Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences
You have accessResearch article

Flexural edge waves generated by steady-state propagation of a loaded rectilinear crack in an elastically supported thin plate

Andrea Nobili

Andrea Nobili

Dipartimento di Ingegneria Enzo Ferrari, Università degli Studi di Modena e Reggio Emilia, via Vivarelli 10, 41125 Modena, Italy

[email protected]

Google Scholar

Find this author on PubMed

,
Enrico Radi

Enrico Radi

Dipartimento di Scienze e Metodi dell’Ingegneria, Università degli Studi di Modena e Reggio Emilia, via Amendola 2, 42122 Reggio Emilia, Italy

Google Scholar

Find this author on PubMed

and
Luca Lanzoni

Luca Lanzoni

Dipartimento di Economia e Tecnologia, Università degli Studi della Repubblica di San Marino, Via Salita alla Rocca 44, 47890 San Marino, Republic of San Marino

Google Scholar

Find this author on PubMed

    Abstract

    The problem of a rectilinear crack propagating at constant speed in an elastically supported thin plate and acted upon by an equally moving load is considered. The full-field solution is obtained and the spotlight is set on flexural edge wave generation. Below the critical speed for the appearance of travelling waves, a threshold speed is met which marks the transformation of decaying edge waves into edge waves propagating along the crack and dying away from it. Yet, besides these, and for any propagation speed, a pair of localized edge waves, which rapidly decay behind the crack tip, is also shown to exist. These waves are characterized by a novel dispersion relation and fade off from the crack line in an oscillatory manner, whence they play an important role in the far field behaviour. Dynamic stress intensity factors are obtained and, for speed close to the critical speed, they show a resonant behaviour which expresses the most efficient way to channel external work into the crack. Indeed, this behaviour is justified through energy considerations regarding the work of the applied load and the energy release rate. Results might be useful in a wide array of applications, ranging from fracturing and machining to acoustic emission and defect detection.

    1. Introduction

    Crack propagation in elastically supported thin structures made of brittle or quasi-brittle material is a common feature of many natural phenomena, such as ice fracturing and calving, rock fault planes and layered material failure, road pavement deterioration, surface coating detachment, to name only a few. In all instances, cracking takes place at the expense of the stored elastic energy, which is rapidly converted into stress waves travelling in the material and providing the so-called acoustic emission (AE). AE can be measured in the far-field and it lends a convenient indirect means to access the internal change in the material status. Besides travelling waves, moving in the bulk of the material, edge waves are usually excited and occur in a localized region near the boundaries [1,2]. Edge waves tend to appear at smaller speed than travelling waves, owing to their lower energy content, and consequently are detected first [3,4]. Furthermore, edge waves are closely related to edge buckling [5].

    When an external load moves on a thin structure, its speed might easily approach some resonant speed and produce dramatic effects [6]. This outcome is further enhanced by the presence of cracks in the structure. An intriguing example of this is the discovery that a ground effect machine may be successfully employed as ice breaker when operated at the system’s critical speed [7]. The analysis of the effect of loads moving on elastic structures has been a long-standing subject of investigation, in the light of its many practical implications. Historically, much of this analysis has been directed by the desire to safely design bridges, rail tracks and road pavements under the ever-increasing demand of high-speed high-capacity transportation [8,9]. Recently, renewed interest has been drawn to model and design floating ice sheets as supporting structures for oil rigs, pipes, roads, runways and platforms [7]. Climate change and extensive investigation of the interaction between ice-shelf cracking and impinging sea-waves, in a process somewhat similar to that leading to edge waves excited by deep water surface waves [10], are also motivating further research in the field [1113].

    Remarkably, despite the broad interest and the wide range of application, only a handful of contributions may be found in the literature concerning fracture dynamics in elastically supported thin plates. The static solution to this problem was first considered in [14] and it was later extended in [15,16] to a weakly non-local foundation. In the classic references [1719], several static problems for finite cracks in unsupported plates or shells are considered. The mathematical problem is related to that appearing in the study of crack propagation in couple-stress materials [20,21], although different boundary conditions (BCs) apply. In all cases, the combined effect of a moving load acting on a moving crack has never been investigated.

    In this paper, the steady-state propagation of a rectilinear crack in a thin plate resting on a Winkler foundation and subject to a moving harmonic load applied at the crack flanks is considered. The spotlight is set on flexural edge waves generation as a result of this combination. In particular, it is observed that, compared with the classic subject of edge wave propagation at the boundary of a semi-infinite plate [22,23], a new edge wave arises out of the fact that propagation is restricted to the crack flanks. Besides, thorough investigation of the stress intensity factors reveals that the loading frequency may be tuned so as to either promote or hinder crack propagation in practical applications (consider, for instance, ice breaking as opposed to road pavement preservation). Finally, this solution may be used as a building block to tackle, through superposition, the problem of a general distributed load in steady-state motion on a cracked plate, where the load application is no longer restricted to the crack flanks, although it still moves with the crack tip.

