ArticlePDF Available

Abstract and Figures

In previous work (Lee et al., 2016, 2017), Lee et al. introduced a new Smooth Particle Hydrodynamics (SPH) computational framework for large strain explicit solid dynamics with special emphasis on the treatment of near incompressibility. A first order system of hyperbolic equations was presented expressed in terms of the linear momentum and the minors of the deformation, namely the deformation gradient, its co-factor and its Jacobian. Taking advantage of this representation, the suppression of numerical deficiencies (e.g. spurious pressure, long term instability and/or consistency issues) was addressed through well-established stabilisation procedures. In Reference Lee et al. (2016), the adaptation of the very efficient Jameson-Schmidt-Turkel algorithm was presented. Reference Lee et al. (2017) introduced an adapted variationally consistent Streamline Upwind Petrov Galerkin methodology. In this paper, we now introduce a third alternative stabilisation strategy, extremely competitive, and which does not require the selection of any user-defined artificial stabilisation parameter. Specifically, a characteristic-based Riemann solver in conjunction with a linear reconstruction procedure is used, with the aim to guarantee both consistency and conservation of the overall algorithm. We show that the proposed SPH formulation is very similar in nature to that of the upwind vertex centred Finite Volume Method presented in Aguirre et al. (2015). In order to extend the application range towards the incompressibility limit, an artificial compressibility algorithm is also developed. Finally, an extensive set of challenging numerical examples is analysed. The new SPH algorithm shows excellent behaviour in compressible, nearly incompressible and truly incompressible scenarios, yielding second order of convergence for velocities, deviatoric and volumetric components of the stress.
Content may be subject to copyright.
A Total Lagrangian upwind Smooth Particle Hydrodynamics
algorithm for large strain explicit solid dynamics
Chun Hean Lee a,1, Antonio J. Gil b, 2, Ataollah Ghavamianb, Javier Bonetc
(a) Glasgow Computational Engineering Centre, School of Engineering
University of Glasgow, Glasgow, G12 8QQ, United Kingdom
(b) Zienkiewicz Centre for Computational Engineering, College of Engineering
Swansea University, Bay Campus, SA1 8EN, United Kingdom
(c) University of Greenwich, London, SE10 9LS, United Kingdom
Abstract
In previous work [1, 2], Lee et al. introduced a new Smooth Particle Hydrodynamics (SPH)
computational framework for large strain explicit solid dynamics with special emphasis on the
treatment of near incompressibility. A first order system of hyperbolic equations was presented
expressed in terms of the linear momentum and the minors of the deformation, namely the
deformation gradient, its co-factor and its Jacobian. Taking advantage of this representation,
the suppression of numerical deficiencies (e.g. spurious pressure, long term instability and/or
consistency issues) was addressed through well-established stabilisation procedures. In Refe-
rence [1], the adaptation of the very efficient Jameson-Schmidt-Turkel algorithm was presented.
Reference [2] introduced an adapted variationally consistent Streamline Upwind Petrov Galer-
kin methodology. In this paper, we now introduce a third alternative stabilisation strategy,
extremely competitive, and which does not require the selection of any user-defined artificial
stabilisation parameter. Specifically, a characteristic-based Riemann solver in conjunction with
a linear reconstruction procedure is used, with the aim to guarantee both consistency and con-
servation of the overall algorithm. We show that the proposed SPH formulation is very similar
in nature to that of the upwind vertex centred Finite Volume Method presented in [3]. In order
to extend the application range towards the incompressibility limit, an artificial compressibi-
lity algorithm is also developed. Finally, an extensive set of challenging numerical examples is
analysed. The new SPH algorithm shows excellent behaviour in compressible, nearly incom-
pressible and truly incompressible scenarios, yielding second order of convergence for velocities,
deviatoric and volumetric components of the stress.
Keywords: Conservation laws, SPH, Upwind, Riemann Solver, Explicit dynamics,
Incompressibility
1. Introduction
The classical (displacement-based) Smooth Particle Hydrodynamics (SPH) Lagrangian for-
malism [4–10] is well-known to suffer from a number of severe drawbacks, namely: (1) tensile
1Corresponding author: chunhean.lee@glasgow.ac.uk. This work was completed whilst at the previous af-
filiation: Zienkiewicz Centre for Computational Engineering, College of Engineering, Swansea University, Bay
Campus, SA1 8EN, United Kingdom
2Corresponding author: a.j.gil@swansea.ac.uk
Preprint submitted to Computer Methods in Applied Mechanics and Engineering September 27, 2018
instability, spurious pressure and hour-glassing [11, 12]; (2) numerical issues associated with
conservation, consistency, long term stability and convergence [13, 14] and (3) the reduced order
of convergence for derived variables (e.g. stresses and strains) [15–18].
Over the last two decades, significant effort has been devoted to enhance the robustness of
classical SPH algorithms. The numerical deficiencies described above can be partially alleviated
through the use of corrected kernel gradient approximations, combined with so-called non-
consistent stabilisation strategies [8] (i.e. artificial viscous fluxes [4, 19, 20] and conservative
strain smoothing regularisation [15, 21]). As reported in References [6, 8], the above enhanced
SPH methodologies ensure the reproducibility of complete polynomial basis in finite domains
[5, 22–24]), but still suffer from persistent artificial mechanisms similar to hour-glassing when
attempting to model problems with predominant nearly incompressible behaviour.
In very recent works [1, 2], some of the authors of the present manuscript have successfully
introduced a mixed-based SPH computational framework for explicit fast solid dynamics, where
the conservation of linear momentum pis solved along with conservation equations for the
deformation gradient F, its co-factor Hand its Jacobian J. Specifically, the SPH discretisation
of the new system of conservation laws {p,F,H, J }was introduced through a family of well-
established stabilisation strategies, namely a Jameson-Schmidt-Turkel (JST) algorithm [25, 26]
and a variationally consistent Streamline Upwind Petrov Galerkin (SUPG) algorithm [2, 27].
Both computational methodologies were capable of eliminating spurious hourglass-like modes
and spurious pressure oscillations in nearly incompressible scenarios.
The main objective of the current paper is to explore a third alternative SPH spatial discre-
tisation of the same system {p,F,H, J}by means of a very competitive Riemann solver based
stabilisation methodology, without the need to introduce any artificial stabilisation parameters.
The mixed-based set of equations is advanced in time by means of an explicit time integrator,
where the time step is controlled through the Courant-Friedrichs-Lewy number [28] dependent
on the volumetric wave speed cp. We also show that the new SPH algorithm is very similar
to an upwind vertex centred FVM [3], where the latter requires the generation of a dual mesh
using the medial dual approach [29]. In the near (or full) incompressibility limit, the wave
speed cpcan reach very high (even infinite in the degenerate case) values potentially leading to
a very inefficient algorithm [26, 30]. To address this issue, we then present an adapted artifi-
cial compressibility algorithm [31–33], supplemented with the use of a (conservative) Laplacian
viscosity term, if necessary, to allow for the explicit simulation of (nearly) incompressible solids.
The paper is organised as follows. Section 2 briefly summarises the system of first or-
der conservation laws {p,F,H, J }for large strain solid dynamics. Section 3 describes the
computational methodology of the Upwind SPH framework. The variational statement of the
{p,F,H, J}system, the SPH approximation, the Riemann solver based SPH discretisation and
the linear reconstruction procedure are presented. For clarity, the complete explicit Upwind
SPH algorithm is summarised in Section 4. In addition, an artificial compressibility algorithm
is also presented in Section 5 to account for truly and nearly incompressible solids. Section 6
summarises the flowchart of the proposed SPH methodology. In Section 7, an extensive set of
challenging numerical examples is examined to assess the performance of the proposed algo-
rithm. Finally, Section 8 presents some concluding remarks and current directions of research.
2. Reversible elastodynamics
Consider the three dimensional deformation of an isothermal body of material density ρ0
moving from its initial undeformed configuration occupying a volume V, of boundary ∂V , to
2
Figure 1: Motion of a deformable continuum domain
a current deformed configuration at time toccupying a volume v, of boundary ∂v (see Figure
1). The time dependent motion is defined through a deformation mapping x=φ(X, t) which
satisfies the following set of Total Lagrangian first order conservation laws [1, 2, 27, 34–38]
p
∂t DIV P=f0; (1a)
F
∂t 0p
ρ0=0; (1b)
H
∂t CURL p
ρ0
F=0; (1c)
∂J
∂t DIV 1
ρ0
HTp= 0.(1d)
Here, p:= ρ0vis the linear momentum per unit of undeformed volume, vis the velocity field,
Fis the deformation gradient (or fibre map), His the co-factor of the deformation (or area
map), Jis the Jacobian of the deformation (or volume map), Pis the first Piola-Kirchhoff
stress tensor and f0is a body force term per unit of undeformed volume. The symbol
represents the tensor cross product between vectors and/or second order tensors in the sense of
[25, 30, 39, 40], DIV and CURL represent the material divergence and material curl operators
as defined in expressions (5) and (7) of Reference [25], respectively, and 0represents the
material gradient operator defined as 0:=
X.
Above equation (1a) represents the conservation of linear momentum, whilst the rest of the
system equations (1b-1d) represent a supplementary set of conservation laws for the geometric
strain measures {F,H, J}. Additionally, appropriate involutions [1, 26] must be satisfied by
some of the strain variables {F,H}[41] of the system as
CURLF=0; DIVH=0.(2)
Notice that, if necessary, the use of these involutions enables the conservation equations for
3
the area and volume maps in (1c) and (1d) to be re-written as [1]
H
∂t F0p
ρ0=0;∂J
∂t H:0p
ρ0= 0.(3)
More generally, the above system of conservation laws (1) can be summarised in a concise
manner as U
∂t +FI
∂XI
=S;I= 1,2,3,(4)
where Udenotes the set of conservation variables, Sthe source term and FIthe flux vector
in the material Cartesian direction I, as follows
U=
p
F
H
J
;FI=
P EI
1
ρ0pEI
F1
ρ0pEI
H:1
ρ0pEI
;S=
f0
0
0
0
,(5)
with EIis the I-th unit vector of the Cartesian basis defined as
E1=
1
0
0
;E2=
0
1
0
;E3=
0
0
1
.(6)
Furthermore, the corresponding flux vector (5b) associated with the material unit outward
normal Ncan be expressed as
FN=FINI=
P N
1
ρ0pN
F1
ρ0pN
H:1
ρ0pN
.(7)
Crucially, in the presence of non-smooth solutions, all of the above conservation laws (1a-1d)
are accompanied by appropriate jump conditions [34, 42] defined as
cJpK=JPKN; (8a)
cJFK=1
ρ0
JpKN; (8b)
cJHK=FAve 1
ρ0
JpKN; (8c)
cJJK=HAve :1
ρ0
JpKN,(8d)
where J·Kdenotes the jump operator across a discontinuity surface with normal Nmoving with
speed cin the reference space and {FAve,HAve}being defined as an average of the geometric
strain measures between the left and right states of a discontinuity surface.
For the particular case of a reversible process, the closure of the system described in (1)
requires the introduction of a suitable constitutive law relating the stress tensor Pwith the
4
strain measures {F,H, J}, obeying the principle of objectivity [43] and thermodynamic consis-
tency (via the Colemann-Noll procedure) [44]. In this work, a polyconvex nearly incompressible
constitutive model is used (see Section 2.1 of Reference [1] on pg. 75). Taking advantage of the
complete set of conservation laws described in (1), the polyconvex nature of the constitutive
model (i.e. a guarantor of material stability [45, 46]) facilitates the transformation of the sy-
stem of conservation laws into a symmetric set of hyperbolic equations expressed in terms of the
entropy conjugates of the conservation variables [25, 30]. Finally, for the complete definition
of the initial boundary value problem, initial and boundary (essential and natural) conditions
must also be specified as appropriate.
Remark 1:
In the present manuscript, consideration of irreversible processes is restricted to the case of
an isothermal elasto-plastic model typically used in metal forming applications [24]. Particu-
larly, thermal effects will be neglected. In order to model irrecoverable plastic behaviour, the
standard rate-independent von Mises plasticity model [43] with isotropic hardening is used and
the basic structure is summarised for completeness in Appendix A.
3. Spatial discretisation
The set of local conservation equations described in (1a-1d) has recently been spatially
discretised by the authors using two well-established (and consistent) stabilisation strategies,
leading to the Jameson-Schmidt-Turkel Smooth Particle Hydrodynamics (JST-SPH) algorithm
[1] and the Streamline Upwind Petrov-Galerkin Smooth Particle Hydrodynamics (SUPG-SPH)
algorithm [2]. These numerical techniques resulted in being very efficient when attempting to
solve problems with predominant nearly incompressible behaviour. JST-SPH requires the eva-
luation of a blend of Laplacian and bi-Laplacian operators for JST stabilisation and SUPG-SPH
requires the generation of a secondary set of particles for the computation of a residual-based
SUPG stabilisation. In this paper, we explore a third and extremely competitive stabilised
SPH method. The new methodology is known as Upwind Smooth Particle Hydrodynamics
(Upwind-SPH) and will be discussed in the following section.
3.1. General remark
In general, a standard weak variational statement for the mixed-based system {p,F,H, J}
(1a-1d) (known as Bubnov-Galerkin contribution AGal) is established by multiplying the local
form of the conservation laws (4) by appropriate work conjugate virtual fields δVand integra-
ting over the volume Vof the body, to give
0 = AGal(U, δV) := ZV
δVU
∂t dV ZV
δVSdV +ZV
δVFI
∂XI
dV. (9)
The symbol is used to denote the inner (dual) product of work conjugate pairs, δV:=
{δv, δΣF, δΣH, δΣJ}represent the virtual work conjugates of the conservation variables U,
δvis the virtual velocity field and {δΣF, δΣH, δΣJ}are appropriate conjugate stresses to
{F,H, J}respectively.
