2D simulations

30 which are representative of 3D.

31 Pacs: 43.35C

2 /18
 Dimensionality of Elastic Wave Scattering

32 I. INTRODUCTION

33 Scattering of elastic waves within heterogeneous media applies to a great number of problems in

34 elastodynamics, including seismology1 and ultrasonic non-destructive evaluation2. For some it enables,

35 for example, the characterisation of the Earth’s crust (e.g. Zhao et al.3) whilst for others it hinders, say

36 the detection of defects buried in metallic materials4. In the latter case, and within the context of this

37 article, scattering arises when the probing wavelength becomes comparable to the characteristic

38 dimension of random spatial variations in the elastic properties of the medium.

39 Irrespective of its application, scattering remains a complex phenomenon. A great variety of possible

40 heterogeneities and altering wave behaviour throughout different scattering regimes, restrict analytical

41 models where multiple scattering5,6 in particular presents difficulties. Numerical techniques7–10 are more

42 flexible and capture both single and multiple scattering, yet are computationally expensive due to the

43 requirement of modelling two length-scales, one corresponding to the wavelength and another to that

44 of the heterogeneity. The continual growth in computational resource however, makes numerical

45 approaches such as the Finite Element (FE) method increasingly attractive11–15 to study wave scattering

46 phenomena. Recent such progress15 has shown FE to be capable of capturing the range of physics

47 required to describe the different scattering regimes.

48 Even so, its computational cost remains significant; full-scale 3D studies are often unfeasible and are

49 likely to remain so for many practical cases of interest. Hence for the time being, the majority of

50 investigations are limited to 2D11–14,16–20 which are not necessarily representative of 3D scattering. At

51 least the implication of a 2D approximation to the scattering problem is not sufficiently understood at

52 present, particularly so across the different scattering regimes. Understanding the dimensional

53 behaviour of scattering therefore becomes important not only for future numerical studies but also for

54 laying the groundwork to achieve more efficient 2D simulations that are usefully representative of 3D.

55 Scattering behaviour can be studied either in terms of the amplitude loss incurred by the coherent wave

56 field, or the intensity gained by its counterpart, the incoherent scattered wave field. The former

57 constitutes the scattering induced attenuation, usually denoted by the attenuation coefficient, α as an

3
 Dimensionality of Elastic Wave Scattering

58 exponential decay in Equation 1, where U describes the coherent wave amplitude for a plane wave

59 propagating at distance z.

𝑈(𝑧) = 𝑈0 𝑒 −𝛼𝑧 (1)

60 This article treats the scattering induced attenuation and studies its governing dependence on

61 wavelength (or wavenumber) within relatively small yet characteristic numerical models to demonstrate

62 the dimensional behaviour across different scattering regimes in 3D, 2D, and 1D. A range of

63 wavelengths are studied such that the long-wavelength Rayleigh regime and the shorter-wavelength

64 stochastic regime are considered for the example of ultrasonic waves propagating within polycrystalline

65 materials.

66 The subsequent Section II presents preliminary theories that describe the dimensional dependence of

67 scattering for both regimes. Section III holds the numerical calculations which verify these postulations.

68 II. THEORETICAL CONSIDERATIONS

69 Scattering within heterogeneous media, including polycrystalline materials, embodies different regimes

70 – each regime describing a different physical approximation to a particular scattering behaviour. The

71 validity of each region can be classified by three measures of scale1, foremost of which is the normalised

72 wavenumber, kd, where k is the wavenumber and d is the length scale of the heterogeneity, in this case

73 the average grain size. The two remaining dimensionless numbers are the normalised propagation

74 length, L/d, where L is the propagation distance; and the perturbation index, n, which is a property solely

75 related to the medium, here defined by the degree of elastic variation within the polycrystalline material.