    The paper is organized as follows: §2 formulates the problem, which is then recast in the frequency domain in §3. The full-field solution is given in §4 and §5 discusses stress intensity factors (SIFs). Energy considerations supporting the non-monotonic behaviour of the SIFs are given in §6 along with the energy release rate (ERR) at the crack tip. Flexural edge wave solutions are considered in §7 and conclusions are drawn in §8. Finally, the electronic supplementary material presents the derivation of a conservative line integral, which extends to elastically supported thin plates the analogous result obtained in [24] for steady-state crack propagation in rate-dependent plastic solids.

    2. Problem formulation

    (a) Field equations

    We consider a semi-infinite rectilinear crack propagating along its length in an infinite Kirchhoff–Love (K-L) thin elastic plate (figure 1). The plate, of thickness h, is elastically supported by a Winkler foundation with stiffness k. A moving Cartesian reference frame, (ξ^1,ξ^2,ξ^3), is attached to the crack tip such that the linear crack corresponds to the negative part of the ξ^1-axis, while the ξ^2-axis measures the distance from the crack line of a point on the plate. The crack is propagating at constant speed c (steady-state propagation) with respect to a fixed reference frame (x1,x2,x3). In this fixed frame, the governing equation for the transverse displacement of the plate, w, reads [7], §4.3

    Dw+kw=ρhttw+q,2.1
    being △=∂x1x1+∂x2x2 the Laplace operator in two dimensions, q the transverse distributed load per unit area, D the plate bending stiffness and ρ the mass density per unit volume. The double Laplace operator is usually named the biharmonic operator and it is denoted by ∇4.
    Figure 1.

    Figure 1. Rectilinear crack propagating at constant speed c in an elastic thin plate resting on a Winkler elastic foundation.

    As is customary in a steady-state analysis, we set ourselves in the constant speed moving frame (ξ^1,ξ^2,ξ^3), with ξ^1=x1ct, and assume w=w(ξ^1,ξ^2), which bears no explicit time dependence. Then, equation (2.1) may be rewritten as

    ^4w+κ2ξ^1ξ^1w+λ4w=qD,2.2
    having let the characteristic lengths
    λ=Dk4andκ=c1Dρh,
    together with the positive dimensionless ratio
    η=λκ=cρhkD4.

    It is worth observing that equation (2.2) corresponds to the governing equation for a supported thin elastic plate subject to an axial compression of magnitude −2 [5] or for an unsupported cylindrical shell [14,25]. We introduce the dimensionless coordinates (ξ1,ξ2)=λ1(ξ^1,ξ^2) and take q≡0, with no loss of generality. Then, equation (2.2) becomes

    4w+η2ξ1ξ1w+w=0,2.3
    where ∇4=(∂ξ1ξ1+∂ξ2ξ2)2 is the biharmonic operator in dimensionless coordinates. The special case η=0 corresponds to the static problem, whose solution is given in [14]. Furthermore, we recall that, in a supported thin elastic plate, travelling wave solutions are admitted beyond a critical speed [7], §4.2
    ccr=2kD4ρhη=ηcr=2,2.4
    corresponding to the group speed and occurring at the dimensionless wavenumber
    μ=μcr=1.

    (b) Boundary conditions

    Within the K-L theory, the bending moment, twisting moment and the equivalent shearing force are given by, respectively, [3]

    m22=(ξ2ξ2+νξ1ξ1)w,2.5a
    m12=ν0ξ1ξ2w,2.5b
    andv2=λ1ξ2[ξ2ξ2+(2ν)ξ1ξ1]w,2.5c
    all of them having being deprived of the common factor −2 and having let the shorthand notation ν0=1−ν. Besides, we let the slope
    ϕ=λ1ξ2w.

    The BCs across the crack line ξ2=0 and ahead of the crack tip, i.e. for ξ1>0, are

    • — of kinematic nature, expressing continuity of displacement and slope,

      w0(ξ1)=ϕ0(ξ1)=0,ξ1>0,2.6

    • — of static nature, demanding continuity for the bending moment and the equivalent shearing force,

      m0(ξ1)=v0(ξ1)=0,ξ1>0,2.7
      where m0 and v0 are shorthands for m220 and v20, respectively.

    Here, f(0) denotes the jump of the function f(ξ2) across the crack line, namely f(0+)−f(0), while a zero subscript means evaluation at the crack line, i.e. w0(ξ1)=w(ξ1,0).

    As is well known, the solution of any linear fracture mechanics problem under general loading conditions may be obtained from the superposition of two simpler set-ups: the first set-up is obtained disregarding the crack and considering the given loading condition, the second set-up takes into account the presence of the crack, which is loaded by the force distribution found at the previous set-up. The latter problem may be further decomposed through harmonic expansion of the crack loading. In this paper, we are interested in investigating the effect of the crack and, consequently, only the second set-up will be considered. Indeed, it is assumed that the crack flanks are loaded in a continuous fashion by a general harmonic term. Then, the BCs at the crack line ξ2=0 are of static nature, namely

    m22(ξ1,0±)=M0 exp(ıaξ1)andv2(ξ1,0±)=λ1V0 exp(ıaξ1),ξ1<0,2.8
    where M0=M0(a), V 0=V 0(a) are complex-valued and the dimensionless forcing frequency (in space) satisfies
    (a)<02.9
    to ensure decay as ξ1. Hereinafter, ı is the imaginary unit, ℜ(s) and ℑ(s) denote the real and the imaginary part of sC, respectively, and a superscript asterisk denotes complex conjugation, i.e. s*=ℜ(s)−ıℑ(s). We observe that equations (2.8), just like (2.6) and (2.7), entail four conditions in total, for they apply at both flanks of the crack (denoted by ξ2=0±). As a consequence of loading continuity, equations (2.7) hold on the entire crack line.