5
Figure 2: Particle approximation
Application of the Green-Gauss divergence theorem on the last term of (9) results in
0 = AGal(U, δV) := ZV
δVU
∂t dV ZV
δVSdV ZV
FI∂δV
∂XI
dV +Z∂V
δVFNdA. (10)
Above Galerkin representation (10) can be particularised to the case of the linear momentum
pand the extended set of geometric strain measures {F,H, J}
0 = Ap
Gal := ZV
δv·p
∂t dV +ZV
P:0δvdV ZV
δv·f0dV Z∂V
δv·tBdA; (11a)
0 = AF
Gal := ZV
δΣF:F
∂t 0p
ρ0dV ; (11b)
0 = AH
Gal := ZV
δΣH:H
∂t F0p
ρ0dV ; (11c)
0 = AJ
Gal := ZV
δΣJ∂J
∂t H:0p
ρ0dV, (11d)
with tBbeing a possible boundary traction vector. Notice that only the linear momentum
conservation equation (11a) has been integrated by parts in order to enable the imposition of
boundary tractions.
3.2. Corrected SPH approximation
For evaluation of the material gradient of any arbitrary vector function f, we employ the
following particle approximation via the use of the “Corrected Gradient of a Corrected Kernel
˜
0˜
W” [24, 47] described as follows
0f(Xa)X
bΛb
a
fbGb(Xa); Gb(Xa) := Vb˜
0˜
Wb(Xa).(12)
Here, Λb
arepresents the set of neighbouring particles bthat lie inside a sphere of a given radius
2haround Xa(see Figure 2), Vband fbrepresent the volume and time varying vector function f
6
stored at particle b. The kernel approximation W being employed in this paper is the standard
quadratic kernel function introduced in [48] (see equation (5) of Section 2.1 on pg. 350).
As shown in Reference [47] (see Section 4.3 on pg. 106-107), the above kernel gradient
evaluation Gb(Xa) reproduces exactly the gradient of any constant and/or linear function. For
this reason, equation (12), if necessary, can also be alternatively expressed by including the
redundant term fain (12) as
0f(Xa)X
bΛb
a
21
2(fa+fb)Gb(Xa).(13)
Remark 2:
Another useful representation for the evaluation of a kernel gradient can be formulated by
artificially adding the redundant term fato (12), resulting in
0f(Xa)X
bΛb
a
21
2(fbfa)Gb(Xa).(14)
One of the key contributions of this paper is to show a relationship between the SPH particle
approximation and the vertex centred Finite Volume Method (FVM) approximation [3], where
the latter requires the definition of a dual mesh3which is constructed using the median dual
approach [29]. This will enable a well-established upwinding finite volume spatial discretisation
[49, 50] commonly used in the CFD community to be adapted to a SPH mesh-free method.
To achieve this, we first recall the discrete gradient evaluation used within the control volume
associated with node aby utilising a discrete version of the Green-Gauss divergence theorem
0f(Xa)1
VaZ∂Va
fNdA, (15)
with Nbeing the outward unit normal vector on the boundary of the control volume Va.
Above discrete boundary integral (15) can then be discretised by means of the FVM where
a (second order) central difference approximation is used for the discretisation of the vector
function f, to give
0f(Xa)1
VaX
bΛb
a1
2(fa+fb)Cab,(16)
where for a given edge connecting nodes aand b(see Figure (3a)), the material area vector Cab
is defined as4
Cab := X
kΛk
ab
NkAk,(17)
where Λk
ab is the set of facets belonging to edge ab,Akis the area of a given facet kand
Nkits outward normal. These material area vectors enable a substantial reduction in the
computational cost when computing the boundary integral used in the Green-Gauss divergence
theorem, since they save an additional loop on facets per edge ab [29, 38].
3The dual mesh is constructed by connecting edge midpoints with element centroids in two dimensions (see
Figure 3a) and edge midpoints with face centroids and element centroids in three dimensions.
4Due to the definition of a dual mesh, the area vectors satisfy Cab = -Cba.
7
(a) (b)
Figure 3: Control volume for (a) vertex centred FVM and (b) SPH mesh free approach in two dimensions. The
red shaded area is the control volume associated to node a. The blue lines are the edges connecting node a
to its neighbouring nodes bi. The vertex centred FVM requires the outward unit normal Nab to be defined
using the material outward normal Nkof every facet kbelonging to edge ab. However, for the SPH mesh free
approach, the normal Nab is defined by a direction vector between particles aand b.
Comparing equation (13) with (16), a useful relationship arises relating the (mesh-based)
material outward area vector with the (mesh-free) SPH kernel gradient evaluation, defined as
Cab := 2VaGb(Xa).(18)
Its reciprocal relationship can also be defined as Cba := 2VbGa(Xb). Due to the anti-symmetric
nature of the SPH gradient correction [7, 47], notice here that Gb(Xa)6=Ga(Xb) which in
turn lead to Cab 6=Cba.
3.3. SPH spatial discretisation
Typically, in the context of Element Free Galerkin schemes [51, 52], the Galerkin weak
statements described in (11a-11d) are accurately evaluated using the necessary distribution of
(Gauss) quadrature points in order to avoid spurious hourglass (or zero-energy) modes [12,
15, 53]. In general, the positions of these quadrature points do not coincide with those of the
material particles [51, 54]. This is however not the case when employing a mesh-free SPH
discretisation. In the latter case, above integrands (11a-11d) are under-integrated at the cloud
of particles regarded as quadrature points [1, 4, 6, 8, 55, 56].
3.3.1. Bubnov-Galerkin contribution
Using a corrected kernel interpolation ˜
W (see Section 3.2 on pg. 77 in Reference [1]) along
with the corresponding discrete gradient evaluation of 0δvdescribed in (14), term Ap
Gal (11a)
results in the following semi-discrete expression
dpa
dt =EaTa.(19a)
8
Here, the nodal external Eaand internal Taforce vectors are defined as
Ea=Aa
Va
ta
B+fa
0;Ta=X
bΛb
a
2VbhPb˜
0˜
Wb(Xa)iAve ,(20)
where hPb˜
0˜
Wb(Xa)iAve := 1
2hPb˜
0˜
Wa(Xb)Pa˜
0˜
Wb(Xa)i.(21)
Here, Pa,b := P(Fa,b,Ha,b , Ja,b). Aaand ta
Bare the tributary area and traction vector of those
particles placed on the boundary, the latter being computed directly from the given traction
boundary conditions [3, 34, 42]. The internal force representation Tadescribed in (20) satisfies
the global conservation of linear momentum, that is PaVaTa=0.
Analogously to the above discretisation procedure, the extended set of geometric conserva-
tion laws AF,H,J
Gal (11b-11d) can now follow, but this time using the gradient approximation of
0p
ρ0presented in (13). This will yield
dFa
dt =X
bΛb
a
2pAve
ρ0Gb(Xa); (22a)
dHa
dt =FaX
bΛb
a
2pAve
ρ0Gb(Xa); (22b)
dJa
dt =Ha:X
bΛb
a
2pAve
ρ0Gb(Xa),(22c)
where the average linear momentum pAve := 1
2(pa+pb).
Remark 3: It is easy to show that the vertex-centred FVM presented in Reference [3] is a parti-
cular case of the above representation (19a, 22a-22c). To show this, we utilise the relationship
described in (18) and replacing the kernel gradient Gb(Xa) in (19a, 22a-22c) with the material
area vector Cab, which gives
dpa
dt =1
Va
Ea1
VaX
bΛb
a
[PbCab]Ave ; (23a)
dFa
dt =1
VaX
bΛb
apAve
ρ0Cab; (23b)
dHa
dt =1
Va
FaX
bΛb
apAve
ρ0Cab; (23c)
dJa
dt =1
Va
Ha:X
bΛb
apAve
ρ0Cab; (23d)
with [PbCab]Ave := 1
2[PbCba PaCab]. However, for the case of the dual mesh used in FVM,
the area vectors become Cab =Cab, and thus satisfying Cba =Cab. As a result, the linear
momentum evolution (23a) reduces to the following expression
dpa
dt =1
Va
Ea+1
VaX
bΛb
a
PAveCab;PAve =1
2(Pa+Pb),(24)
9
X,x
Y,y
Z,z
N+
N
n
n+
n
n+
c
sc+
s
c+
p
c
p
Time t=0
Time t
Figure 3: Contact mechanics (To be replaced!!!)
(23a) can be reduced to
dpa
dt =1
Va
Ea+1
VaXb
a
4.3.2. Riemann-based upwiding stabilisation
In this paper, by taking advantage of the conservation structure of the mixed-based set of
equations (1a-1d) to be solved, a well-established numerical stabilisation algorithm is introduced
via a characteristic-based Riemann solver previously explored in the context of finite volume
method [33, 41, 45]. In what follows, some of the results presented in Reference [41] (see Section
4.3 on pages. 417-420) will be employed.
10
However, the resulting system, either SPH discretisation described in (19a, 22a-22c) or FVM
discretisation described in (24, 23b-23d), suffers from severe numerical instabilities (e.g. hour-
glassing, pressure and tensile instabilities) [12, 18, 24, 52] and requires some form of numerical
viscosity to counterbalance the negative diffusion introduced by the classical Bubnov-Galerkin
approximation [25–27, 29, 33, 34, 37, 41, 45]. In this work, an adapted (particle) Riemann-
based upwinding stabilisation strategy is introduced, crucial to circumvent this shortcoming.
This will be addressed in the next section.
a
1
PAveCab;PAve =
b
V(t)
V(t)
a
b
V
a
φ (X, t)
φ (X, t)
Vb
Figure 4: Contact mechanics
and the rest of the evolution equations for the geometric strain measures (23b-23d) remain
exactly the same but replacing Cab with Cab. Notice that expressions (24, 23b-23d) are exactly
the semi-discrete equations presented for a vertex centred FVM [3].
However, the resulting spatially discretised systems, either the SPH discretisation described
in (19a, 22a-22c) or the FVM discretisation in (24, 23b-23d), suffer from severe numerical
instabilities (e.g. hour-glassing, pressure and tensile instabilities) [12, 18, 24, 57] and require
some form of numerical dissipation [3, 25–27, 30, 34, 35, 38, 42]. In this work, an adapted
(particle) Riemann-based upwinding stabilisation strategy is introduced to circumvent this
shortcoming. This will be presented in the next section.
3.3.2. Riemann-based upwinding stabilisation
By taking advantage of the conservation structure of the mixed-based set of equations (1a-
1d), numerical stabilisation is introduced via a characteristic-based Riemann solver previously
explored in the context of finite volume methods [3, 34, 42]. In what follows, some of the results
presented in Reference [42] (see Section 4.3 on pages. 417-420) will be employed.
In Lagrangian dynamics, it is often possible for two domains apart (i.e. Vaand Vb) to
come into contact with each other after some time t. The impact would generate two types of
shock waves (p-wave cpand s-wave cs) travelling within the domains, as illustrated in Figure
4. Similar waves propagate in the fictitious interface between neighbouring particles when
considering piecewise discontinuous solutions.
Referring to Section 4.3 of [42], the numerical (interface) fluxes FC
Nab across a discontinuity
10
surface defined by the outward unit normal vector Nab can be read as follows
FC
Nab :=
PCNab
1
ρ0pCNab
Fa1
ρ0pCNab
Ha:1
ρ0pCNab
.(25)
These numerical fluxes can be additively decomposed into the summation of an average state
and an upwinding numerical stabilisation term, expressed as [·]C:= [·]Ave + [·]stab.
In order to ensure the discrete satisfaction of the involutions described by equation (2), we
must not introduce any numerical dissipation into (25b) and (25c) by setting the values of pC=
pAve. This implies that the update for {F,H}are naturally curl- and divergence-free as their
semi-discrete equations are formulated in terms of a material discrete gradient of a continuous
velocity field [3, 30, 38]. As a result, above flux approximations (25) now reduce to
FC
Nab =
PAveNab
1
ρ0pAve Nab
Fa1
ρ0pAve Nab
Ha:1
ρ0pAve Nab
| {z }
Average
PstabNab
0
0
Ha:1
ρ0pstab Nab
| {z }
Upwinding stabilisation
.(26)
The first term on the right hand side of (26) represents the average (Galerkin) state, whereas
the second term (being high order corrections) can be interpreted as an upwinding numerical
stabilisation that damps the possible instabilities arising from the first term [34]. A detailed
derivation of the upwinding stabilisation term described in (26) can be found in Reference [42].
For completeness, the expressions for the stabilised first Piola Kirchhoff stresses and linear
momentum are [42]
Pstab := Sp
ab p+
fp
fNab;pstab := SP
ab P+
fP
fNab,(27)
with the (acoustic) stabilisation matrices {Sp
ab,SP
ab}being defined as
Sp
ab =1
2[cp(nab nab) + cs(Inab nab )] ; SP
ab =1
2cp
(nab nab).(28)
Here, the outward spatial normal is defined by the current direction vector between particles
aand bsuch as nab := xbxa
kxbxakand {p,+
f,P,+
f}denote the left and right states of {p,P}at
fictitious interface f. The procedure of evaluating the reconstructed states {p,+
f,P,+
f}will
be discussed in Section 3.3.3.
It is now possible to introduce Riemann based stabilisation terms to the Galerkin contribu-
tions described in (24) and (23d). To achieve this, we need to replace {PAve,pAve}of (24) and
(23d) with {PC,pC}via (26), and after some simple algebraic manipulation gives
dpa
dt =
1
Va
Ea+1
VaX
bΛb
a
1
2(PbCba PaCab)
+D(pa); (29a)
dJa
dt =
1
Va
Ha:X
bΛb
apAve
ρ0Cab
+D(Ja),(29b)
11
where the Riemann based dissipative terms are
D(pa) = 1
VaX
bΛb
a
PstabCab ;D(Ja) = 1
VaX
bΛb
a
1
ρ0
pstab ·[HaCab].(30)
Substitution of the fundamental relation (18) described by Cab = 2VaGb(Xa) (and its
reciprocal relation Cba = 2VbGa(Xb)) into (29) and (30), the squared bracket terms in (29a,b)
recover the Galerkin discrete expressions for the conservation equations of linear momentum
(19a) and Jacobian (22c), whereas the particle upwinding stabilisation terms can be found as
D(pa) = X
bΛb
a
2VbPstab ˜
0˜
Wb(Xa); D(Ja) = X
bΛb
a
2Vb
pstab
ρ0·hHa˜
0˜
Wb(Xa)i.(31)
Since in this case ˜
0˜
Wb(Xa)6=˜
0˜
Wa(Xb), the dissipation term D(pa) presented in (31a)
will not automatically satisfy the global conservation requirement [47]. However, this can be
easily fulfilled by replacing ˜
0˜
Wb(Xa) with its average counterpart ˜
0˜
WAve
b(Xa) defined as
˜
0˜
WAve
b(Xa) := 1
2h˜
0˜
Wb(Xa)˜
0˜
Wa(Xb)i.(32)
Similarly, in order to ensure the stabilisation term D(Ja) is globally conservative, the term
hHa˜
0˜
Wb(Xa)ipresented in (31b) must also be replaced by hHa˜
0˜
Wb(Xa)iAve defined by
hHa˜
0˜
Wb(Xa)iAve := 1
2hHb˜
0˜
Wa(Xb)Ha˜
0˜
Wb(Xa)i.(33)
3.3.3. Linear reconstruction procedure
In order to guarantee second order accuracy in space, a linear reconstruction procedure
[34] is employed for the evaluation of the neighbouring states of the Riemann values, that is
{p,+
f,P,+
f}.