76 Within the practical confines of ultrasonic NDE, three scattering regimes can be readily identified21.

77 Firstly, as a long wavelength approaches dimensional similarity to the heterogeneities, but when k still

78 remains approximately two orders of magnitude smaller than d, the medium ceases to be perceived as

79 a homogenous effective medium and Rayleigh scattering emerges. As the wavelength continues to

80 reduce, Rayleigh scattering transitions into an intermediate scattering regime21 which eventually settles

81 into a stochastic scattering region when kd~1. Beyond this scope, lies the Geometric-stochastic

82 transition21, occurring at kd~1/n, where ray theory approximations become applicable and the medium

4
 Dimensionality of Elastic Wave Scattering

83 once again recedes to an, albeit piecewise, homogenous effective medium. Scattering here is

84 subsequently independent of the wavelength for all dimensions; this leaves us to discuss the two

85 aforementioned regimes.

86 A. Rayleigh scattering

87 Mason and McSkimmin22 experimentally discovered the fourth order frequency dependence of the

88 scattering induced attenuation for ultrasonic waves within polycrystalline materials, more commonly

89 observed for electromagnetic waves as Rayleigh Scattering23. This finding established that to an elastic

90 wave, a polycrystalline material appears as a homogenous medium containing randomly distributed

91 scattering spheres24. It is perhaps unsurprising then that in a 2D representation, a scattering sphere would

92 be replaced by a circle, and in 1D by a line. It is thereby recognised that the wavenumber dependence

93 of Rayleigh scattering scales with D+125, which is a consequence of the respective Green’s functions

94 for a spherical, circular, and a planar scatterer. It therefore seems fair to postulate the same dimensional

95 behaviour should be observed within polycrystalline materials.

96 We can use the rule of similitude26 to further infer the dimensional dependence on grain size, d, which

97 relates the dependencies of k and d such that the attenuation is proportional to the i - th power of d i, and

98 the (i + 1)-th power of ki+1. The above arguments can be cast into a unified form for D dimensional

99 scattering attenuation,  R , within the Rayleigh regime.

 R  d D k D 1 (2)

100 Whereas it is generally well known that attenuation follows a fourth power dependence on frequency

101 in the Rayleigh regime, it is not well appreciated that this dependence is only of the third order in a 2D

102 representation. Accordingly one should not expect predictions of scattering from inhomogeneities by

103 2D models to be representative of 3D behaviour either.

104 B. Stochastic scattering

105 Once the wavelength becomes dimensionally similar to the grain size, kd~1, the attenuation eventually

106 converges to a stochastic asymptote. This was first described by Huntington27 for the ultrasonic

5
 Dimensionality of Elastic Wave Scattering

107 propagation within polycrystalline materials by proposing a phase scrambling mechanism. A similar

108 approach known as the phase screen approximation28 is used here to derive the attenuation coefficient

109 of the coherent wave.

110 Let us consider a monochromatic plane wave traversing in the z-direction through a phase screen where

111 random fluctuations of phase rather than amplitude are imposed on the wave. Such a wave at one (x, y,

112 z) location can be described as an unperturbed wave field 𝑈0 perturbed by a phase perturbation 𝛿𝜃.

𝑈(𝑥, 𝑦, 𝑧) = 𝑈0 exp(𝑖𝛿𝜃(𝑥, 𝑦, 𝑧)) (3)

113 These changes in phase cause a loss in amplitude when summing the contributions that describe the

114 coherent wavefront emerging from the screen. This is denoted by ⟨ ⟩ to represent the lateral spatial

115 average along x,y denoted by:

⟨𝑈(𝑥, 𝑦, 𝑧)⟩ = U0 ⟨exp(𝑖𝛿𝜃(𝑥, 𝑦, 𝑧))⟩ (4)

116 Assuming the phase perturbation obeys a Gaussian distribution, the exponential term can be further

117 evaluated, as shown by Eckart28.