    3. Analysis in the frequency domain

    Let us define the bilateral (or full) Fourier transform of w(ξ1,ξ2) along ξ1 in the usual way [26]

    w¯(s,ξ2)=+w(ξ1,ξ2) exp(ısξ1)dξ1.

    In a similar fashion, the unilateral (or generalized, or half-range) transforms are introduced. The plus transform is defined as

    w¯+(s,ξ2)=0+w(ξ1,ξ2) exp(ısξ1)dξ1
    and it is analytic in the complex half-plane ℑ(s)>α1, provided that α1R exists such that w(ξ1,ξ2) exp(α1ξ1) is absolutely integrable with respect to ξ1 in the interval (0,+). Likewise, the minus transform of w
    w¯(s,ξ2)=0w(ξ1,ξ2) exp(ısξ1)dξ1
    is analytic in the complex half-plane ℑ(s)<α2 provided that α2 can be found such that w(ξ1,ξ2) exp(α2ξ1) is absolutely integrable with respect to ξ1 in the interval (,0). Consequently, assuming α1<0<α2, the bilateral Fourier integral is related to the unilateral transforms through the connection
    w¯(s,ξ2)=w¯+(s,ξ2)+w¯(s,ξ2),3.1
    valid in the strip of analyticity S={sC:α1<(s)<α2} containing the real axis. In this strip, the inverse of the bilateral Fourier transform may be defined as:
    w(ξ1,ξ2)=12π+w¯(s,ξ2) exp(ısξ1)ds3.2
    and similarly for the inverse of the half-transforms.

    Taking the Fourier transform of equation (2.3) in the ξ1 variable, a linear constant coefficient ODE is obtained whose general solution is

    w¯(s,y)=A1 exp(Λ1|ξ2|)+B1 exp(Λ1|ξ2|)+A2 exp(Λ2|ξ2|)+B2 exp(Λ2|ξ2|),3.3
    wherein
    Λ1,2=s2R(s)andR(s)=η2s21.3.4

    We observe that the factorization holds

    Λ1Λ2=s4η2s2+1=(s2β2)(s2β2),β=η22+ı1η44,
    in which the roots ±ββ* are the branch points of the double square roots in (3.3). It is further observed that β is a complex number with unit modulus located in the first quadrant of the complex plane inasmuch as 0≤η<ηcr, while it sits on the real axis for ηηcr. For a strip of analyticity about the real axis to be accessible and the inverse Fourier transform (3.2) meaningful, we need to have
    η<ηcr.3.5

    This constraint amounts to requiring c<ccr, where ccr is the critical speed (2.4).

    Let w(ξ1,ξ2+) and w(ξ1,ξ2) be the restrictions of the displacement w(ξ1,ξ2) in the upper and in the lower half of the (ξ1,ξ2)-plane, respectively, where it is understood that ξ2+(0,+) and ξ2(,0). The general solution of the ODE (2.3), bounded at infinity, retains only the A-terms,

    w¯(s,ξ2±)=A1± exp(Λ1|ξ2±|)+A2± exp(Λ2|ξ2±|),3.6
    where A1± and A2± are four complex-valued functions of s to be determined. The square root in (3.6) is made defined by choosing the Riemann sheet such that (Λ1,2)>0. To this aim, we locate the cuts according to the requirements
    (Λ1)<0and(Λ1)=0,3.7
    where the equality sets the cuts position, while the inequality warrants the proper orientation, see [3], §6.2.2 for more details. It is observed that adopting Λ2 in (3.7) would lead to the same cut location in the light of the two-valued nature of the square root R(s). Interestingly, the cut location is independent of ν. We assume that A1± and A2± split into a symmetric and skew-symmetric part [15]:
    Ai±=12(A¯i±ΔAi),i=1,2,3.8
    whence we have, for the restriction of (3.6) onto the ξ1-axis,
    w¯0±(s)=A1±+A2±.3.9

    Equation (2.6) may be written in terms of plus Fourier transforms

    w¯0+=ϕ¯0+=0,
    whence, using the general solution (3.6) and in view of equations (3.1), (3.8), (3.9), it is
    ΔA1+ΔA2=w¯03.10a
    andΛ1A¯1+Λ2A¯2=λϕ¯0.3.10b