The reconstruction procedure can be achieved using two different strategies. The first
standard approach is to reconstruct both the linear momentum and the deformation gradient
tensor (e.g. {p
f,F
f}and {p+
f,F+
f}) at the mid-edge fconnecting particles aand b, followed
by the computation of the corresponding stresses {P(F
f),P(F+
f)}. The second strategy is to
first compute the stresses at each particle (i.e. P(Fa) and P(Fb)) and then reconstruct the
computed stresses to the interface f, namely {P
f,P+
f}. The latter is of particular interest in
this paper since it requires fewer evaluations of the stresses and, as complex constitutive models
are typically used in solid mechanics, this can result in a faster algorithm.
For any individual component Uof the variables pand P, the linear reconstructed value
at any position Xis in the form of
U(X) = Ua+Ga·(XXa).(34)
To achieve this, it is necessary to obtain an appropriate (particle) gradient operator Gathrough
the least squares minimisation process [34, 42].
Introducing first the objective functional Π given as
Π(Ga) = 1
2X
bΛb
a
1
kdabk2[Ub(Ua+Ga·dab )]2;dab =XbXa.(35)
12
Here, bΛb
arepresents a set of neighbouring (particle) values bassociated with particle a, and
dab represents a material vector measured from position ato b. Taking directional derivative
of (35) with respect to Ga, yields
0=DΠ[Ga] = X
bΛb
a
1
kdabk2[Ub(Ua+Ga·dab )] dab.(36)
Rearranging equation (36) gives the following expression for Ga
Ga=
X
bΛb
a
Nab Nab
1
X
bΛb
aUb− Ua
kdabkNab ,(37)
where Nab is the unit vector in the direction of dab described in (C.3) (see Figure 3b).
4. Complete Upwind-SPH algorithm
Addition of the discrete upwinding stabilisation terms (31) to the Bubnov-Galerkin discrete
expressions (see (19a) and (22a-22c)), finally yields the complete semi-discrete set of governing
equations for {p,F,H, J}as
dpa
dt =EaTa+D(pa); (38a)
dFa
dt =X
bΛb
a
2pAve
ρ0Gb(Xa); (38b)
dHa
dt =FaX
bΛb
a
2pAve
ρ0Gb(Xa); (38c)
dJa
dt =Ha:X
bΛb
a
2pAve
ρ0Gb(Xa) + D(Ja).(38d)
The Riemann-based stabilising terms {D(pa),D(Ja)}and the internal Taand external Ea
force vectors are already defined in Section 3.3. Notice here that the upwinding stabilisation
is only applied to the linear momentum evolution D(pa) (38a) and the volume map evolution
D(Ja) (38d). The former alleviates the appearance of spurious zero-energy (hourglass-like [8])
modes due to rank deficiency inherent to the use of Galerkin particle integration, whereas the
latter removes pressure instabilities in near incompressibility [26].
Remark 4:
Alternatively, it is also possible to show that the (particle) discrete system (38a-38d) can
13
be re-written as
dpa
dt =EaX
bΛb
a
Vb
Va
PbGa(Xb) + D(pa); (39a)
dFa
dt =X
bΛb
a
pb
ρ0Gb(Xa); (39b)
dHa
dt =FaX
bΛb
a
pb
ρ0Gb(Xa); (39c)
dJa
dt =Ha:X
bΛb
a
pb
ρ0Gb(Xa) + D(Ja).(39d)
Insofar as the mixed-based system {p,F,H, J}, either (38) or (39), is rather large, it will
only be suitable to employ an explicit type of time integrator. For simplicity, an explicit one-
step two-stage Total Variation Diminishing Runge-Kutta (TVD-RK) scheme has been used
[1–3, 25–27, 34, 38, 42]. This is described by the following time update equations from time
step tnto tn+1:
U?
a=Un
a+ ∆t˙
Un
a(Un
a, tn);
U??
a=U?
a+ ∆t˙
U?
a(U?
a, tn+1);
Un+1
a=1
2(Un
a+U??
a).
(40)
In this manuscript, the geometry is also updated through the above TVD-RK algorithm
[1, 2, 42]. This results in a monolithic time integration procedure where the unknowns U=
(p,F,H, J, )Talong with the geometry xare all updated via (40).
The maximum time step ∆t:= tn+1 tnis governed by a standard Courant-Friedrichs-Lewy
(CFL) condition [28] given as
t=αCF L
hmin
cp,max
,(41)
where cp,max is the maximum p-wave speed, hmin is the characteristic particle spacing within the
computational domain and αCF L is the CFL stability number. For the numerical computations
presented in this paper, a value of αCF L = 0.3, unless otherwise stated, has been chosen to
ensure both accuracy and stability [38] of the algorithm.
It is important to point out that the resulting Upwind-SPH algorithm (either (38) or
(39)) does not intrinsically fulfil conservation of angular momentum, since the strain measures
{F,H, J}are no longer exclusively computed from the current geometry [42]. For instance,
the deformation gradient F6=Fx:= 0x, its co-factor H6=Hx:= 1
2FxFxand its Jacobian
J6=Jx:= det Fx. In this paper, a monolithic discrete angular momentum projection algorithm
presented in Lee et al. [1] is employed and is summarised in Appendix B. Specifically, both the
internal nodal force Taand the particle-based upwinding stabilisation term D(pa), described
in (38a), are modified in a least-square sense in order to preserve the total angular momentum,
whilst still ensuring the global conservation of linear momentum.
14
5. Artificial compressibility
5.1. General remark
As it is well known, in the case of truly incompressible or nearly incompressible materials,
the volumetric wave speed cpcan reach very large values leading to prohibitively small time
steps [28]. This can have a very negative effect on the computational efficiency of any time-
explicit algorithm. One of the most popular approaches to address this issue is the artificial
compressibility method, originally developed for the Navier-Stokes equations [31–33]. Taking
inspiration from the fractional step algorithm presented in Gil et al. [26, 30], the artificial
compressibility approach is here adapted to the set of first order conservation law equations
presented in (1).
To achieve this, it is first important to re-write the volume map conservation law (3b) in
terms of the pressure unknown by utilising the volumetric constitutive law used in this paper,
namely p=κ(J1) (refer to equation (11) on pg. 75 in Reference [1]). This equation now
reads 1
κ
∂p
∂t =H:0p
ρ0.(42)
In this case, the new unknowns of the problem are {p,F,H, p}(with Jbeing replaced by the
pressure variable p). For a truly incompressible material, κ≈ ∞ and the left hand side of
equation (42) vanishes, resulting in an incompressibility constraint [31].
In this approach, we first discretise the continuum equations (1a, 1b, 1c, 42) in time and
then proceed to discretise in space via a Riemann-based upwinding stabilisation as presented in
Section 3.3.2. For satisfaction of the incompressibility constraint, we can update the evolution
equations of the linear momentum (1a) and pressure (42) as follows [26, 30]
pn+1 pn
tDIVP(Fn,Hn, pn+1)fn
0=0; (43a)
pn+1 pn
κtHn:0pn+1
ρ0= 0.(43b)
Both the pressure and linear momentum conservation variables are solved implicitly in tn+1,
assuming the deformation gradient Fand its cofactor Hto be frozen at time tn.
In order to solve equations (43a,b), a predictor-corrector algorithm is used. The algorithm
is first advanced explicitly yielding intermediate variables {pint, pint}which are then projected
after iteratively solving an implicit system known as pressure correction. Therefore, the first
(predictor or intermediate) step of the scheme over a time step ∆tis defined as
pint pn
tDIVP(Fn,Hn, pn)fn
0=0; (44a)
Fn+1 Fn
t0pn
ρ0=0; (44b)
Hn+1 Hn
tFn0pn
ρ0=0; (44c)
1
κ
pint pn
tHn:0pn
ρ0= 0.(44d)
15
The second (corrector or projection) step becomes
pn+1 pint
tDIV pn+1 pnHn=0; (45a)
1
κpn+1 pint
tHn:0pn+1
ρ0pn
ρ0= 0.(45b)
Here, the maximum time step ∆tof the predictor-corrector algorithm described above is go-
verned by the maximum p-wave speed dependent on the bulk modulus κof a material.
For nearly (and truly) incompressible materials, the bulk modulus present in the first term
of equation (44d) can potentially reach very high values (even infinite), leading to extremely
small time steps. This will then destroy the explicit nature of the predictor step of the scheme.
It is for this reason that a fictitious bulk modulus ˜κis used in its place, yielding
1
˜κ
pint pn
tHn:0pn
ρ0= 0.(46)
As a consequence, the projection step of the pressure equation (45b) now becomes
1
κ
(pn+1 pn)
t1
˜κ
pint pn
tHn:0pn+1
ρ0pn
ρ0= 0.(47)
Notice that the summation of both equations (46) and (47) recovers the original assumption in
which the linear momentum variable is treated implicitly in the formulation.
Finally, in order to solve the implicit system for the correction (refer to (45a) and (47)),
one approach is to employ an iterative artificial compressibility algorithm [29]. This can be
achieved through the introduction of a pseudo time derivative term
∂τ to expressions (45a) and
(47), and after rearranging gives
p
∂τ = DIV pn+1 pnHnpn+1 pint
t; (48a)
1
γ
∂p
∂τ =1
˜κ
pint pn
t+Hn:0pn+1
ρ0pn
ρ01
κ
(pn+1 pn)
t,(48b)
where γrepresents the artificial compressibility parameter. In this work, the pseudo terms
are also advanced in time using exactly the same time integrator described in (40). The SPH
spatial discretisation for the predictor-corrector system (44a, 44b, 44c, 46, 48a, 48b) will be
presented in the following section.
Remark 5:
In contrast to the artificial compressibility algorithm used in system (48a-48b), an alterna-
tive approach based on a pressure Poisson equation [26, 30] is also presented here for comple-
teness. In this approach, the pressure correction can be re-formulated by combining (46) into
(47), to give
1
κ
(pn+1 pn)
tHn:0pn+1
ρ0.(49)
Substitution of (45a) into (49) for pn+1 yields
1
κ
(pn+1 pn)
ttHn:0pint
ρ0Hn:0DIV 1
ρ0pn+1 pnHn= 0.(50)
16
For evaluation of the pressure increment, equation above however requires the solution of a
system of nonlinear equations at each time step. This is indeed not computationally efficient
when employing a mesh-free SPH discretisation, hence not pursued in this work.
5.2. SPH artificial compressibility algorithm
Following the SPH discretisation procedure presented in Section 3, the predictor step of the
mixed-based system {p,F,H, p}is
pint
apn
a
t=En
aTn
a+D(pn
a); (51a)
Fn+1
aFn
a
t=X
bΛb
apn
b
ρ0Gb(Xa); (51b)
Hn+1
aHn
a
t=Fn
aX
bΛb
apn
b
ρ0Gb(Xa); (51c)
1
˜κ
pint
apn
a
t=Hn
a:X
bΛb
apn
b
ρ0Gb(Xa) + D(Jn
a),(51d)
with the stabilising terms {D(pa),D(Ja)}and the internal and external force vectors {Ta,Ea}
defined in Section 3.3.
In addition, the corrector step of the discrete system {p,F,H, p}becomes
dpa
=X
bΛb
apn+1
bpn
bHn
bGb(Xa)pn+1
apint
a
t; (52a)
1
γ
dpa
=1
˜κ
pint
apn
a
t1
κ
(pn+1
apn
a)
t+Hn
a:X
bΛb
apn+1
b
ρ0pn
b
ρ0Gb(Xa).(52b)
At each time step ∆t, above system (52) is iteratively solved for the linear momentum and
pressure within the pseudo time integration, aims at obtaining convergence to a pseudo steady
state (e.g. dpa
0and dpa
0).
5.3. Iteration speed-up procedure
To accelerate the speed of convergence within the iterative process, we can incorporate an
additional Laplacian (or harmonic) dissipative operator to equation (52b), which results in
1
γ
dpa
=1
˜κ
pint
apn
a
t1
κ
(pn+1
apn
a)
t+Hn
a:X
bΛb
apn+1
b
ρ0pn
b
ρ0Gb(Xa) + Dpseudo(pa).(53)
The discrete pseudo viscosity term at particle acan be written as
Dpseudo(pa) := ηL[p(Xa)] ; η=αc2
st
µ,(54)
where αis a dimensionless user-defined parameter in the range of [0,1]. A challenging aspect
remains as to how to approximate the (particle based) Laplacian dissipative operator to ensure
fulfillment of the global conservation requirement, that is PaVaDpseudo(pa) = 0.
17
Referring to Appendix C, the discrete Laplacian viscosity operator (C.4) used in this work
is approximated as [64]
L[p(Xa)] 2X
bΛb
a
Vbpbpa
kXbXakNab·˜
0˜
Wb(Xa).(55)
Unfortunately, the evaluation of Laplacian operator prevents the exact global conservation due
to the lack of symmetry of the kernel gradient correction, that is ˜
0˜
Wb(Xa)6=˜
0˜
Wa(Xb).
This issue can also be addressed by replacing ˜
0˜
Wb(Xa) with ˜
0˜
WAve
b(Xa) (32).