1 21
⟨𝑈(𝑥, 𝑦, 𝑧)⟩ ≈ 𝑈0 exp (− ⟨𝛿𝜃 2 (𝑥, 𝑦, 𝑧) ⟩ ) ≈ 𝑈0 exp (− 𝛿𝜃𝑅𝑀𝑆 (𝑧) ) (5)
2 2

118 Nagy and Adler29 found that for non-Gaussian distributions, the governing statistical property is still

119 the r.m.s. phase perturbation, but for loss values above 3 dB the relationship between the logarithmic

120 loss and RMS is not strictly quadratic. However, by virtue of the central limit theorem, the phase

121 perturbation over a large number of grains tends to be inherently Gaussian, therefore Eq. (5) is still

122 applicable although the total loss is well over 3dB, in excess of 10dB. These assumptions can break

123 down when considering, usually, unnaturally perturbed phase distributions, such as those encountered

124 by reflections from triangulated periodic surfaces30.

125 By comparison to the definition of the attenuation coefficient, exp(−𝛼𝑧), we can then write:

1 (6)
α≈ 𝛿𝜃 2 (𝑧)
2𝑧 𝑅𝑀𝑆

126 The phase perturbation in our case relates to a change in the local elastic properties of the medium,

127 causing a positive or negative fluctuation in wavenumber, 𝛿𝑘, away from the mean wavenumber 𝑘0 .

6
 Dimensionality of Elastic Wave Scattering

128 The phase then, after z-direction propagation through 𝑁 grains, is the summation of all the individual

129 phases through each n-th grain with an average grain size d.

𝑁 (7)
𝛿𝜃(𝑧) = ∑ 𝛿𝑘𝑛 𝑑
𝑛

130 This can be rewritten as an effective wavenumber, 𝛿𝑘′, in the form 𝛿𝜃(𝑧) = 𝛿𝑘 ′ (𝑧)𝑧, such that:

𝑁 (8)
′ (𝑧)
1
𝛿𝑘 = ∑ 𝛿𝑘𝑛 𝑑
𝑧
𝑛

131 In order to evaluate the r.m.s. phase perturbation featured in Equation 5, we can calculate the r.m.s value
′
132 of the effective wavenumber, 𝛿𝑘𝑅𝑀𝑆 , by substituting the summation in Equation 8 by its incoherent

133 sum.

(9)
√𝑁 𝑑
′
𝛿𝑘𝑅𝑀𝑆 (𝑧) = 𝛿𝑘𝑅𝑀𝑆 𝑑 = √ 𝛿𝑘𝑅𝑀𝑆
𝑧 𝑧

134 Which yields:

𝛿𝜃𝑅𝑀𝑆 (𝑧) = √𝑧𝑑 𝛿𝑘𝑅𝑀𝑆 (10)

135 From Equation 6, we can thus describe the attenuation coefficient in terms of 𝛿𝑘𝑅𝑀𝑆 .

2
𝑑𝛿𝑘𝑅𝑀𝑆 (11)
α≈
2

136 The relevance of this simplistic model lies in the prediction that, as long as the lateral correlation length

137 of 𝛿𝑘𝑅𝑀𝑆 (𝑥, 𝑦) is larger than the wavelength, the scattering-induced loss is proportional to the

138 propagation distance z in a way that is independent of the dimensionality of the model. This is similar

139 to Eckart’s original finding28 which stated that the losses incurred from acoustic waves reflected from

140 a rough surface are merely dependent on the properties of its roughness, the r.m.s. height in that case.

141 In our case, and within our approximations, Eckart’s ‘roughness’ is solely equivalent to the perturbation

142 in elastic constants.

143 It is therefore expected that, whilst we have seen dimensional dependencies in the Rayleigh regime, this

144 is not the case for the stochastic regime; the attenuation varies with the second power of frequency in

7
 Dimensionality of Elastic Wave Scattering

145 any representation: 3D, 2D or 1D. Notably in the 1D case, the attenuation follows a second order

146 dependence with frequency throughout both regimes. In summary, regardless of dimensions, the

147 dependence of the attenuation in the stochastic region is described by:

 S  d k2 (12)

148 The above result may seem encouraging for 2D models to approximate 3D scattering behaviour, but it

149 is important to note that despite their shared dimensional dependence, this analysis does not go as far

150 as to say that the absolute scattering and hence also attenuation should be equal.