    Likewise, the unilateral Fourier transform of equation (2.8) gives

    m¯0=ıM0s+aandv¯0=ıλ1V0s+a,3.11
    valid in the semi-infinite region S1={s:(s)<(a)}, which, by the inequality (2.9), contains the real axis. It is observed that M0 and V 0 have dimension of length. Taking the full Fourier transform of the bending moment (2.5a) and of the shearing force (2.5c), we get
    m¯=(ξ2ξ2νs2)w¯andv¯=λ1ξ2[ξ2ξ2(2ν)s2]w¯,
    which, employing the solution (3.6) and using equations (3.1), (3.8), (3.11), gives
    [(Λ1νs2)A1±+(Λ2νs2)A2±]=m¯0+ıM0s+a3.12a
    and±λ1{Λ1[Λ1(2ν)s2]A1±+Λ2[Λ2(2ν)s2]A2±}=v¯0+ıλ1V0s+a3.12b
    in the strip S0=SS1. Finally, Fourier transformation of equations (2.7), which really hold on the entire crack line, gives
    (Λ1νs2)ΔA1+(Λ2νs2)ΔA2=03.13a
    andΛ1[Λ1(2ν)s2]A¯1+Λ2[Λ2(2ν)s2]A¯2=0,3.13b
    according to which equations (3.12) reduce to
    12[(Λ1νs2)A¯1+(Λ2νs2)A¯2]=m¯0+ıM0s+a3.14a
    and12{Λ1[Λ1(2ν)s2]ΔA1+Λ2[Λ2(2ν)s2]ΔA2}=λv¯0+ıV0s+a.3.14b

    Conditions (3.13) are immediately fulfilled through letting

    ΔA1=(Λ2νs2)ΔA,3.15a
    ΔA2=(Λ1νs2)ΔA,3.15b
    A¯1=Λ2[Λ2(2ν)s2]A¯,3.15c
    andA¯2=Λ1[Λ1(2ν)s2]A¯,3.15d
    whence equations (3.10) become
    (Λ1Λ2)ΔA=w¯0,3.16a
    andΛ1Λ2(Λ1Λ2)A¯=λϕ¯0.3.16b

    Similarly, equations (3.14) give

    (Λ2Λ1)K(s)A¯=m¯0+ıM0s+a,3.17a
    and(Λ2Λ1)K(s)ΔA=λv¯0+ıV0s+a,3.17b
    where the kernel function K(s) is let as follows:
    2(Λ2Λ1)K(s)=Λ1[Λ1(2ν)s2](Λ2νs2)+Λ2[Λ2(2ν)s2](Λ1νs2).3.18

    In particular, in the limit as η0, the function 4ıK(s) in equation (3.18) reduces to the kernel (24) of [14]. Solving the system (3.16), which is linear in the unknown functions A¯,ΔA and plugging the result in equations (3.17) provides the following pair of uncoupled inhomogeneous Wiener–Hopf (W-H) equations

    K(s)w¯0+λv¯0+=ıV0s+a,3.19a
    and(Λ1Λ2)1/2K(s)λϕ¯0+m¯0+=ıM0s+a.3.19b

    4. Full-field solution

    The kernel K(s) is an even function of s and it possesses six roots all of which are, in the general case, of order unity (figure 2)

    K(s)=0s=±s1,±s1,±ır1.4.1
    Figure 2.

    Figure 2. Branch cuts (dashed-dotted curves), branch points (solid dots) and zeros (circles) in the complex plane for the kernel function K(s) in the regime η<ηe. Branch points and zeros are located on a circle (dotted) of radius 1 and γe1, respectively. The strip of analyticity S is also shown, together with the loading frequency a (cross).

    Here, s1 is taken to sit in the first quadrant of the complex plane,

    s1=γe1(ηηe)2+ı1(ηηe)4,ηe=2γe,
    having let γe=(1ν)(3ν1+22ν22ν+1)4, which is a well-known bending edge wave constant [22,23]. We observe that γe[2(52)1/4,1] and its maximum γe=1 is attained at ν=0. It is easy to see that s1 is complex-valued for η<ηe and it sits in the first quadrant of the complex plane on the circle of radius γe1. It can be proved that ℑ(s1)>ℑ(β) for η<1, almost irrespectively of ν. According to the conditions (3.7), the double roots in the kernel K(s) bring in two non-straight branch cuts which extend from s=β,β* to −β,−β* through ı, respectively. It is remarked that, for ν=0, it is γe=1 and the roots ±s1,±s1 coincide with the branch points for the double roots ±ββ*, whence their order goes down to 1/2. Besides, in this case, ηe=ηcr.

    In a similar fashion, we find the location of the purely imaginary roots s=±ır1, being

    r1=γm1(ηηm)2+(ηηm)4+1,ηm=2γm
    and we have defined the new constant
    γm=(1ν)(3ν+1+22ν22ν+1)4.

    We note that γm[(221)1/4/2,2(5+2)1/4] is a monotonic decreasing function of ν (figure 3). Conversely, r1 is a real-valued monotonic increasing function of η, whose minimum r1=γm1 is attained in the static case η=0, i.e. unlike s1, this root never reaches the real axis. Indeed, in the special case η=0 (stationary crack), we have

    r1=γm1ands1=γe122(1+ı).
    Figure 3.

    Figure 3. γm (solid, black) and γe (dashed, blue) as a function of Poisson ratio ν. (Online version in colour.)

    This situation is considered in [14], where the roots ±ır1 seem to have gone amiss. By contrast, for ηeη<ηcr, s1 turns real-valued and the root landscape (4.1) switches to (figure 4)

    ±s,±s+,±ır1,
    where now 0<s<γe1<s+ and
    s=γe1(ηηe)2(ηηe)41.
    Figure 4.