Remark 6: Following [1], an alternative approach to approximate the (globally conservative)
Laplacian evaluation of the pressure variable pcan be expressed as
L[p(Xa)] X
bΛb
a
Vb[pbpa]˜
0WAve
b(Xa),(56)
where ˜
0WAve
b(Xa) is defined as
˜
0WAve
b(Xa) := 1
2h˜
0Wb(Xa) + ˜
0Wa(Xb)i.(57)
Here, ˜
0is a corrected Laplacian approximation which can be obtained through a least-square
minimisation procedure (see Appendix A in [1] for further details). This requires the solution
of a system of equations and for this reason not pursued in this paper.
Thus, the discrete pseudo viscosity operator can be described as
Dpseudo(pa)2ηX
bΛb
a
Vbpbpa
kXbXakNab·˜
0˜
WAve
b(Xa).(58)
Notice here that the only purpose of adding viscosity operator Dpseudo(pa) to (53) is to accelerate
the speed of convergence within the pseudo time integration process when iteratively solving
the implicit system for pressure correction (52a, 53). This is in clear contrast to the upwinding
stabilisation terms {D(pa),D(Ja)}introduced in the predictor step (51), crucial to ensure the
robustness (stability) of the algorithm.
6. Algorithmic description
For ease of understanding, Algorithm 1 summarises the complete algorithmic description of
the mixed-based {p,F,H, J}Upwind Smooth Particle Hydrodynamics (Upwind-SPH) metho-
dology, with all the necessary numerical ingredients. Notice that simpler {p,F}and {p,F, J}
versions of the algorithm can be easily obtained by neglecting the relevant geometric conserva-
tion laws (i.e. (39b) and/or (39c)).
7. Numerical examples
An ample spectrum of numerical examples is presented in order to examine the performance
of the proposed SPH methodology in compressible, nearly incompressible and truly incompres-
sible scenarios. Specifically, three stabilised mixed-based SPH methodologies are analysed,
namely {p,F},{p,F, J}and {p,F,H, J }Upwind-SPH. For validation purposes, some of
the results are benchmarked against other in-house mixed-based numerical schemes, including:
Finite Element Method [25–27, 30, 35], Finite Volume Method [3, 34, 38, 42] and SPH [1, 2].
18
Algorithm 1: Complete stabilised Upwind-SPH mixed methodology
Input : Un
awhere U= [p F H J]T
Output: Un+1
a,Pn+1
a,xn+1
a
(1) ASSIGN old primary variables: Uold
a=Un
aand xold
a=xn
a
(2) EVALUATE p-wave speed: cp(see References [30, 42])
(3) COMPUTE time increment: t
for TVD-RK time integrator = 1 to 2do
(4) COMPUTE right-hand-side of the mixed-based system:
˙
pa(38a), ˙
Fa(38b), ˙
Ha(38c) and ˙
Ja(38d)
(5) APPLY discrete angular momentum preserving algorithm (see Section 6 of [1])
(6) EVOLVE {Ua,xa}via TVD-RK (40)
(7) COMPUTE first Piola Pa(see Eqn. (12) on pg. 75 in Reference [1])
end
(8) UPDATE {Un+1
a,xn+1
a}(see 40)
(9) COMPUTE first Piola Pn+1
a(see Eqn. (12) on pg. 75 in Reference [1])
Table 1: Numerical values for the relative error of the components of linear momentum and stress as compa-
red to the exact solution, measured with the L2norm. Comparison between the {p,F,H, J }Upwind-SPH,
{p,F,H, J }SUPG-SPH and {p,F,H, J}JST-SPH. Convergence rate calculated using the results of the two
finest meshes.
Error values px
h{p,F,H, J}Upwind {p,F,H, J}SUPG-SPH {p,F,H, J }JST-SPH
1/4 2.490 ×1032.690 ×1036.635 ×103
1/8 8.123 ×1048.987 ×1041.986 ×103
1/16 2.198 ×1042.578 ×1045.697 ×104
1/24 9.342 ×1051.177 ×1042.510 ×104
conv. rate 2.110 1.934 2.022
Error values PxX
h{p,F,H, J}Upwind {p,F,H, J}SUPG-SPH {p,F,H, J }JST-SPH
1/4 4.422 ×1033.964 ×1039.976 ×103
1/8 2.269 ×1032.051 ×1034.156 ×103
1/16 5.554 ×1045.805 ×1041.184 ×103
1/24 2.513 ×1042.456 ×1045.248 ×104
conv. rate 1.956 2.122 2.007
19
(a) (b)
Figure 5: Swinging cube. L2norm convergence of components of (a) Linear momentum; and (b) Stresses at a
particular time t= 0.002 s. Results obtained using a neo-Hookean constitutive model with A=B=C= 1
and U0= 5 ×104using the proposed SPH methodologies. The material properties used are Young’s modulus
E= 17 MPa, density ρ0= 1100 kg/m3, Poisson’s ratio ν= 0.3 and αCFL = 0.3.
7.1. Conservation, consistency and convergence
7.1.1. Swinging cube
As already explored in References [2, 3, 25–27, 30, 34, 38, 42, 58], the first example shows
a cube of unit side length with symmetric boundary conditions (e.g. restricted normal dis-
placement) at faces X= 0, Y= 0 and Z= 0 and skew-symmetric boundary conditions (i.e.
restricted tangential displacement) at faces X= 1m, Y= 1m and Z= 1m (see Figure 3 on
pg. 526 in [2]). The main objective of this example is to show the convergence behaviour of
the proposed {p,F},{p,F, J }and {p,F,H, J}Upwind-SPH methodologies5. A linear elastic
material with Young’s modulus E= 17 MPa and Poisson’s ratio ν= 0.3 is considered. The den-
sity of the material is ρ0= 1100 kg/m3and the Courant-Friedrich-Levy number is αC F L = 0.3.
Following Section 8.2 on pg. 85 of Reference [1], the convergence analysis is carried out by
computing the L2norm of the error between the analytical solution of this problem and the
numerical solution obtained for different values of the particle spacing. Table 1 shows a global
L2convergence analysis of the linear momentum pand the first Piola Kirchhoff stress tensor P
simulated using the {p,F,H, J}Upwind-SPH method6, as compared to the analytical solution
described in equation (52) of [1] (see Section 8.2 on pg. 85). Their corresponding graphical
representations are depicted in Figure 5. As expected, all of the SPH schemes show equal
second order convergence for both linear momentum and the components of the stress tensor.
Both Upwind-SPH and SUPG-SPH methodologies show very similar convergence behaviour,
as the stabilisation terms introduced in both methodologies are very well designed from a mat-
hematical standpoint (e.g. Riemann solver and residual based stabilisation). However, a slight
disadvantage of the SUPG-SPH method is the need to establish a secondary cloud of particles
(e.g. stress points) for the evaluation of the residual based stabilisation term, which would
5Nearly identical results have been obtained for {p,F}and {p,F, J}Upwind-SPH methodologies, hence
not presented.
6Given the fact that A=B=C, all the three components of linear momentum and stresses are of the same
magnitude, namely px=py=pzand PxX =PyY =PzZ .
20
Figure 6: Satellite problem setup
require a non-trivial intervention of the user prior to the analysis run. More interestingly, the
proposed Upwind-SPH method shows better accuracy than the JST-SPH algorithm previously
reported in [1], with the same slope but with a lower translation error (see Figure 5). This is
due to the fact that the Upwind-SPH method requires careful design (via an acoustic Riemann
solver) of the stabilisation terms, without relying on the use of any user defined stabilisation
parameters.
7.1.2. Satellite-like structure
In order to demonstrate the momentum conservation characteristics of the proposed al-
gorithm, the motion of a satellite-like structure originally proposed by [59] is studied. The
structure of a truncated cone of base radius 4 m, of top radius 2 m and of height 3 m, along
with four attached arms of unit cross-section that extend 6.5 m from the centre of the structure
(see Figure 6). The satellite is released without any initial deformation but with an initial
angular velocity of Ω = 1 rad/s about the centre of mass. The velocity field relative to the
centre of mass Xcm is given as
v0(X) = ω×(XXcm) ; ω= (0,0,Ω)T;X= (X, Y , 0) .(59)
A neo-Hookean model is chosen with the material parameters ρ0= 1000 kg/m3, Young’s
modulus E= 50.05 kPa and Poisson’s ratio ν= 0.3. For visualisation purposes, a sequence of
snapshots capturing the deformation of the satellite-like structure is shown in Figure 7. Figure
8 shows the time history of the components of the global angular and linear momenta, where
both fields can be seen to remain constant whilst the structure deforms free from external
effects. Activation of the angular momentum preserving algorithm enables the global angular
momentum to remain constant. We show how the new Riemann-based {p,F}Upwind-SPH
algorithm fulfill global conservation, namely PaVaD(pa) = 0and PaVaTa=0(see Figure
9).
7.2. Spurious pressure
7.2.1. Tensile test
A perforated block is clamped on its bottom surface and is left free on the rest of the
boundaries (see Figure 10a). The main aim of this example is to illustrate the capability of
the proposed Total Lagrangian Upwind-SPH algorithm in surpressing spurious pressure modes.
The block is initially pulled with a linear variation in initial velocity field v0=V[0,0,(Z/H)]T
where V= 100 m/s and H= 0.5 m. This problem is solved using a nearly incompressible
21
(a) t = 0 s (b) t = 1.66 s
(c) t = 5.94 s (d) t = 11.11 s
(e) t = 16 s (f) t = 23.33 s
Figure 7: Satellite-like structure. Time evolution of the deformation along with the pressure distribution using
the mixed-based {p,F}Upwind-SPH. Results obtained with an angular velocity of ω= [0,0,1]Trad/s. A
neo-Hookean constitutive model is employed with Young’s modulus E= 50046 Pa, density ρ0= 1000 kg/m3,
Poisson’s ratio ν= 0.3 and αCFL = 0.3. 22
(a) (b)
Figure 8: Satellite-like structure. Time evolution of the components of (a) Global angular momentum; and (b)
Global linear momentum using the mixed-based {p,F}Upwind-SPH. Results obtained with an angular velocity
of ω= [0,0,1]Trad/s. A neo-Hookean constitutive model is employed with Young’s modulus E= 50046 Pa,
density ρ0= 1000 kg/m3, Poisson’s ratio ν= 0.3 and αCFL = 0.3.
(a) (b)
Figure 9: Satellite-like structure. Time evolution of the components of (a) Global internal force PaVaTa0
and (b) Global Riemann-based dissipation PaVaD(pa)0of linear momentum evolution using the mixed-
based {p,F}Upwind-SPH algorithm. Results obtained with an angular velocity of ω= [0,0,1]Trad/s. A
neo-Hookean constitutive model is employed with Young’s modulus E= 50046 Pa, density ρ0= 1000 kg/m3,
Poisson’s ratio ν= 0.3 and αCFL = 0.3.
23
Z, z
v0=V[0,0,(Z/H)]Tm/s
(b)
Table 3: Final radii of copper bar at t= 80µs. Results obtained using the proposed {p,F}Upwind-SPH
mesh-free algorithm, benchmarked against other published methodologies.
Method Final radius (cm)
Standard 4-Noded Tetrahedra [15] 0.555
Under-integrated 8-Noded Hexahedra [15] 0.695
Average Nodal Pressure 4-Noded Tetrahedra [15] 0.699
Jameson-Schmidt-Turkel Vertex Centred FVM [37] 0.698
Petrov-Galerkin FEM [27] 0.700
Upwind Cell Centred FVM [41] 0.700
Upwind-SPH mesh-free method 0.689
8.2.3. Taylor impact plasticity
Previously explored in References [27, 37, 41], a copper bar of initial length L= 0.0324
m and of initial radius r= 0.0032 m impacts against a rigid wall with a dropping velocity
of 227 m/s, as shown in Figure 17. The main objective of this benchmark problem is to
assess the performance of the proposed mesh-free algorithm in capturing plastic deformation
with application to metal forming. A von Mises hyperelastic-plastic material with isotropic
hardening is chosen to simulate the material. The material parameters are Young’s modulus
E= 117 GPa, density ρ0= 8.930 ×103kg/m3, Poisson’s ratio ν= 0.35, yield stress ¯τ0
y= 0.4
GPa and hardening modulus H= 0.1 GPa. Figure 18 shows the contours of both plastic strain
and pressure at eight different time instants. Extremely smooth pressure field is observed. For
comparison purposes, the final radius of copper bar at time t= 80µs predicted by the {p,F}
Upwind-SPH algorithm is shown in Table 3, benchmarked against other published numerical
results [15, 27, 37]. As reported in References [15, 16], the solutions obtained using the linear 4-
noded tetrahedra suffers from volumetric locking and pressure instabilities. The new mesh-free
method clearly circumvents these issues.
24
Y, y
D= 0.2m
H= 0.5m
Z, z
v0Tm/s
Y, y
D= 0.2m
H= 0.5m
(0.5,0.5,0.5) m(0.5,0.5,0.5) m
X, x (0.5,0.5,0) mX, x (0.5,0.5,0) m
(a)
=V[0,0,(Z/H)]
(a)
Z, z
v0=V[0,0,(Z/H)]Tm/s
(b)
Table 3: Final radii of copper bar at t= 80µs. Results obtained using the proposed {p,F}Upwind-SPH
mesh-free algorithm, benchmarked against other published methodologies.
Method Final radius (cm)
Standard 4-Noded Tetrahedra [15] 0.555
Under-integrated 8-Noded Hexahedra [15] 0.695
Average Nodal Pressure 4-Noded Tetrahedra [15] 0.699
Jameson-Schmidt-Turkel Vertex Centred FVM [37] 0.698
Petrov-Galerkin FEM [27] 0.700
Upwind Cell Centred FVM [41] 0.700
Upwind-SPH mesh-free method 0.689
8.2.3. Taylor impact plasticity
Previously explored in References [27, 37, 41], a copper bar of initial length L= 0.0324
m and of initial radius r= 0.0032 m impacts against a rigid wall with a dropping velocity
of 227 m/s, as shown in Figure 17. The main objective of this benchmark problem is to
assess the performance of the proposed mesh-free algorithm in capturing plastic deformation
with application to metal forming. A von Mises hyperelastic-plastic material with isotropic
hardening is chosen to simulate the material. The material parameters are Young’s modulus
E= 117 GPa, density ρ0= 8.930 ×103kg/m3, Poisson’s ratio ν= 0.35, yield stress ¯τ0
y= 0.4
GPa and hardening modulus H= 0.1 GPa. Figure 18 shows the contours of both plastic strain
and pressure at eight different time instants. Extremely smooth pressure field is observed. For
comparison purposes, the final radius of copper bar at time t= 80µs predicted by the {p,F}
Upwind-SPH algorithm is shown in Table 3, benchmarked against other published numerical
results [15, 27, 37]. As reported in References [15, 16], the solutions obtained using the linear 4-
noded tetrahedra suffers from volumetric locking and pressure instabilities. The new mesh-free
method clearly circumvents these issues.