151 III. NUMERICAL VERIFICATION

152 In verification of the preceding predictions, we numerically calculate the wavenumber dependence of

153 the scattering-induced attenuation for plane longitudinal waves propagating through random

154 heterogeneous elastic media in 3D, 2D, and 1D. Waves of millimetre-scale wavelengths with

155 frequencies in the MHz range are modelled to represent typical ultrasonic waves within polycrystalline

156 materials. The perceived attenuation within our models is solely due to scattering, as is the case

157 practically for polycrystalline materials, where other sources of attenuation, through for example

158 absorption, are negligible. The results are expected to be equally valid however for seismic waves31

159 which experience a similar scattering problematic albeit on a different scale.

160 A. Numerical methodology

161 The methodology employed here has been previously detailed in15. The procedure starts by generating

162 a numerical model with a spatial domain of randomly sized and oriented elastic crystals whose average

163 properties are representative of those occurring within polycrystalline morphologies. Voronoi diagrams

164 were relatively recently established for this purpose in various fields including Materials Science32–34

165 and Ultrasonic NDE12,14,15. Generating a Voronoi tessellation requires a set of spatially randomly

166 distributed points, which each serve as a seed for an eventual Voronoi cell, which will in turn represent

167 the geometry for an individual crystallite, or grain. Once such a grain morphology is obtained, in order

168 to emulate polycrystalline material properties, each crystallite is assigned a fixed density but a random

169 crystallographic orientation to determine its elastic properties. When both the seed distribution and

8
 Dimensionality of Elastic Wave Scattering

170 crystallographic orientations are spatially randomised, as is done here, this produces a macroscopically

171 isotropic material with equiaxed grains and no texture. The authors use the software Neper35 for this

172 part of the procedure.

173 The main obstacle to this numerical approach is the requirement to incorporate two length-scales, one

174 associated with the wavelength and a second with that of the heterogeneity. Namely, it has been shown

175 that to achieve numerical convergence15, the model mesh must be sampled to satisfy both a minimum

176 spacing of d/10 and /10 where  denotes the wavelength . Depending on the d/ ratio, this most often

177 results in a mesh sampling which is drastically finer, and thus more computationally expensive, than

178 conventional wave propagation modelling36. Here however, efficient solutions for the explicit time-

179 domain computations are found using the FE solver Pogo37 that exploits parallel processing on graphics

180 processors (GPU).

181 Once the mesh is generated, symmetry boundary conditions are applied to create a plane longitudinal

182 wave as detailed in15. Plane waves are desirable as they avoid influences of diffraction, such as those

183 usually associated with sources of finite length, and hence simplify the calculation for attenuation. In

184 the case of a 3D cuboid model exposing six surfaces, a toneburst is applied simultaneously to all of the

185 nodes which lie on one of these surfaces - as a force in the direction normal to the surface. The four

186 surfaces which share an edge with this excitation surface require a rolling boundary condition to

187 approach a plane wave as closely as possible. A perfect plane wave is achieved in the case of a

188 homogeneous medium, but only an approximated plane wave is achieved for the heterogeneous

189 medium; the approximation improves with increasing size of the lateral dimension. The remaining

190 surface which opposes the excitation surface receives the plane wave, hence simulating a pitch-catch

191 configuration. The same method can be extended to 2D and 1D models as illustrated in Figure 1.

9
 Dimensionality of Elastic Wave Scattering

(a) (b) (c)

Figure 1: Illustration of numerical models in (a) 1D (b) 2D (c) 3D. For clarity, the illustrations are not
representative of the model size.
Model size: 400mm(×400) 10mmx200mm 12mmx12mmx100mm

Grain size: 100µm 100µm 500µm

Centre Freq: 2MHz 2MHz 1MHz

DOF 2x105(×400) 42x106 345x106

Table 1: Details of FE models used for Rayleigh scattering calculations.