    Figure 4. Branch cuts (dash-dot curves), branch points (solid dots) and zeros (circles) in the complex plane for the kernel function K(s) in the regime ηeη<ηcr. The strip of analyticity S is also shown as well as the loading frequency a (cross).

    In this case, the strip of analyticity S is taken to warrant the radiation (or Sommerfeld) condition of energy flowing from the load application zone to ξ1.

    Let us define, for d=ν0(3+ν)/4 and for any chosen value b>α1,2=min{(s1),(β),r1},

    F(s)=s2+b2s2β2s2β2d(s2s12)(s2s12)(s2+r12)K(s).4.2

    The function F(s) is even, deprived of zeros in a semi-infinite strip of analyticity S, and it is such that lim|s|F(s)=1 in this strip. For such F(s), the W-H logarithmic factorization [26], §3.2 is applicable and it gives a plus and a minus function, respectively, denoted by F+(s) and F(s), with the properties (see [27] for more details)

    F(s)=F+(s)F(s)andF(s)=F+(s).

    Accordingly, system (3.19) reads

    K(s)w¯0+λv¯0+K+(s)=ıV01(s+a)K+(s),4.3a
    andK(s)sβs+βλϕ¯0+s+βsβK+(s)m¯0+=ıM0s+βsβ(s+a)K+(s),4.3b
    where
    K±(s)=d(s±s1)(ss1)(s±ır1)s±ıbs±βsβF±(s)±ı,
    with the properties K(s)=K+(s)K(s) and K(s)=K+(−s). Here, it is understood that ±ı= exp(±ıπ/4). For large values of |s|, we observe the asymptotic behaviour K±(s)∼s3/2. Finally, the r.h.s. in equation (4.3) are split in terms of the sum of a plus and a minus function
    λv¯0+K+(s)ıV0s+a(1K+(s)1K(a))=ıV0(s+a)K(a)K(s)w¯0,4.4a
    ands+βsβK+(s)m¯0+ıM0s+a(s+βsβK+(s)aβa+βK(a))=ıM0s+aaβa+βK(a)K(s)sβs+βλϕ¯0.4.4b

    As, in system (4.4), the left (right) hand is represented by a function which is analytic in the upper (lower) complex half-plane with a common strip of regularity S0, it can be analytically continued into the whole complex plane. Indeed, continuation brings in two entire functions, E1(s) and E2(s), which are holomorphic over the whole complex plane. Appealing to Liouville’s theorem, it is E1(s)≡E2(s)≡0, in the light of the fact that the behaviour for large |s| of either hands in equation (4.4) decays at least as fast as s−1. Thus,

    w¯0=ıV0K(a)1(s+a)K(s)4.5
    andλϕ¯0=ıM0aβa+βK(a)sβs+β(s+a)K(s).4.6

    Expressions for the unilateral Fourier transform of the bending moment and of the shearing force on the crack line follow immediately:

    m¯0+=ıM0s+a(1aβa+βs+βsβK+(s)K(a))4.7a
    andλv¯0+=ıV0s+a(1K+(s)K(a)).4.7b

    It is observed that, according to Jordan’s lemma [26], equations (4.5) and (4.6) satisfy both BCs (2.6) and, by the same argument, equations (4.6a) and (4.6b) imply the conditions (2.7).

    Equations (3.6), (3.8), (3.15), (3.16) allow writing the full Fourier transform of the plate deflection

    w¯(s,ξ2±)=12(Λ2(2ν)s2Λ1(Λ1Λ2)λϕ¯0Λ2νs2Λ1Λ2w¯0)e1(s,ξ2±)12(Λ1(2ν)s2Λ2(Λ1Λ2)λϕ¯0Λ1νs2Λ1Λ2w¯0)e2(s,ξ2±),
    having let the shorthand notation e1,2(s,ξ2)= exp(Λ1,2|ξ2|). In particular, on the crack line, it is e1,2(s,0)=1, whence
    w¯0±(s)=(ν0s2R(s)s2R(s)ν0s2+R(s)s2+R(s))λϕ¯04R(s)±12w¯0.

    5. Stress-intensity factors

    Stress-intensity factors (SIFs) can be determined from the behaviour of the relevant Fourier transform for large |s|. Indeed, equation (4.6a) gives

    m¯0+ıM0aβa+βK(a)ıds1/2,
    whence, making use of the connection between the asymptotic expansion of a function and the expansion of its Fourier half-transform [26], §2.14.B, we get
    m0ıM0aβa+βΓ(1/2)K(a)ıdξ1,asξ10+.

    By the definition of the stress intensity factor [17] and recalling that σ22=6 h−2m22 times the omitted term −2, we find

    k^1=limξ10+2ξ1σξ2=6ıDM0λ2h2aβa+βK(a)2ıdπ.5.1

    The modulus of the dimensionless stress intensity factor k1=(λh)2k^1/(M0D) is plotted in figure 5 at fixed ν=0.25 as a function of the loading frequency. As expected, |k1| asymptotes to zero as a−1/2 yet, remarkably, its decay is monotonic only for small speed η. Indeed, for η close to the critical speed, it displays an absolute maximum for ℜ(a) near 1, which becomes greater as ηηcr. The same resonant behaviour appears in figure 6, which shows the dependence of |k1| from the crack speed η. The role of ν is illustrated in figure 7 according to which resonance is stronger near the ends of the admissible range ν[1,12].