24
Y, y
D= 0.2m
H= 0.5m
Z, z
v0=V[0,0,(Z/H)]Tm/s
Y, y
D= 0.2m
H= 0.5m
(0.5,0.5,0.5) m(0.5,0.5,0.5) m
X, x (0.5,0.5,0) mX, x (0.5,0.5,0) m
(a)
(b)
Figure 10: Problem setup: (a) Tensile test and (b) Punching test
neo-Hookean material with Young’s modulus E= 1.7×107Pa, density ρ0= 1.1×103kg/m3
and Poisson’s ratio ν= 0.45.
As illustrated in Figure 11, all of the proposed Total Lagrangian SPH methodologies, namely
{p,F},{p,F,J}and {p,F,H,J}Upwind-SPH, are capable of eliminating tensile instability
without showing pressure oscillations. Figure 12 shows the time evolution of the deformation
pattern along with its pressure contour plot with three different views: isometric view, front
view and top view. In this problem, extremely large distortions around the holes of the block
are displayed7. Notice that in this type of problems, mesh-based methods as presented in
[3, 25–27, 30, 34, 38, 42] are not suitable unless adaptive mesh refinement [60] is carried out.
To further examine the efficiency of the algorithm, the same problem is now assessed using
a larger value of Poisson’s ratio ν= 0.499. As presented in Figure 13, the {p,F}Upwind-
SPH methodology shows excessive pressure fluctuations which eventually lead to an incorrect
deformation path. For this reason, the conservation law for the Jacobian Jis incorporated in
order to stabilise the formulation, as already explored in previous publications [1, 2]. A smooth
pressure spatial representation can then be observed when using {p,F, J}and {p,F,H, J}
Upwind-SPH methodologies. For completeness, Figure 14 depicts the time evolution of the
deformation of the problem simulated using the {p,F,H, J}algorithm, displaying a smooth
pressure contour.
7.2.2. Punch test
Similar to the tensile test described in Section 7.2.1, the perforated block is now left free
on its top face and constrained with roller supports (i.e. symmetric boundary conditions) on
the rest of the boundaries (see Figure 10b). The main objective of this example is to show
the ability of the proposed methodology in alleviating severe pressure oscillations in highly
constrained problems. The block is punched on a quarter of the domain with a linear variation
of a velocity field v0=V[0,0,(Z/H)]Twhere V= 250 m/s and H= 0.5 m. A nearly
incompressible neo-Hookean material is used where Young’s modulus E= 1.7×107Pa, density
ρ0= 1.1×103kg/m3and Poisson’s ratio ν= 0.499.
As shown in [1, 2], addition of the volume map conservation law to the {p,F}mixed-based
system seems to be very efficient when solving problems with predominant nearly incompressible
7For simplicity, the inter-particle contact algorithm [56] is not considered in the present manuscript.
24
t = 0.0058 s t = 0.0171 s t = 0.0211 s
(a) Mixed-based {p,F}Upwind-SPH
(b) Mixed-based {p,F,J}Upwind-SPH
(c) Mixed-based {p,F,H,J}Upwind-SPH
Figure 11: Tensile cube. A sequence of deformed configurations using the mixed-based (a) {p,F}, (b) {p,F,
J}and (c) {p,F,H,J}Upwind-SPH. Results obtained with velocity field v0=V[0,0,(Z/H)]Twhere V= 100
m/s and H= 0.5 m. A neo-Hookean constitutive model is employed with Young’s modulus E= 1.7×107Pa,
density ρ0= 1.1×103kg/m3, Poisson’s ratio ν= 0.45 and αCFL = 0.3.
25
t = 0.0266 s t = 0.0330 s t = 0.0417 s
(a) Isometric view
(b) Front view
(c) Top view
Figure 12: Tensile cube. A sequence of deformed configurations in three different views via the {p,F}Upwind-
SPH, namely: (a) Isometric view, (b) Front view, and (c) Top view. Results obtained with velocity field
v0=V[0,0,(Z/H)]Twhere V= 100 m/s and H= 0.5 m. A neo-Hookean constitutive model is employed with
Young’s modulus E= 1.7×107Pa, density ρ0= 1.1×103kg/m3, Poisson’s ratio ν= 0.45 and αCFL = 0.3.
26
t = 0.0058 s t = 0.0171 s t = 0.0211 s
(a) Mixed-based {p,F}Upwind-SPH
(b) Mixed-based {p,F,J}Upwind-SPH
(c) Mixed-based {p,F,H,J}Upwind-SPH
Figure 13: Tensile cube. A sequence of deformed configurations using the mixed-based (a) {p,F}, (b) {p,
F,J}and (c) {p,F,H,J}Upwind-SPH. Results obtained with velocity field v0=V[0,0,(Z/H)]Twhere
V= 100 m/s and H= 0.5 m. A nearly incompressible neo-Hookean constitutive model is employed with
Young’s modulus E= 1.7×107Pa, density ρ0= 1.1×103kg/m3, Poisson’s ratio ν= 0.499 and αCFL = 0.3.
27
t = 0.0266 s t = 0.0330 s t = 0.0417 s
(a) Isometric view
(b) Front view
(c) Top view
Figure 14: Tensile cube. A sequence of deformed configurations in three different views via the {p,F,H,
J}Upwind-SPH: (a) Isometric view, (b) Front view, and (c) Top view. Results obtained with velocity field
v0=V[0,0,(Z/H)]Twhere V= 100 m/s and H= 0.5 m. A nearly incompressible neo-Hookean model is
used with Young’s modulus E= 1.7×107Pa, density ρ0= 1.1×103kg/m3, Poisson’s ratio ν= 0.499 and
αCFL = 0.3. 28
behaviours. In Figure 15, it is clear that the {p,F}Upwind-SPH introduces spurious pressure
modes. These non-physical pressure instabilities can be effectively removed by using alternative
{p,F, J}and {p,F,H, J }stabilised schemes. For completeness, Figures 16 and 17 show a
sequence of deformed states of the problem for a relatively long period of time. Very smooth
pressure field is observed surrounding the holes of the perforated block.
7.2.3. Taylor impact plasticity
Previously presented in References [27, 38, 42], a copper bar of initial length L= 0.0324
m and of initial radius r= 0.0032 m impacts against a rigid wall with a dropping velocity of
227 m/s, as shown in Figure 18. The main objective of this benchmark problem is to assess
the performance of the proposed mesh-free algorithm in capturing large plastic deformations
with application to metal forming [43]. A von Mises hyperelastic-plastic material with isotropic
hardening is chosen for the simulation of this problem. The material parameters are Young’s
modulus E= 117 GPa, density ρ0= 8.930 ×103kg/m3, Poisson’s ratio ν= 0.35, yield stress
¯τ0
y= 0.4 GPa and hardening modulus H= 0.1 GPa. Figure 19 shows the contours of both
plastic strain and pressure at eight different time instants. Extremely smooth pressure field is
observed. For verification purposes, the final radius of the copper bar at time t= 80µs predicted
by the {p,F}Upwind-SPH algorithm is shown in Table 2, benchmarked against other published
numerical results [15, 27, 38]. As shown in References [15, 16], the solutions obtained using
the standard linear 4-noded tetrahedra (being widely used in commercial software) typically
suffers from volumetric locking and pressure instabilities. The new mesh-free method clearly
circumvents these issues.
Table 2: Final radii of copper bar at t= 80µs. Results obtained using the proposed {p,F}Upwind-SPH
mesh-free algorithm, benchmarked against other published methodologies.
Method Final radius (cm)
Standard 4-Noded Tetrahedra [15] 0.555
Under-integrated 8-Noded Hexahedra [15] 0.695
Average Nodal Pressure 4-Noded Tetrahedra [15] 0.699
Jameson-Schmidt-Turkel Vertex Centred FVM [38] 0.698
Petrov-Galerkin FEM [27] 0.700
Upwind Cell Centred FVM [42] 0.700
Upwind-SPH mesh-free method 0.689
7.3. Robustness
7.3.1. Twisting column
In order to examine the robustness of the algorithm, the challenging twisting column ex-
ample previously explored in References [1, 2, 42] is now considered (see Figure 20). The
problem is initialised with a sinusoidal angular velocity field relative to the origin given by
ω0= [0,Ω sin(πY /2L),0]T, where Ω = 105 rad/s and L= 6 m is the length of the column.
A nearly incompressible neo-Hookean material is considered and the material parameters are
Young’s modulus E= 1.7×107Pa, density ρ0= 1.1×103kg/m3and Poisson’s ratio ν= 0.4995.
A particle refinement study is shown in Figure 21. Interestingly, both the deformation
and pressure resolution obtained are practically identical, showing optimal convergence for the
29
t = 0.0018 s t = 0.0066 s t = 0.0113 s
(a) Mixed-based {p,F}Upwind-SPH
(b) Mixed-based {p,F,J}Upwind-SPH
(c) Mixed-based {p,F,H,J}Upwind-SPH
Figure 15: Punch test. A sequence of deformed configurations using the mixed-based (a) {p,F}, (b) {p,F,
J}and (c) {p,F,H,J}Upwind-SPH. Results obtained with velocity field v0=V[0,0,(Z/H)]Twhere
V= 250 m/s and H= 0.5 m. A nearly incompressible neo-Hookean constitutive model is employed with
Young’s modulus E= 1.7×107Pa, density ρ0= 1.1×103kg/m3, Poisson’s ratio ν= 0.499 and αCFL = 0.3.
30
(a) t = 0.0016 s (b) t = 0.0023 s (c) t = 0.0030 s
(d) t = 0.0046 s (e) t = 0.0063 s (f) t = 0.0073 s
(g) t = 0.0089 s (h) t = 0.0100 s (i) t = 0.0113 s
Figure 16: Punch test. A sequence of deformed configurations using the mixed-based {p,F,H, J}Upwind-
SPH. Results obtained with velocity field v0=V[0,0,(Z/H)]Twhere V= 250 m/s and H= 0.5 m. A nearly
incompressible neo-Hookean constitutive model is employed with Young’s modulus E= 1.7×107Pa, density
ρ0= 1.1×103kg/m3, Poisson’s ratio ν= 0.499 and αCFL = 0.3.
31
(a) t = 0.0011 s (b) t = 0.0021 s (c) t = 0.0032 s
(d) t = 0.0041 s (e) t = 0.0054 s (f) t = 0.0068 s
(g) t = 0.0096 s (h) t = 0.0104 s (i) t = 0.0113 s
Figure 17: Punch test. A sequence of deformed configurations (emphasising bottom view) using the mixed-
based {p,F,H, J }Upwind-SPH. Results obtained with velocity field v0=V[0,0,(Z/H)]Twhere V= 250
m/s and H= 0.5 m. A nearly incompressible neo-Hookean constitutive model is employed with Young’s
modulus E= 1.7×107Pa, density ρ0= 1.1×103kg/m3, Poisson’s ratio ν= 0.499 and αCFL = 0.3.
32
X,x
Y,y
v0
(0.0032, 0, 0)
(0.0032, 0.0324, 0)
Z,z
r0
Figure 18: Taylor bar configuration
proposed method. It is also interesting to notice how the methodology preserves perfect axial
rotation without introducing out-of-axis characteristics. This can be easily shown in Figure 22
by monitoring the evolution of the horizontal velocity and horizontal displacement components
at point X= [0.5,0.5,6]Tm, both within zero machine accuracy. For benchmarking purposes,
Figure 23 depicts a comparison of the new Upwind-SPH methodology against the recently
proposed JST-SPH [1], SUPG-SPH [2] mesh-free methods and other in-house mixed-based
methodologies. All of the schemes described above produce very similar results in terms of
deformed shape and pressure field.
In comparison to the Upwind-SPH and SUPG-SPH schemes, it is clear that the JST-SPH
(see Figure 23c) requires a considerably larger number of particles in order to capture the
correct deformation pattern of the column. The results simulated using the SUPG-SPH (see
Figure 23b) are in good agreement with the Upwind-SPH scheme (see Figure 23a). However,
the SUPG-SPH method requires the generation of a secondary set of particles for the evaluation
of the SUPG stabilisation, thus resulting in a greater computational cost.
The problem becomes significantly more challenging by increasing the initial angular velocity
now to Ω = 200 rad/s with a Poisson’s ratio of ν= 0.4995. A refinement study is also carried
out in Figure 24 using a sequentially refined number of particles of 5 ×5×30, 6 ×6×36 and
7×7×42. Notably, the number of twists shown in the column is captured extremely well even
with a small number of particles. As illustrated in the top view of the simulation of Figure 24,
no out-of-plane deformation can be observed.
Finally, we can further examine the robustness of the algorithm by increasing the value of
Poisson’s ratio to the limit such that ν0.5. Its main objective is to show the efficiency of the
artificial compressibility approach (see Section 5) in comparison to the explicit Upwind-SPH,
especially in problems characterised by nearly or truly incompressible behaviours. Figure 25
shows the qualitative comparison between those two approaches. As expected, the artificial
compressibility algorithm introduces a larger numerical dissipation due to the use of the (con-
33
(a) t= 10µs (b) t= 20µs (c) t= 30µs (d) t= 40µs
(e) t= 50µs (f) t= 60µs (g) t= 70µs (h) t= 80µs
Figure 19: Taylor impact. Time evolution of plastic strain in the left quarter and pressure distribution in the
right quarter of the domain along with the deformation. Results obtained using the proposed {p,F}Upwind-
SPH algorithm with dropping velocity v0= (0,227,0)Tm/s. A hyperelastic-plastic constitutive model is
employed with Young’s modulus E= 117 GPa, density ρ0= 8930 kg/m3, Poisson’s ratio ν= 0.35, yield stress
¯τ0
y= 0.4 GPa, hardening modulus H= 0.1 GPa and αCFL = 0.3.