192 Remaining parameters such as model size also need to be considered to ensure a numerically accurate

193 solution. First in terms of noise, even a numerical measurement in this case contains a contribution of

194 both the desired coherent field and undesired incoherent field. The contribution of the incoherent field

195 however reduces as the lateral extent of the model is increased, due to (coherent) spatial averaging.

196 Secondly, the model must also be of optimal length to ensure the attenuation is within a desirable range

197 for measuring38. Both these constraints impose limitations on the attainable scattering regimes given

198 finite computational resources.

199 Finally, obtaining a value for attenuation involves calculating the received and sent time-amplitude

200 signals, by ensemble averaging the time-dependent nodal displacements across the send and receive

201 surface. The frequency dependent attenuation can be obtained (for example as shown in38) from the

202 Fourier Transform of each signal.

203 B. Rayleigh scattering results

204 The study of attenuation within the Rayleigh regime requires operating at a kd<<1 length scale. As

205 mentioned previously, relatively long (in the propagation direction) models are therefore preferred in

10
 Dimensionality of Elastic Wave Scattering

206 order for the grains to significantly reduce the received amplitude. This requirement is more taxing in

207 3D, and hence a larger grain size (500µm), is modelled than in the 1D and 2D models, where smaller

208 grain size (100µm) is used. The respective kd values for 3D, 2D, and 1D are 0.5, 0.2, and 0.2.

209 Illustrations and details of each model are shown in Figure 1 and Table 1. The material, Inconel, is

210 described in Table 2.

Material A ρ (kg/m3) C11 (GPa) C12 (GPa) C44 (GPa)

Inconel 2.8 8000 234.6 145.4 126.2

211 Table 2: Material properties used for Rayleigh scattering calculations.

212 Figure 2 plots the wavenumber dependent attenuation in (a) 1D (b) 2D (c) 3D where errorbars indicate

213 the standard deviation from at least 10 realisations of a random material. The mean wavenumber power

214 law is calculated including its standard deviation, which produces frequency exponents of 1.96, 3.01,

215 and 3.95 for 1D, 2D, and 3D respectively. These values are within satisfactory bounds to the expected

216 integer values of 2, 3, and 4, which thereby confirms the expected kD+1 dependence.

217 A. Stochastic scattering results

218 Different 3D and 2D models were created to assess attenuation in the stochastic regime; their properties

219 are shown in Table 3. In comparison to the previous section, a broader range of frequencies is now

220 investigated (2-20MHz) and the material used is shown in Table 4.

2D 2D 3D

Centre frequencies 1-5MHz 6-20MHz 2-20MHz

Model size 30mmx100mm 300mmx10mm 12mmx12mmx10mm

Grain size 500µm 500µm 500µm

DOF 3x106 15x106 278x106

221 Table 3: Details of FE models used for stochastic scattering calculations.

11
 Dimensionality of Elastic Wave Scattering

222

(a) (b)

(c)

Figure 2: Power law calculations for Rayleigh scattering induced attenuation, plotted as normalised

attenuation versus normalised wavenumber in (a) 1D (b) 2D (c) 3D.

223

Material ρ (kg/m3) CLV (m/s) A C11 (GPa) C12 (GPa) C44 (GPa)

Fictitious 8000 6000 1.5 262.1 136.5 95.3

224 Table 4: Material properties used for stochastic scattering calculations.

12
 Dimensionality of Elastic Wave Scattering

225 It is important to note that a question arises at this stage regarding the appropriate grain size distribution

226 to enable an objective comparison of 3D and 2D. In essence, a degree of ambiguity exists when

227 comparing a random distribution of spheres to that of cylinders. Stereology39 is occupied with the

228 inverse problem, namely, how to interpret a 2D measurement of a cross-sectional distribution onto that

229 of the 3D material. Since a 2D slice will not necessarily behave like its 3D parent, the simplest case is

230 assumed here and both media are compared with matching grain size distributions (see Figure 3). This

231 can be interpreted as our 2D model being a slice of a 3D material where every grain is sectioned through

232 its centre and therefore exhibits its largest radius. Whilst this assumption should not influence the

233 wavenumber dependence which is our primary interest, it will affect the absolute attenuation.