    Figure 5.

    Figure 5. Modulus of the dimensionless stress intensity factors |k1| (a) and |k3| (b) versus ℜ(a) for ℑ(a)=−0.1ı, ν=0.25 and η=0.5 (solid, black), 1(dashed, blue) and ηcr (dotted, red). (Online version in colour.)

    Figure 6.

    Figure 6. |k1| (a) and |k3| (b) versus speed η for ℑ(a)=−0.1ı, ν=0.25 and ℜ(a)=0.25 (solid, black), 0.5 (dashed, blue) and 1 (dotted, red).(Online version in colour.)

    Figure 7.

    Figure 7. |k1| versus ℜ(a) for ℑ(a)=−0.1ı, ν=−1 (solid, black), 0 (dashed, blue), 0.5 (dotted, red) and η=1 (a), η=ηcr (b). (Onlineversion in colour.)

    Along the same line, it is easy to observe that, for large |s|, equation (4.6b) gives

    λv¯0+ıV0ıdK(a)s1/2,
    whence we get
    λv0ıV0ıdΓ(1/2)K(a)ξ13/2,
    and, therefore, in the light of the connection σ23=32h1v2,
    k^3=limξ10+ξ13/2σ23=34ıDV0λ3hıdπK(a).5.2

    As shown in figures 5 and 6, the behaviour of k3=λ3h/DV 0 is similar to that of k1, although it asymptotes to zero faster, as |a|−3/2. Figure 8 brings along the role of ν at η=1 and η=ηcr. In general, compared with |k1|, |k3| appears much smaller.

    Figure 8.

    Figure 8. |k3| versus ℜ(a) for ℑ(a)=−0.1ı, ν=−1 (solid, black), 0 (dashed, blue), 0.5 (dotted, red) and η=1 (a), η=ηcr (b). (Onlineversion in colour.)

    Determination of k2 requires dealing with the asymptotics of a full Fourier transform. Indeed, Fourier transformation of the twisting moment (2.5b) gives

    m¯12=ıν0sξ2w¯=ıν0s[Λ1A1±e1(s,ξ2)+Λ2A2±e2(s,ξ2)]=±ıν0s2(Λ1Λ2){[(Λ2(2ν)s2)λϕ¯0±Λ1(Λ2νs2)w¯0]e1(s,ξ2)[(Λ1(2ν)s2)λϕ¯0±Λ2(Λ1νs2)w¯0]e2(s,ξ2)},
    which, on the crack line, reduces to
    m¯120=ıν02s(λϕ¯0Λ1(Λ2νs2)Λ2(Λ1νs2)Λ2Λ1w¯0).

    It is observed that the first term in parenthesis in this equation is a minus function, which brings no contribution for ξ1>0. Conversely, the second term is analytic in a semi-infinite strip around the real axis and it is neither plus nor minus. Some straightforward asymptotic analysis of equation (4.5) gives, for large |s|,

    w¯0ıV01ıdK(a)s5/2,
    whence
    m¯12014V01ν2ıdK(a)s1/2,as|s|+.

    As the asymptotics of the full Fourier transform is available, we employ Abel’s theorem [28] to get

    m120=14V02π(1ν2)ıdK(a)ξ11/2,asξ10+,
    whence
    k^2=limξ10+2ξ1σ12=3DV0λ2h2π(1ν2)ıdK(a),5.3
    which is proportional to k3. It is remarked that SIFs have been determined within the framework of the K-L theory, which neglects shear deformation.

    6. Stored energy and energy release rate

    The resonant behaviour of the SIFs can be most easily explained evaluating the energy fed into the system by the applied load, which, according to Betti’s theorem, is given by

    W=12Dλ20(m0ϕ0+v0w0)λdξ1=12Dλ20(M0λϕ0+V0w0) exp(ıaξ1)dξ1=14ıDλ2(M02W1+V02W2),
    where
    W1=(aβ)(a+β)aK(a)2andW2=1aK(a)2.

    The dimensionless energies introduced by the bending moment and by the shearing force, respectively, W1 and W2, are plotted in figure 9 as a function of the applied load frequency a. It appears that a local maximum in the energy input occurs for a≈1, which is responsible for the non-monotonic behaviour of the SIFs. Similarly, figure 10 presents W1 and W2 as a function of the crack propagation speed η.

    Figure 9.

    Figure 9. Bending moment dimensionless energy input |W1| (a) and shearing force dimensionless energy input |W2| (b) versus ℜ(a) for ℑ(a)=−0.1ı, η=ηcr and ν=−1 (solid, black), 0 (dashed, blue) and 0.5 (dotted, red). (Online version in colour.)

    Figure 10.

    Figure 10. Bending moment dimensionless energy input |W1| (a) and shearing force dimensionless energy input |W2| (b) versus η for ℑ(a)=−0.1ı, ν=0.25 and ℜ(a)=0.25 (solid, black), 0.5 (dashed, blue) and 1 (dotted, red). (Online version in colour.)