34
X, x
Y, y
(0.5,0,0.5)m
(0.5,6,0.5)m
Z, z
ω0= Ω [0,sin(πY /2H),0]Trad/s
H= 6 m
Figure 20: Twisting column problem setup
servative) Laplacian viscosity term described in (58). A particle refinement study is carried
out in Figure 26, with a progressive level of refinement. It is remarkable that the deformation
pattern predicted with a small number of particles agrees extremely well with those results
obtained using a fine discretisation. The accuracy of the pressure contour is clearly enhanced
as we increase the number of particles. Figure 27 shows the importance of adding (particle ba-
sed) pseudo viscosity term to the artificial compressibility approach, with the primary aim to
speed up the iterative process of the algorithm. As presented in Figure 27, the use of viscosity
term dramatically reduces the number of iterations required for convergence within the pseudo
time integration when solving truly incompressible solids.
7.3.2. Complex geometry
In the last example of this paper, we demonstrate the robustness of the proposed SPH
algorithm on a complex geometry displayed in Figure 28. As reported in Reference [61], the
geometry used in this example is a simplified version of a cardiovascular stent used in biomedical
applications. The structure has an initial outer diameter of 20 mm, a thickness of 0.5 mm and
a total length of 20 mm. The diameter of every hole is 2 mm. In this problem, we study
the deformation pattern of this stent-like structure by applying a velocity field at the top and
bottom of the structure, described as follows
v0=(h0,0,abγ
(γ+1)2iTif t0.03 s
0otherwise ;γ= exp [a(ct)],(60)
where a= 800, b= 0.006 and c= 0.015. Due to the presence of three symmetry planes, only
one eighth of the problem is solved with appropriate boundary conditions. The structure is
made of a nearly incompressible neo-Hookean material with density ρ= 1100 kg/m3, Young’s
Modulus E= 17 MPa and Poisson’s ratio ν= 0.45. Figure 29 shows the overall deformation of
a structure at time t= 0.0216 s. It is remarkable that the pressure is extremely well captured
35
Top view
(a) 4 ×4×24 (b) 5 ×5×30 (c) 6 ×6×36
Figure 21: Twisting column. A sequence of particle refinement using the mixed-based {p,F,H, J }Upwind-
SPH at time step t= 0.1s: (a) 4 ×4×24, (b) 5 ×5×30, and (c) 6 ×6×36 particles. Results obtained with an
angular velocity field ω0= [0,Ω sin (πY /2L),0] where Ω = 105 rad/s and L= 6 m. A nearly incompressible
neo-Hookean constitutive model is employed with Young’s modulus E= 1.7×107Pa, density ρ0= 1.1×103
kg/m3, Poisson’s ratio ν= 0.4995 and αCFL = 0.3.
36
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
-5
510-10
(a)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
-1
0
110-10
(b)
Figure 22: Twisting column. Time evolution of the components of (a) Velocity and (b) Displacement at point
X= [0.5,0.5,6]Tm via the mixed-based {p,F,H, J }Upwind-SPH algorithm. A nearly incompressible neo-
Hookean constitutive model is employed with Young’s modulus E= 1.7×107Pa, density ρ0= 1.1×103kg/m3,
Poisson’s ratio ν= 0.4995 and αCFL = 0.3.
even with a minimum of two particles across the thickness. As can be seen in Figure 30, the
mixed-based {p,F,H, J}Upwind-SPH method produces reliable results that are freed from
zero energy modes. This will open up interesting possibilities for modelling in the field of
biomechanics [62, 63], where this consideration is very relevant.
8. Conclusions
This paper presents a new Upwind Smooth Particle Hydrodynamics (Upwind-SPH) com-
putational methodology for the analysis of nearly and truly incompressible large deformations
in explicit fast solid dynamics. The methodology is established starting from a mixed-based
system of Total Lagrangian first order conservation laws, where the linear momentum pconser-
vation law is solved along with conservation equations for the extended set of geometric strain
measures {F,H, J}.
In this work, a Riemann solver based spatial discretisation has been employed for stabi-
lisation, offering a series of advantages over the recently proposed JST-SPH method [1] and
SUPG-SPH method [2]. First, the nature of the Riemann based upwinding stabilisation does
not explicitly rely on the use of any artificial stabilisation parameters. This differs from the
JST-SPH and SUPG-SPH methodologies, as the former requires the evaluation of user-defined
harmonic and bi-harmonic operators and the latter requires the selection of a number of user-
defined stabilisation parameters for the evaluation of the SUPG stabilisation. Second, the
Upwind-SPH framework is more accurate than its JST-SPH counterpart, and its accuracy is
comparable to that of the SUPG-SPH but with a reduced computational cost.
It has also been shown that the resulting Upwind-SPH algorithm overcomes a number of
numerical difficulties posed by the classical SPH method, namely non-physical hydrostatic pres-
sure fluctuations, hour-glassing and numerical issues associated with conservation, consistency,
long term instability and convergence. As a result, the new algorithm provides a good balance
37
(a) (b) (c) (d) (e)
Pressure (Pa)
Figure 23: Twisting column. Comparison of deformed shapes plotted with pressures at time t= 0.1 s using:
(a) Mixed-based {p,F}Upwind-SPH; (b) Mixed-based {p,F,H, J}SUPG-SPH-H1 (τF= ∆t,ξF= 0.2,
τp= 0.1∆t); (c) Mixed-based {p,F}JST-SPH (ε(2)
p= 0 and ε(4)
p=1
8); (d) PG-FEM [26]; and (e) Constrained-
TOUCH [42]. Results obtained with an angular velocity field ω0= [0,Ω sin(πY/2L),0] where Ω = 105 rad/s
and L= 6 m. A neo-Hookean material is used with density ρ0= 1100 kg/m3, Young’s modulus E= 17 MPa
and Poisson’s ratio ν= 0.495.
38
Top view
(a) 5 ×5×30 (b) 6 ×6×36 (c) 7 ×7×42
Figure 24: Twisting column. A sequence of particle refinement using the mixed-based {p,F,H, J }Upwind-
SPH at time step t= 0.1 s: (a) 5 ×5×30, (b) 6 ×6×36, and (c) 7 ×7×42 particles. Results obtained with an
angular velocity field ω0= [0,Ω sin (πY /2L),0] where Ω = 200 rad/s and L= 6 m. A nearly incompressible
neo-Hookean constitutive model is employed with Young’s modulus E= 1.7×107Pa, density ρ0= 1.1×103
kg/m3, Poisson’s ratio ν= 0.4995 and αCFL = 0.3.
39
Top view
(a) Artificial compressibility approach (b) Explicit Upwind-SPH
Figure 25: Twisting column. A comparison between (a) {p,F,H, p}artificial compressibility approach and
(b) Explicit {p,F,H, J }Upwind-SPH at time step t= 0.1s. Results obtained with an angular velocity
field ω0= [0,Ω sin (πY /2L),0] where Ω = 105 rad/s and L= 6 m. A nearly incompressible neo-Hookean
constitutive model is employed with Young’s modulus E= 1.7×107Pa, density ρ0= 1.1×103kg/m3,
Poisson’s ratio ν= 0.4999 and αCFL = 0.3.
40
Top view
(a) 4 ×4×24 (b) 6 ×6×36 (c) 8 ×8×48
Figure 26: Twisting column. A sequence of particle refinement using the mixed-based {p,F}Upwind-SPH
through artificial compressibility at time step t= 0.1s: (a) 4 ×4×24, (b) 6 ×6×36, and (c) 8 ×8×48 particles.
Results obtained with an angular velocity field ω0= [0,Ω sin (πY /2L),0] where Ω = 105 rad/s and L= 6 m.
A truly incompressible neo-Hookean constitutive model is employed with Young’s modulus E= 1.7×107Pa,
density ρ0= 1.1×103kg/m3, Poisson’s ratio ν= 0.5 and αCFL = 0.3.
41
0 0.1 0.2 0.3 0.4 0.5
8
8.2
8.4
8.6
8.8
9
0 2 4 6 8
4
4.5
5
5.5
6106
(a)
22
24
26
28
30
32
1
1.5 107
0.5
0 10 20 30
20
0 0.1 0.2 0.3 0.4 0.5
(b)
0 0.1 0.2 0.3 0.4 0.5
20
40
60
80
100
120
140
160
180
0 10 20 30
0.5
1
1.5
2107
(c)
0 0.1 0.2 0.3 0.4 0.5
0
100
200
300
400
500
600
700
800
900
0.1 0.2 0.3 0.4 0.5
20
30
40
0 10 20 30
0.5
1
1.5
2107
(d)
Figure 27: Twisting column. The effect of incorporating Laplacian viscosity (54) to the artificial compressibility
approach with the primary aim to enhance the iterative process of the algorithm in various scenarios: (a)
ν= 0.499, (b) ν= 0.4999, (c) ν= 0.49999 and (d) ν= 0.5. Results obtained for the first step of Runge-Kutta
time integrator with an angular velocity field ω0= [0,Ω sin (πY /2L),0] where Ω = 105 rad/s and L= 6 m.
The green curve represents the number of iterations required to achieve steady state convergence in pseudo-
time for a fixed Laplacian viscosity α. A neo-Hookean constitutive model is employed with Young’s modulus
E= 1.7×107Pa, density ρ0= 1.1×103kg/m3and αCFL = 0.3.
42
Figure 28: Stent-like structure [61]. The Computer Aided Design (CAD) file can be downloaded from GrabCAD.
(a) Front view (b) Top view (c) Interior view
Figure 29: Stent-like structure. Deformed configuration of the structure at t= 0.0216 s simulated using the
mixed-based {p,F,H,J}Upwind-SPH algorithm. A nearly incompressible neo-Hookean constitutive model
is employed with Young’s modulus E= 0.9 MPa, density ρ0= 1000 kg/m3, Poisson’s ratio ν= 0.45 and
αCFL = 0.3.
43
t = 0 s t = 0.0128 s t = 0.0148 s
t = 0.0168 s t = 0.0189 s t = 0.0208 s
t = 0.0233 s t = 0.0248 s t = 0.0026 s
Figure 30: Stent-like structure. A sequence of deformed configurations using the mixed-based {p,F,H,J}
Upwind-SPH algorithm. A nearly incompressible neo-Hookean constitutive model is employed with Young’s
modulus E= 0.9 MPa, density ρ0= 1000 kg/m3, Poisson’s ratio ν= 0.45 and αCFL = 0.3.
44
between accuracy and speed of computation. In terms of computational efficiency, an adapted
artificial compressibility approach has been formulated particularly useful when dealing with
extremely large (or even infinite) wave speeds in the incompressible limit.
Finally, a comprehensive set of numerical examples has been presented in order to ben-
chmark the results obtained against other numerical methodologies, including Finite Element
Method [25–27, 30, 35], Finite Volume Method [3, 34, 38, 42] and Smooth Particle Hydrody-
namics Method [1, 2]. Both velocities and (volumetric and deviatoric) stresses display equal
second order of convergence, advantageous in those scenarios where stresses are of primary
interest. The consideration of thermoelasticity within the current mesh-free computational fra-
mework is the next step of our work.
Acknowledgements
The authors gratefully acknowledge the financial support provided by the Sˆer Cymru Natio-
nal Research Network for Advanced Engineering and Materials, United Kingdom. The third
author acknowledges the financial support received through the European Commission EA-
CEA Agency, Framework Partnership Agreement 2013-0043 Erasmus Mundus Action 1b, as a
part of the EM Joint Doctorate Simulation in Engineering and Entrepreneurship Development
(SEED).
Appendix A. von Mises plasticity model
The standard algorithm to ensure that the Kirchhoff stress satisfies a von Mises type of
plastic constraint is summarised here for completeness in Algorithm 2.
Appendix B. Discrete angular momentum preserving algorithm
The resulting Upwind-SPH algorithm does not intrinsically fulfil conservation of angular
momentum, since the strain measures {F,H, J}are no longer exclusively obtained from the
current geometry [42]. Specifically, the deformation gradient F6=Fx:= 0x, its co-factor
H6=Hx:= 1
2FxFxand its Jacobian J6=Jx:= det Fx.
In this paper, and following the work of Jibran et al. [42], a monolithic discrete angular
momentum projection algorithm is presented. Specifically, the local internal nodal force Tais
modified (in a least-square sense) in order to preserve the total angular momentum, whilst still
ensuring the global conservation of linear momentum.
Following Reference [42], sufficient conditions for the global preservation of the discrete
linear and angular momentum within a time step are enforced at each stage of the one-step
two-stage Runge Kutta time integrator (40) described as:
X
a
VaTχ
a=0;X
a
VaXa×Tχ
a=0;Xa=xn
a, χ =n
xn
a+t
2ρ0(pn
a+p?
a), χ =?; (B.1)
where χ={n, ?}.
A least-square minimisation procedure is used to obtain a modified set of internal nodal
forces ˆ
Tathat satisfy the above conditions (B.1). This can be achieved by computing the
45
Algorithm 2: Time update of first Piola Kirchoff stress tensor
Input : Fn+1,C1
pn, εn
p
Output: Pn+1
(1) Obtain Jacobian of deformation: Jn+1
F= det Fn+1
(2) Evaluate pressure: pn+1 =κln (Jn+1
F)
Jn+1
F
(3) Compute trial elastic left strain tensor: ¯
bn+1
e=Fn+1 C1
pnFTn+1
(4) Spectral decomposition of ¯
bn+1
e:¯
λe,i ,¯ni¯
bn+1
e=P3
i=1 ¯
λe,i2(¯ni¯ni)
(5) Obtain trial deviatoric Kirchoff stress tensor:
¯τ0=P3
i=1 ¯τ0
ii (¯ni¯ni),¯τ0
ii = 2µln ¯
λe,i2
3µln Jn+1
F
(6) Obtain yield criterion: f¯τ0, εn
p=q3
2(¯τ0:¯τ0)τ0
y+Hεn
p
(7) Compute direction vector and plastic multiplier:
if f¯τ0, εn
p>0then
Direction vector: υn+1
i=¯τ0
ii
2
3(¯τ0:¯τ0)
Plastic multiplier: γ=f(¯τ0n
p)
3µ+H
else
υn+1
i= ∆γ= 0
end
(8) Evaluate elastic stretch: λn+1
e,i = exp ln(¯
λe,i)γυn+1
i
(9) Set spatial normals: nn+1
i=¯ni
(10) Compute Kirchoff stresss tensor:
τn+1 =P3
i=1 τii nn+1
inn+1
i
τii =τ0
ii +Jn+1
Fpn+1, τ 0
ii =12µγ
2
3(¯τ0:¯τ0)¯τ0
ii
(11) Evaluate first Piola Kirchoff stress tensor: Pn+1 =τn+1 FTn+1
(12) Update elastic left strain tensor: bn+1
e=P3
i=1 λn+1
e,i 2nn+1
inn+1
i
(13) Compute plastic right Cauchy Green tensor: C1
pn+1 =F1n+1 bn+1
eFTn+1
(14) Update plastic strain: εn+1
p=εn
p+ ∆γ
46
minimum of the following functional [38, 42] (ignoring time arguments for brevity):
ΠT(ˆ
Ta,λang,λlin) = 1
2X
a
Va(ˆ
TaTa)·(ˆ
TaTa)
λang · X
a
VaXa׈
Ta!λlin · X
a
Vaˆ
Ta!.