234

235 Figure 3: Grain size distribution of the 3D and 2D FE models used to investigate stochastic scattering.

236 The resulting normalised attenuation versus the normalised kd is plotted in Figure 4. Starting from a

237 high kd, it can be seen that within the stochastic region both the 3D and 2D curves remain parallel and

238 exhibit the same gradient. This result confirms the prediction that the wavenumber dependent behaviour

239 maintains dimensional insensitivity within the stochastic regime. An absolute shift in attenuation

240 remains however, in this case in the order of 10 to 20%.

13
 Dimensionality of Elastic Wave Scattering

241

242 Figure 4: Normalised attenuation against normalised wavenumber for 3D and 2D scattering. Both

243 curves exhibit independent transitional scattering behaviour, characterised by the ‘hump’, before

244 converging to a stochastic asymptote at kd~8.

245 As the wavenumber decreases from the stochastic regime, the attenuation mechanism falls into a

246 transitional regime21, here notable in Figure 4 by a characteristic but different ‘hump’ for both the 3D

247 and 2D results. This region constitutes scattering behaviour with rapidly varying dependencies on kd,

248 and also exhibits the largest absolute difference in attenuation between the two models. This is an

249 important observation for numerical modelling, as the transitional behaviour makes it difficult to deploy

250 analytical models here, yet it also shows 2D numerical models have the largest shortcomings within

251 this region.

252 Finally, at low kd and from the supporting evidence in the previous section, we can observe the 2D

253 attenuation curve failing to attain the steeper slope which is characteristic of Rayleigh scattering in 3D.

14
 Dimensionality of Elastic Wave Scattering

254 Whilst the absolute attenuation is lower in 2D within the stochastic regime, this trend reverses once

255 attenuation approaches the Rayleigh regime. Necessarily the two curves subsequently intersect in-

256 between the Rayleigh and stochastic regime.

257 IV. CONCLUSION

258 Our aim was to demonstrate and understand the dimensional behaviour of elastic wave scattering within

259 heterogeneous media. The fundamental dependence of the scattering induced attenuation on

260 wavenumber was postulated for various scattering regimes in 3D, 2D, and 1D. Numerical calculations

261 based on the Finite Element (FE) method simulated scattering from random spatial variations in elastic

262 properties of the medium, such as those experienced by ultrasonic waves propagating within

263 polycrystalline materials. As supported by established theory, it is numerically verified that within the

264 long-wavelength Rayleigh regime, the attenuation is proportional to kD+1 where D denotes the

265 dimension. For smaller wavelengths in the stochastic regime however, the calculations show a k2

266 dependence regardless of the dimensions. It is hoped that these findings will be useful for future studies

267 of scattering, and that they may lay the groundwork for developing an approach to achieve efficient 2D

268 simulations which are usefully representative of 3D phenomena.