    A more rigorous argument pertains to the ERR at the crack tip, Gtip, which may be determined through the conservative integral I. A derivation of this integral in the case of elastically supported thin plates is presented in the electronic supplementary material (see also [29]). To relate I to the near-tip fields, we consider a suitable contour Γ constituted by a vanishingly thin rectangular box, centred at the crack tip, with sides 2δ1≪2δ2 parallel to the ξ1 and ξ2 axes, respectively, by the crack flanks Γ+ and Γ and finally closed by the far-field circle ΓR, with radius R (figure 11). As we shrink the rectangular box down to the crack tip, i.e. δ1,20+, and simultaneously let R+, we get

    I=limδ1,20+Γ+(m22ξ1ϕ2+q2ξ1w+m12ξ1ϕ1)λdξ1+Gtip=0,
    in the light of the fact that limR+ΓR=0 and Gtip is the contribution of the small box centred at the crack tip. Taking the limit of the integral term and part integrating twice leads to
    Gtip=0(m22ξ1ϕ2+v2ξ1w)λdξ1=Dλ20ξ1(M0λϕ0+V0w0) exp(ıaξ1)dξ1=ıDλ2a0(M0λϕ0+V0w0) exp(ıaξ1)dξ1=2ıaW,
    for boundary terms vanish and recalling the definition of the Kirchhoff equivalent shearing force v2=q2+∂ξ1m12. It is concluded that, for steady-state crack propagation, the ERR, which is readily seen to be proportional to the SIFs squared, may be related to the work of the applied loads. Therefore, the resonant behaviour displayed by the SIFs represents the most efficient way to channel energy from the applied load to the crack tip and thus promote fracturing.
    Figure 11.

    Figure 11. Integration path for evaluation of Gtip.

    7. Edge waves at the crack flanks

    In this section, we provide some physical insight into the afore-obtained results. Looking for solutions of equation (2.3) in the form of a edge wave [23] for, say, the top half-plate,

    w(ξ1,ξ2+)= exp[μ(ıξ1+ζξ2+)]7.1
    and, plugging this expression in the governing equation (2.3), we obtain, for the attenuation index,
    ζ1,2=1±η2μ21μ2.

    The sign for ζ is chosen so as to warrant decay as ξ2+, i.e. it is such that ℜ(μζ)>0. Indeed, μ is generally a complex number such that

    (μ)>0.7.2

    Consideration of homogeneous BCs gives the dispersion relation

    (1ν2)μ4+(2ν01η2μ2+μ4η2)μ2+1=0,
    which, upon rationalizing, gives the solution curve
    η2=γe4μ2+μ2.7.3

    This quadratic equation in μ2 is plotted in figure 12 and it possesses two positive real roots provided that the moving frame speed η exceeds the threshold speed ηeηcr. This threshold corresponds to the minimum speed of the phase velocity (see [23]) and it occurs at the dimensionless wavenumber μ=γe11. For η<ηe, μ is complex-valued and we find a pair of decaying edge waves whose wavenumber corresponds to the single complex solution of (7.3) complying with (7.2) and whose attenuation indexes are given by the pair of positive real values

    ζ1,2=1±γe2,7.4
    irrespectively of the wavenumber. This solution exists only along the crack flanks, i.e. behind the propagating crack tip, and for it the sign of ℜ(μ) is irrelevant.
    Figure 12.

    Figure 12. Edge wave dispersion curves (a) and decaying edge wave dispersion curves (b) for ν=0.5 (solid, black), ν=0 (dashed, red), ν=−0.5 (dashed-dotted, blue). A minimum threshold speed ηe=2γe exists forpropagating edge waves to appear (here plotted as a dotted curve only for ν=0). (Online version in colour.)

    For ηηe, edge waves become propagating along the crack flanks with a pair of real wavenumbers s,s+ and the same pair of attenuation indices (7.4), for a total of four edge wave solutions. These occur at smaller speed yet lower (higher) wavenumber s (respectively, s+) than travelling wave solutions. In the special case ν=0, they occur simultaneously, for η=ηe=ηcr and equation (7.3) admits the double root μ=1 (figure 12) and the attenuation index ζ is either 2 or zero, whence only one proper edge wave solution really exists, the other solution corresponding to a travelling wave.

    In the frequency domain analysis, edge wave solutions are closely related to the complex root s1, which expresses the wavenumber μ (as already remarked, for this class of solutions, the sign of ℜ(μ) is immaterial, which amounts to considering either s1 or s1). Indeed, this root is a pole for the minus transforms (4.5), (4.6) and, when considered in the inversion integral (3.2), it gives a solution of the form (7.1). Such solution represents decaying (along ξ1) edge vibrations, which turn into proper edge waves, propagating at ξ1, provided that ηηe. Indeed, beyond the minimum speed ηe, s1 moves onto the real axis and it separates in the pair of poles s+ and s. In this context, when ν=0, the roots s1 and s1 become of fractional order and no longer correspond to exponential solutions.