(B.2)
After some simple algebra, a modified set of internal nodal forces ˆ
Taarise:
ˆ
Ta=Ta+λang ×Xa+λlin.(B.3)
The Lagrange multipliers {λang,λlin}are the solutions of the following system of equations
PaVa[(Xa·Xa)IXaXa]PaVaˆ
Xa
PaVaˆ
XaPaVaλang
λlin =PaVaXa×Ta
PaVaTa,(B.4)
with the indicial notation hˆ
Xaiik =Eijk [Xa]j.
In addition, an extra condition must be added for the satisfaction of the global angular
momentum preservation, namely
X
a
VaD(pa) = 0;X
a
VaXa×D(pa) = 0.(B.5)
Similarly to the least-square minimisation procedure described above, a set of modified upwin-
ding dissipation terms can now be obtained by replacing {ˆ
Ta,Ta,λlin,λang}in (B.3) and (B.4)
with {ˆ
D(pa),D(pa),µDp,λDp}.
Appendix C. Laplacian evaluation
The artificial compressibility procedure to be employed in this paper to effectively handle
near incompressibility regimes may require the evaluation of the Laplacian (harmonic operator
[1]) of a solution function. For this purpose, consider the Laplacian of any arbitrary scalar
function gto be numerically approximated as
L[g(Xa)] := 0·[0g(Xa)] X
bΛb
a
0g(Xb)·Gb(Xa).(C.1a)
Addition of the redundant term 0g(Xa) to (C.1a) yields an alternative expression as
L[g(Xa)] X
bΛb
a
20g(Xa) + 0g(Xb)
2·Gb(Xa).(C.2)
It is worth pointing out that the squared bracket term shown in (C.2) represents the gradient
approximation at the mid-edge connecting particles aand b. This can be further approximated
via a second order central difference scheme as [64]
0g(Xa) + 0g(Xb)
2gbga
kXbXakNab,Nab := XbXa
kXbXak.(C.3)
47
For the SPH method under consideration, the normal Nab is defined through a direction vector
between particles aand b, as depicted in Figure 3b.
Finally, substitution of (C.3) into (C.2) yields the discrete version of the Laplacian evalua-
tion
L[g(Xa)] 2X
bΛb
agbga
kXbXakNab·Gb(Xa).(C.4)
48
[1] C. H. Lee, A. J. Gil, G. Greto, S. Kulasegaram, J. Bonet, A new Jameson-Schmidt-Turkel
Smooth Particle Hdrodynamics algorithm for large strain explicit fast dynamics, Computer
Methods in Applied Mechanics and Engineering 311 (2016) 71–111.
[2] C. H. Lee, A. J. Gil, O. I. Hassan, J. Bonet, S. Kulasegaram, A variationally consistent
Streamline Upwind Petrov Galerkin Smooth Particle Hydrodynamics algorithm for large
strain solid dynamics, Computer Methods in Applied Mechanics and Engineering 318
(2017) 514–536.
[3] M. Aguirre, A. J. Gil, J. Bonet, C. H. Lee, An upwind vertex centred finite volume solver
for Lagrangian solid dynamics, Journal of Computational Physics 300 (2015) 387–422.
[4] P. W. Randles, L. D. Libersky, Smoothed Particle Hydrodynamics: Some recent impro-
vements and applications, Computer Methods in Applied Mechanics and Engineering 139
(1996) 375–408.
[5] G. R. Johnson, S. R. Beissel, Normalized smoothing functions for SPH impact computa-
tions, International Journal for Numerical Methods in Engineering 39 (1996) 2725–2741.
[6] Y. Vidal, J. Bonet, A. Huerta, Stabilized updated Lagrangian corrected SPH for explicit
dynamic problems, International Journal for Numerical Methods in Engineering 69 (2006)
2687–2710.
[7] J. Bonet, S. Kulasegaram, Remarks on tension instability of Eulerian and Lagrangian
Corrected Smooth Particle Hydrodynamics (CSPH) methods, International Journal for
Numerical Methods in Engineering 52 (2001) 1203–1220.
[8] G. C. Ganzenm¨uller, An hourglass control algorithm for Lagrangian Smooth Particle
Hydrodynamics, Computer Methods in Applied Mechanics and Engineering 286 (2015)
87–106.
[9] G. C. Ganzenm¨uller, S. Hiermaier, M. May, On the similarity of meshless discretizations of
Peridynamics and Smooth-Particle Hydrodynamics, Computers and Structures 150 (2015)
71–78.
[10] S. A. Silling, Reformulation of elasticity theory for discontinuities and long-range forces,
Journal of the Mechanics and Physics of Solids 48 (2000) 175–209.
[11] J. J. Monaghan, SPH without a tensile instability, Journal of Computational Physics 159
(2000) 290–311.
[12] M. A. Puso, J. S. Chen, E. Zywicz, W. Elmer, Meshfree and finite element nodal integration
methods, International Journal for Numerical Methods in Engineering 74 (2008) 416–446.
[13] J. W. Swegle, D. L. Hicks, S. W. Attaway, Smoothed Particle Hydrodynamics stability
analysis, Journal of Computational Physics 116 (1995) 123–134.
[14] T. Belytschko, Y. Guo, W. K. Liu, S. P. Xiao, A unified stability analysis of meshless
particle methods, International Journal for Numerical Methods in Engineering 48 (2000)
1359–1400.
49
[15] J. Bonet, A. J. Burton, A simple average nodal pressure tetrahedral element for incom-
pressible and nearly incompressible dynamic explicit applications, Communications in
Numerical Methods in Engineering 14 (1998) 437–449.
[16] M. W. Gee, C. R. Dohrmann, S. W. Key, W. A. Wall, A uniform nodal strain tetrahedron
with isochoric stabilization, International Journal for Numerical Methods in Engineering
78 (2009) 429–443.
[17] C. R. Dohrmann, M. W. Heinstein, J. Jung, S. W. Key, W. R. Witkowski, Node-based
uniform strain elements for three-node triangular and four-node tetrahedral meshes, In-
ternational Journal for Numerical Methods in Engineering 47 (2000) 1549–1568.
[18] M. A. Puso, J. Solberg, A stabilized nodally integrated tetrahedral, International Journal
for Numerical Methods in Engineering 67 (2006) 841–867.
[19] J. J. Monaghan, Smoothed particle hydrodynamics, Annual Review of Astronomy and
Astrophysics 30 (1992) 543–574.
[20] P. W. Randles, L. D. Libersky, Normalized SPH with stress points, International Journal
for Numerical Methods in Engineering 48 (2000) 1445–1462.
[21] J. S. Chen, C. T. Wu, S. Yoon, Y. You, A stabilized conforming nodal integration for
Galerkin mesh-free methods, International Journal for Numerical Methods in Engineering
50 (2001) 435–466.
[22] M. B. Liu, G. R. Liu, K. Y. Lam, Constructing smoothing functions in smoothed particle
hydrodynamics with applications, Jounal of Computational and Applied Mathematics 155
(2003) 263–284.
[23] J. K. Chen, J. E. Beraun, T. C. Carney, A corrective smoothed particle method for
boundary value problems in heat conduction, International Journal for Numerical Methods
in Engineering 46 (1999) 231–252.
[24] J. Bonet, S. Kulasegaram, Correction and stabilization of smooth particle hydrodyna-
mics methods with applications in metal forming simulations, International Journal for
Numerical Methods in Engineering 47 (2000) 1189–1214.
[25] J. Bonet, A. J. Gil, C. H. Lee, M. Aguirre, R. Ortigosa, A first order hyperbolic frame-
work for large strain computational solid dynamics. Part I: Total Lagrangian isothermal
elasticity, Computer Methods in Applied Mechanics and Engineering 283 (2015) 689–732.
[26] A. J. Gil, C. H. Lee, J. Bonet, M. Aguirre, A stabilised Petrov-Galerkin formulation for
linear tetrahedral elements in compressible, nearly incompressible and truly incompressible
fast dynamics, Computer Methods in Applied Mechanics and Engineering 276 (2014) 659–
690.
[27] C. H. Lee, A. J. Gil, J. Bonet, Development of a stabilised Petrov-Galerkin formulation
for conservation laws in Lagrangian fast solid dynamics, Computer Methods in Applied
Mechanics and Engineering 268 (2014) 40–64.
[28] R. Courant, K. Friedrichs, H. Lewy, On the partial difference equations of mathematical
physics, Mathematische Annalem 100 (1928) 32–74.
50
[29] A. J. Gil, Z. Zhang, O. Hassan, K. Morgan, Parallel multigrid detached eddy simula-
tion algorithm for three-dimensional unsteady incompressible flows on unstructured grids,
Journal of Aerospace Engineering 19 (2006) 271–280.
[30] A. J. Gil, C. H. Lee, J. Bonet, R. Ortigosa, A first order hyperbolic framework for large
strain computational solid dynamics. Part II: Total Lagrangian compressible, nearly in-
compressible and truly incompressible elasticity, Computer Methods in Applied Mechanics
and Engineering 300 (2016) 146–181.
[31] A. J. Chorin, The numerical solution of the Navier-Stokes equations for an incompressible
fluid, Bulletin of the American Mathematical Society 73 (1967) 928–931.
[32] A. J. Chorin, Numerical solution of the Navier-Stokes equations, Mathematics of Compu-
tation 22 (1968) 745–762.
[33] A. J. Chorin, On the convergence of discrete aapproximations to the Navier-Stokes equa-
tions, Mathematics of Computation 23 (1969) 341–353.
[34] C. H. Lee, A. J. Gil, J. Bonet, Development of a cell centred upwind finite volume algorithm
for a new conservation law formulation in structural dynamics, Computers and Structures
118 (2013) 13–38.
[35] I. A. Karim, C. H. Lee, A. J. Gil, J. Bonet, A two-step Taylor Galerkin formulation for
fast dynamics, Engineering Computations 31 (2014) 366–387.
[36] G. Kluth, B. Despr´es, Discretization of hyperelasticity on unstructured mesh with a cell-
centered Lagrangian scheme, Journal of Computational Physics 229 (2010) 9092–9118.
[37] N. C. Nguyen, J. Peraire, Hybridizable Discontinuous Galerkin methods for partial diffe-
rential equations in continuum mechanics, Journal of Computational Physics 231 (2012)
5955–5988.
[38] M. Aguirre, A. J. Gil, J. Bonet, A. A. Carre˜no, A vertex centred finite volume Jameson-
Schmidt-Turkel (JST) algorithm for a mixed conservation formulation in solid dynamics,
Journal of Computational Physics 259 (2014) 672–699.
[39] J. Bonet, A. J. Gil, R. Ortigosa, On a tensor cross product based formulation of large
strain solid mechanics, International Journal of Solids and Structures 84 (2016) 49–63.
[40] J. Bonet, A. J. Gil, R. Ortigosa, A computational framework for polyconvex large strain
elasticity, Computer Methods in Applied Mechanics and Engineering 283 (2015) 1061–
1094.
[41] J. A. Trangenstein, P. Colella, A higher-order Godunov method for modelling finite de-
formation in elastic-plastic solids, Communications on Pure and Applied Mathematics 44
(1991) 41–100.
[42] J. Haider, C. H. Lee, A. J. Gil, J. Bonet, A first order hyperbolic framework for large
strain computational solid dynamics: An upwind cell centred Total Lagrangian scheme,
International Journal for Numerical Methods in Engineering 109 (2017) 407–456.
51
[43] J. Bonet, A. J. Gil, R. D. Wood, Nonlinear Solid Mechanics for Finite Element Analysis:
Statics, Cambridge University Press, 2016.
[44] J. E. Marsden, T. J. R. Hughes, Mathematical foundations of elasticity, Dover Publications,
1994.
[45] J. M. Ball, Energy-minimising configurations in nonlinear elasticity, Archive for Rational
Mechanics and Analysis 63 (1976) 337–403.
[46] J. M. Ball, Convexity conditions and existence theorems in nonlinear elasticity, Archive
for Rational Mechanics and Analysis 63 (1976) 337–403.
[47] J. Bonet, T.-S. Lok, Variational and momentum preservation aspects of smooth particle
hydrodynamic formulations, Computer Methods in Applied Mechanics and Engineering
180 (1999) 97–115.
[48] G. R. Johnson, R. A. Stryk, S. R. Beissel, SPH for velocity impact computations, Computer
Methods in Applied Mechanics and Engineering 139 (1996) 347–373.
[49] R. L. Leveque, Finite volume methods for hyperbolic problems, Cambridge University
Press, 2002.
[50] E. F. Toro, Riemann solvers and numerical methods for fluid dynamics: A practical intro-
duction, Springer-Verlag, second edition, 2006.
[51] A. Huerta, S. Fern´andez-M´endez, Enrichment and coupling of the finite element and
meshless methods, International Journal for Numerical Methods in Engineering 48 (2000)
1615–1636.
[52] A. Huerta, S. Fern´andez-M´endez, Locking in the incompressible limit for the element-free
Galerkin method, International Journal for Numerical Methods in Engineering 51 (2001)
1361–1383.