15
 Dimensionality of Elastic Wave Scattering

269 REFERENCES
1
270 R.-S. Wu and K. Aki, Pure and Applied Geophysics, "Introduction: Seismic wave scattering in three-
271 dimensionally heterogenous earth", 128(1), 1-6 (1988).
2
272 E.P. Papadakis, in Ultrasonics: Methods in Experimental Physics, edited by P.D. Edmonds (Academic
273 Press, New York, 1981), pp. 237–298.
3
274 D. Zhao, Z. Wang, N. Umino, and A. Hasegawa, "Mapping the mantle wedge and interplate thrust
275 zone of the northeast Japan arc", Tectonophysics 467(1), 89-106 (2009).
4
276 R.B. Thompson, F.J. Margetan, P. Haldipur, L. Yu, A. Li, P. Panetta, and H. Wasan, "Scattering of
277 elastic waves in simple and complex polycrystals", Wave Motion 45(5), 655-674 (2008).
5
278 A. Ishimaru, Wave Propagation and Scattering in Random Media (Academic Press, New York, 1978)
279 pp.1-272.
6
280 M. Van Der Mark, M. van Albada, and A. Lagendijk, "Light scattering in strongly scattering media:
281 Multiple scattering and weak localization", Physical Review B 37(7), 3575 (1988).
7
282 T.R. Hughes, The Finite Element Method: Linear Static and Dynamic Finite Element Analysis (Dover
283 Publications, Mineola, NY, 1987) pp.1-676.
284 8
F. Ihlenburg and I. Babuška, "Finite element solution of the Helmholtz equation with high wave
285 number Part I: The h-version of the FEM", Computers & Mathematics with Applications 30(9), 9-37
286 (1995).
9
287 J.M. Carcione, G.C. Herman, and A.P.E. ten Kroode, "Seismic Modeling", Geophysics 67(4), 1304-
288 1325 (2002).
10
289 K. Kelly, R. Ward, S. Treitel, and R. Alford, "Synthetic seismograms: a finite-difference approach",
290 Geophysics 41(1), 2-27 (1976).
11
291 B. Chassignole, V. Duwig, M.-A. Ploix, P. Guy, and R. El Guerjouma, "Modelling the attenuation in
292 the ATHENA finite elements code for the ultrasonic testing of austenitic stainless steel welds",
293 Ultrasonics 49(8), 653-8 (2009).
12
294 S. Shahjahan, F. Rupin, A. Aubry, B. Chassignole, T. Fouquet, and A. Derode, "Comparison between
295 experimental and 2-D numerical studies of multiple scattering in Inconel600 by means of array probes",
296 Ultrasonics 54(1), 358-367 (2014).
13
297 N. Feuilly, O. Dupond, B. Chassignole, J. Moysan, and G. Corneloup, "Relation between ultrasonic
298 backscattering and microstructure for polycrystalline materials", Review of Progress in Quantitative
299 NDE (AIP Conference Proceedings Vol 1096) pp.1216-1223 (2009).
14
300 G. Ghoshal and J.A. Turner, "Numerical model of longitudinal wave scattering in polycrystals", IEEE
301 Trans. Ultrason. Ferroelect. Freq. Control 56(7), 1419-1428 (2009).
15
302 A. Van Pamel, P. Huthwaite, C.R. Brett, and M.J.S. Lowe, "Finite element modelling of elastic wave
303 scattering within a polycrystalline material in two and three dimensions", Journal of the Acoustical
304 Society America 138(4), 2326 (2015).
16
305 A. Van Pamel, P. Huthwaite, C.R. Brett, and M.J.S. Lowe, "Numerical simulations of ultrasonic array
306 imaging of highly scattering materials", NDT&E International 81, 9-19 (2016).
17
307 A. Frankel and R.W. Clayton, "Finite-difference simulation of wave propagation in two-dimensional
308 random media", Bulletin of the Seismological Society of America 74(6), 2167-2186 (1984).
18
309 A. Frankel and R. Clayton, "Finite difference simulations of seismic scattering: Implications for the