    Alongside the edge wave solution (7.1), we look for solutions in the form of exponentially decaying localized waves

    w(ξ1,ξ2+)= exp[μ(ξ1ζξ2+t)],ξ10,7.5
    assuming μ>0 in consideration of the fact that the crack flank extends along the negative ξ1-axis. Plugging this solution into the plate equation (2.3) gives, for the attenuation index, the complex-conjugated pair
    ζ1,2=1±ıη2μ2+1μ2,
    where the square root is chosen as to have ℜ(ζ1,2)>0 and, accordingly, decay as ξ2+. Consideration of load-free BCs yields a novel dispersion relation
    (1ν2)μ4+(η22ν0μ4+η2μ2+1)μ2+1=0,
    whose single positive solution is plotted in figure 12. This dispersion curve corresponds to the solution curve
    η2=γm4μ2μ2,μ>γm1.
    which, just like (7.3), describes a parabola in μ2. Clearly, for any given speed η, this equation defines two real wavenumbers, although only one of these complies with the inequality, namely
    μ=r1>γm1.

    For this wavenumber, we see that the attenuation index ζ is given by the complex-conjugated pair

    ζ1,2=1±ıγm2.7.6

    This class of edge disturbances is related to the root ır1 through Fourier inversion. Figure 13 compares the attenuation rates ℜ(μζ) of all solutions. It appears that, in the regime η<ηe, any far-field condition (i.e. at large ξ2) is satisfied by a linear combination of one decaying wave with ζ=1γe2 and one localized wave. Conversely, in the speed range ηeη<ηcr, any far-field condition is realized by a linear combination of two propagating edge waves.

    Figure 13.

    Figure 13. Attenuation rate ℜ(μζ) for edge wave solutions of the form (7.1) and ζ=1±γ22 (respectively, dashed-dotted, blue and dashed, red) compared with localized solutions (7.5) with ζ=1+ıγm2 (solid, black), at ν=0.5. In thespeed range η<ηe (a), only decaying edge waves are present, whereas in the range ηeη<ηcr (b), propagating edge waves appear and the corresponding curves bifurcate. (Online version in colour.)

    To recapitulate, three edge wave solutions exist, which correspond to the zeros of the kernel function (3.18) in the complex plane, namely

    • s1 (or −s1) corresponds to the wavenumber of a pair of decaying (along the crack flank) edge waves with real attenuation indexes, provided that the crack moving speed η rests below the threshold speed ηe;

    • — conversely, for ηeη<ηcr, s1 separates into a pair of real numbers, s≤1≤s+, that describe the wavenumbers of two pairs of propagating edge waves, with the same pair of real attenuation indexes occurring at the previous regime;

    • — for any speed, ır1 corresponds to a pair of exponentially decaying solutions, localized at the back of the crack tip, associated with the positive wavenumber r1>γm1 and with a complex-conjugated pair of attenuation indexes.

    8. Conclusion

    In this paper, the full-field solution for the steady-state propagation of a rectilinear crack in an elastically supported thin plate is given through the W-H method. A harmonic moving load is applied at the moving crack flanks. Focus is set on the analysis of flexural edge waves propagating as a result of the combined effect of crack extension and load motion. It is found that this combination brings in two regimes and three types of waves. Indeed, the solution identifies two threshold speeds, namely the critical speed ηcr, for travelling waves to appear in the bulk of the plate, and the edge wave speed ηeηcr, which corresponds to the speed of edge waves at the boundary of a semi-infinite thin plate. These two threshold speeds coincide when ν=0, for it is shown that edge waves collapse into travelling waves. When the propagation speed η is smaller than ηe, a pair of edge waves exist that decay in an oscillatory manner along the crack-flanks and rapidly fade off away from the crack line with real attenuation index. Rapid attenuation away from the crack line remains for ηηe, yet edge waves become four and they propagate indefinitely along the crack flanks. For any propagation speed, a new type of edge wave is met which is highly localized behind the crack tip. Indeed, two such waves exist which decay with the real exponent along the crack, yet they are associated with a complex-conjugated pair of attenuation indexes, which amounts to oscillatory decay away from the crack line. For this localized edge wave, a novel dispersion relation is given and it is shown that its attenuation stands between the attenuation of decaying and propagating waves in the speed regime η<ηe. Consequently, this wave may be put to advantage for defect detection. This localized solution seems somewhat connected with the dynamic edge effect in cylindrical shells [30], for which curvature plays the role that is here taken by the elastic support.

    Dynamic stress intensity factors are also obtained and they show a remarkable resonant behaviour, which is explained in the light of the ERR at the crack tip. Resonance may be successfully exploited in many practical applications, for instance to speed up ice breaking or sawing, or carefully avoided in many others, for example, to prevent rapid deterioration in pavements or layered materials. Finally, the work of the applied loading is shown to be proportional to the ERR at the crack tip through developing a conservative integral for elastically supported thin plates.

    Data accessibility

    This work does not have any experimental data.

    Author contributions

    E.R. and A.N. developed the model and the full-field solution; L.L. carried out the computational work. All authors gave their final approval for publication.

    Competing interests

    We have no competing interests.

    Funding statement

    This work was supported by ‘Fondazione Cassa di Risparmio di Modena’ within the framework of the Progetti di Ricerca finalizzata all’innovazione 2014, Sime n.2013.0662.

    Footnotes

    Electronic supplementary material is available online at http://dx.doi.org/10.6084/m9.figshare.c.3855910.

    Published by the Royal Society. All rights reserved.