16
 Dimensionality of Elastic Wave Scattering

310 propagation of short‐period seismic waves in the crust and models of crustal heterogeneity", Journal of
311 Geophysical Research 91(B6), 6465-6489 (1986).
19
312 M. Roth and M. Korn, "Single scattering theory versus numerical modelling in 2-D random media",
313 Geophysical Journal International 12(1), 124-140 (1993).
20
314 J. Virieux, "P-SV wave propagation in heterogeneous media: Velocity-stress finite-difference
315 method", Geophysics 51(4), 889-901 (1986).
21
316 F.E. Stanke and G.S. Kino, "A unified theory for elastic wave propagation in polycrystalline
317 materials", The Journal of the Acoustical Society of America 75(3), 665-681 (1984).
22
318 W.P. Mason and H.J. McSkimin, "Attenuation and Scattering of High Frequency Sound Waves in
319 Metals and Glasses", The Journal of the Acoustical Society of America 19(3), 464 (1947).
23
320 J.W. Strutt, "On the light from the sky its polarization and colour", Philosophical Magazine 41(272),
321 107-120 (1871).
24
322 C.F. Ying and R. Truell, "Scattering of a Plane Longitudinal Wave by a Spherical Obstacle in an
323 Isotropically Elastic Solid", Journal of Applied Physics 27(9), 1086 (1956).
25
324 P. Sheng, Introduction to Wave Scattering, Localization and Mesoscopic Phenomena (Academic
325 Press, New York, 1995) pp.1-329.
26
326 J.W.S. Rayleigh, The Theory of Sound (Macmillan, London, 1877) pp.1-370.
27
327 H.B. Huntington, "On Ultrasonic Scattering by Polycrystals", The Journal of the Acoustical Society
328 of America 22(3), 362 (1950).
28
329 C. Eckart, "The scattering of sound from the sea surface", The Journal of the Acoustical Society of
330 America 25(3), 566-570 (1953).
29
331 P.B. Nagy and L. Adler, "Surface roughness induced attenuation of reflected and transmitted
332 ultrasonic waves", Journal of the Acoustical Society of America 82(1), 193-197 (1987).
30
333 K. Mampaert and O. Leroy, "Reflection and transmission of normally incident ultrasonic waves on
334 periodic solid–liquid interfaces", Journal of the Acoustical Society of America 83(4), 1390 (1988).
31
335 J.A. Turner, "Scattering and diffusion of seismic waves", Bulletin of the Seismological Society of
336 America 88(1), 276-283 (1998).
32
337 P. Zhang, D. Balint, and J. Lin, "An integrated scheme for crystal plasticity analysis: Virtual grain
338 structure generation", Computational Materials Science 50(10), 2854-2864 (2011).
33
339 K. Zhang, B. Holmedal, O.S. Hopperstad, S. Dumoulin, J. Gawad, A. Van Bael, and P. Van Houtte,
340 "Multi-level modelling of mechanical anisotropy of commercial pure aluminium plate: Crystal
341 plasticity models, advanced yield functions and parameter identification", International Journal of
342 Plasticity 66, 3 (2015).
34
343 S. Dancette, L. Delannay, K. Renard, M.A. Melchior, and P.J. Jacques, "Crystal plasticity modelling
344 of texture development and hardening in TWIP steels", Acta Materialia 60(5), 2135-2145 (2012).
35
345 R. Quey, P.R. Dawson, and F. Barbe, "Large-scale 3D random polycrystals for the finite element
346 method: Generation, meshing and remeshing", Computer Methods in Applied Mechanics and
347 Engineering 200(17-20), 1729-1745 (2011).
36
348 M.B. Drozdz, "Efficient Finite Element Modelling of Ultrasound Waves in Elastic Media", Imperial
349 College London Mechanical Engineering PhD Thesis (2008) pp.1-220.
37
350 P. Huthwaite, "Accelerated finite element elastodynamic simulations using the GPU", Journal of
351 Computational Physics 257, 687-707 (2014).

17
 Dimensionality of Elastic Wave Scattering

38
352 A. Kalashnikov and R.E. Challis, "Errors and Uncertainties in the Measurement of Ultrasonic Wave
353 Attenuation and Phase Velocity", IEEE Trans. Ultrason. Ferroelect. Freq. Control 52(10), 1754-1768
354 (2005).
39
355 E.E. Underwood, Quantitative Stereology (Addison-Wesley, Reading, 1970) pp.1-274.

356

18