[53] K. J. Bathe, Finite element procedures, Prentice Hall, 1996.
[54] S. Fern´andez-M´endez, A. Huerta, Continuous blending of SPH with finite elements, Com-
putes and Structures 83 (2005) 1448–1458.
[55] J. P. Gray, J. J. Monaghan, R. P. Swift, SPH elastic dynamics, Computer Methods in
Applied Mechanics and Engineering 190 (2001) 6641–6662.
[56] R. Vignjevic, J. Campbell, L. Libersky, A treatment of zero-energy modes in the smoothed
particle hydrodynamics method, Computer Methods in Applied Mechanics and Engineer-
ing 184 (2000) 67–85.
[57] M. A. Puso, A highly efficient enhanced assumed strain physically stabilized hexahedral
element, International Journal for Numerical Methods in Engineering 49 (2000) 1029–1064.
[58] G. Scovazzi, B. Carnes, X. Zeng, S. Rossi, A simple, stable, and accurate linear tetrahedral
finite element for transient, nearly and fully incompressible solid dynamics: A dynamic va-
riational multiscale approach, International Journal for Numerical Methods in Engineering
106 (2016) 799–839.
52
[59] J. C. Simo, N. Tarnow, A new energy and momentum conserving algorithm for the non-
linear dynamics of shells, International Journal for Numerical Methods in Engineering 37
(1994) 2527–2549.
[60] Y. Onishi, K. Amaya, A locking-free selective smoothed finite element method using
tetrahedral and triangular elements with adaptive mesh rezoning for large deformation
problems, International Journal for Numerical Methods in Engineering 99 (2014) 354–371.
[61] E. Boatti, G. Scalet, F. Auricchio, A three-dimensional finite-strain phenomenological
model for shape-memory polymers: Formulation, numerical simulations, and comparison
with experimental data, International Journal of Plasticity 83 (2016) 153–177.
[62] A. J. Gil, A. A. Carre˜no, J. Bonet, O. Hassan, The Immersed Structural Potential Method
for haemodynamic applications, Journal of Computational Physics 229 (2011) 8613–8641.
[63] A. J. Gil, , A. A. Carre˜no, J. Bonet, O. Hassan, An enhanced Immersed Structural
Potential Method for fluid-structure interaction, Journal of Computational Physics 250
(2013) 178–205.
[64] J. P. Morris, P. J. Fox, Y. Zhu, Modeling low Reynolds number incompressible flows using
SPH, Journal of Computational Physics 136 (1997) 214–226.
53
... External body force term C 0 Corrective matrix that ensures first-order consistency C 1 − C 4 Integration constants C C F L The Courant number C cp Compression or volumetric wave speed C s Linearized sound speed in the structure C sh Shear wave speed d Displacement D ...
... The issue of tensile instability or unphysical modes in tensile stress states is likely linked to the use of an Eulerian kernel function at the current nodal positions which may potentially lead to negative eigenvalues [1,2]. The Total Lagrangian SPH (TLSPH) [3,4] or Updated Reference Lagrangian SPH [5,6] are free from tensile instability because the kernel and its derivative remain invariant with respect to a constant reference configuration. Regarding the rank deficiency or zero-energy modes, this issue is not exclusively related to SPH and is an issue with computational methods that are based on collocated integration, including approximations of field variables and their derivatives at the same computational points. ...
... Despite the improved accuracy in computing deformation measures, expression (14) still exhibits spurious hourglass or zero-energy modes [4,8,9] due to the inherent underintegration of the SPH method. To address this issue, a consistent upwinding or Riemann-based numerical stabilization, denoted by a i , is introduced to ensure the discrete satisfaction of the second law of thermodynamics within the SPH scheme and to mitigate the hourglass or zero-energy modes. ...
Article
Full-text available
The paper presents a variationally consistent Total Lagrangian SPH (TLSPH) model for non-linear and finite strain elastic structural dynamics. To enhance the approximation of stresses (dynamics) and thus accelerations (kinematics), the deformation gradient computation is enhanced to ensure second-order completeness and accordingly the discretized momentum equation is also reformulated so that the accelerations are consistently obtained with respect to the targeted second-order completeness. In addition, a novel Riemann stabilization term is proposed with respect to the rank deficiency of TLSPH particularly for challenging structural dynamics problems including fast dynamics and kinematical discontinuities. The new Riemann term is second-order accurate in space and includes a limiter for effective stabilization. This stabilization term is derived from the material acoustic tensor of the solid model employed, carefully accounting for both volumetric (compression) and shear wave contributions. The proposed TLSPH including the C2nd (2nd-order Completeness) and cR (complete Riemann term including volumetric and shear wave contributions) schemes is verified by considering six classical benchmark tests. Graphical abstract
... The smoothed particle hydrodynamics (SPH) [1,2] is a purely particle-based method where all physical quantities are carried and updated by particles. Since its inception in 1977, over several decades, SPH has evolved into a robust numerical method capable of simulating fluid [3][4][5], solid [6][7][8][9][10], and fluid-structure interaction [11][12][13][14] problems effectively. The SPH method can be categorized into updated Lagrangian SPH (ULSPH) [7,15] and total Lagrangian SPH (TLSPH) [16], with ULSPH requiring the update of particle configurations (neighboring particles) at each step, while TLSPH does not necessitate this. ...
... The velocity gradient in Eq. (8) can be discretized as [42] ...
Article
Full-text available
Hourglass modes, characterized by zigzag particle and stress distributions, are a common numerical instability encountered when simulating solid materials with updated Lagrangian smoothed particle hydrodynamics (ULSPH). While recent solutions have effectively addressed this issue in elastic materials using an essentially non-hourglass formulation, extending these solutions to plastic materials with more complex constitutive equations has proven challenging due to the need to express shear forces in the form of a velocity Laplacian. To address this, a generalized non-hourglass formulation is proposed within the ULSPH framework, suitable for both elastic and plastic materials. Specifically, a penalty force is introduced into the momentum equation to resolve the disparity between the linearly predicted and actual velocity differences of neighboring particle pairs, thereby mitigating the hourglass issue. The stability, convergence, and accuracy of the proposed method are validated through a series of classical elastic and plastic cases, with a dual-criterion time-stepping scheme to improve computational efficiency. The results show that the present method not only matches or even surpasses the performance of the recent essentially non-hourglass formulation in elastic cases but also performs well in plastic scenarios.
... In this section, we extend the bending column problem to encompass a twisting column, which represents a highly nonlinear scenario, following Refs. [35,36,[71][72][73][74]. As illustrated in Fig. 21, the twisting is initiated with a sinusoidal rotational velocity field given by ω = [0, 0 sin (π y 0 /2 L) , 0] with 0 = 105 rad/s. ...
... It is worth noting that 0 = 100 rad/s and 200 rad/s are already considered challenging, as indicated in Refs. [71,73]. As shown in Fig. 23, the proposed formulation demonstrates stability, contrasting with the visibly disordered particle results from SPH. Conducting a convergence study involving sequential refinement of spatial resolution from H /dp = 4 to H /dp = 8 and H /dp = 12, and an analysis of anisotropic behavior across varying anisotropic ratios (a f /a = 0.1, a f /a = 0.5, and a f /a = 1.0), ...
Article
Full-text available
The persistence of hourglass modes poses a significant numerical instability issue in total Lagrangian smoothed particle hydrodynamics (TLSPH) solid dynamics, especially when dealing with substantial deformations, regardless of material properties. However, existing hourglass control methods have shown effectiveness only within limited applications. Thus far, a comprehensive solution capable of addressing hourglass issues across a wide range of material models, including elasticity, plasticity, and anisotropy, remains elusive. In this study, we introduce a unified TLSPH formulation grounded in volumetric-deviatoric stress decomposition, aimed at fundamentally mitigating hourglass modes in general simulations. Different conceptually from previous approaches using stress points or extra viscous or hourglass-control stresses within the momentum equation, our formulation is based on the weighted average of a standard but hourglass-prone formulation and an essentially non-hourglass formulation for elastic materials, employing a single limiter to dynamically adjust the weighting between the two formulations. Crucially, the dimensionless characteristic of the formulation enables seamless handling of complex material models. To validate the effectiveness of our formulation, we conduct simulations across a range of benchmark cases involving elastic, plastic, and anisotropic materials. To illustrate its versatility, we apply the formulation to simulate a complex scenario involving viscous plastic Oobleck material, contacts, and very large deformation. Our work addresses a critical gap in TLSPH simulations by offering a unified approach to mitigate hourglass modes, enhancing the reliability and accuracy of simulations across diverse material models and complex scenarios.
... The smoothed particle hydrodynamics (SPH) [1,2] is a purely Lagrangian particle-based method where all physical quantities are carried and updated by particles. Since its inception in 1977, over several decades, SPH has evolved into a robust numerical method capable of simulating fluid [3,4,5], solid [6,7,8,9,10], and fluid-structure interaction [11,12,13,14] problems effectively. The SPH method can be categorized into updated Lagrangian SPH (ULSPH) [7,15] and total Lagrangian SPH (TLSPH) [16], with ULSPH requiring the update of particle configurations (neighboring particles) at each step, while TLSPH does not necessitate this. ...
... The velocity gradient in Eq. (8) can be discretized as [40] ∇v = ...
Preprint
Full-text available
Hourglass modes, characterized by zigzag particle and stress distributions, are a common numerical instability encountered when simulating solid materials with updated Lagrangian smoother particle hydrodynamics (ULSPH). While recent solutions have effectively addressed this issue in elastic materials using an essentially non-hourglass formulation, extending these solutions to plastic materials with more complex constitutive equations has proven challenging due to the need to express shear forces in the form of a velocity Laplacian. To address this, a generalized non-hourglass formulation is proposed within the ULSPH framework, suitable for both elastic and plastic materials. Specifically, a penalty force is introduced into the momentum equation to resolve the disparity between the linearly predicted and actual velocities of neighboring particle pairs, thereby mitigating the hourglass issue. The stability, convergence, and accuracy of the proposed method are validated through a series of classical elastic and plastic cases, with a dual-criterion time-stepping scheme to improve computational efficiency. The results show that the present method not only matches or even surpasses the performance of the recent essentially non-hourglass formulation in elastic cases but also performs well in plastic scenarios.
... The impact efficiency in the technological process of grinding is often characterized by its purpose, that is, the production of a larger number of crushed particles. At the same time, the harbinger of separation, the destruction of [8][9][10] pieces into particles by impact, is the production of cracks and cracks in the crushed material, including cracks and cracks of a larger size, which in turn will quickly lead to the separation of the crushed piece into particles. Hence, cracking and target formation is the cause of softening. ...
Article
Full-text available
The object of research: technology for grinding particles of bone meal. Investigated problem: intensive production of finely ground particles of bone meal. The main scientific results: as a result of consideration of the theory and mechanics of destruction, the theory of crack formation and the Griffith theory, impact-toothed hammers were developed and patented. It was found that the penetration of impact-toothed surfaces of hammers into pieces of crushed material forms cracks. The size of cracks on bone raw material – eggshell, obtained by impact-toothed hammer, exceeds the crack obtained by impact-flat hammer, which occurred mainly due to the penetration of hammer teeth. The area of practical use of the research results: feed preparation shops specializing in feed production. Innovative technological product: technology for grinding particles of bone meal by intensive grinding with a crusher using impact-toothed hammers. Scope of the innovative technological product: hammer mills.
... Khayyer et al. [7], Morikawa and Asai [8]) and solid mechanics (e.g. Bonet and Lok [9], Lee et al. [10,11]), to name a few. Unlike grid-based methods, SPH represents the materials with discrete Lagrangian particles, allowing for efficient and accurate modeling of large deformations. ...
Article
This study presents a novel approach for simulating geotechnical problems including the initiation and post-failure behavior of discontinuities. The developed method is constituted by a mixed total Lagrangian--updated Lagrangian Smoothed Particle Hydrodynamics (SPH) method, which the main characteristic is to distinguish between internal forces within a body and contact forces from interactions with other bodies as internal stress and collision stress, respectively. Internal stress effects are calculated using total Lagrangian SPH interpolations, while collision stress effects are computed with updated Lagrangian. Fractures are simulated by employing plastic deformation as a damage measure, with fully damaged particles detached from their original body and treated as a separate particulate material, interacting via collision stress. Numerical tests confirm the method's capability in accurately modeling elastic collisions, friction forces and tension and shear fractures, validated against experimental data. Finally, we show the applicability of the proposed by simulating a real-scale landslide scenario, the Selborne experiment. This final numerical test demonstrates the capability of the method to simulate a landslide detached from the main soil mass as observed in the experiment.
... Extension to consider path-dependent constitutive models will be explored in Section 4. Interestingly, through the imposition of suitable kinematic conditions [26], the ALE system (15) can degenerate into three alternative systems of first-order conservation equations. As shown in Table 1, these formulations include the well-established Total Lagrangian formulation [27,29,30,37,41,[48][49][50][51][52][53][54][55][56][57][58][59][60], the Eulerian formulation and the recently proposed Updated Reference Lagrangian formulation [32,37], which incorporates the concept of incremental kinematics. Thus, the ALE system (15) emerges as an elegant generalisation of various existing continuum conservation laws descriptions. ...
Article
Key features for the versatile industrial applications of smoothed particle hydrodynamics (SPH) are numerical accuracy and stability, computational effciency, and ease of implementation. To meet these requirements, this study presents a straightforward algorithm to treat large-strain elastoplasticity within Total Lagrangian SPH (TLSPH) framework. In terms of accuracy and stability, the total Lagrangian formulation in SPH completely eliminates tensile instability related to the use of Eulerian kernel functions. This study adopts a multiplicative hyperelastic-based plasticity model, enabling the model to treat from purely elastic to elastoplastic structural deformations. A common strategy of effcient simulations in structural analysis is utilizing predefned local resolution refnement with non-uniform particle spacing. However, within TLSPH framework, numerical accuracy and effciency of this strategy were not investigated suffciently. To maintain a proper number of neighboring particles in the presence of non-uniform spacing, an anisotropic kernel and its derivatives are incorporated in SPH approximations. Beyond its inherent stability, TLSPH can offer great convenience for the multi-resolution implementation with minimized computational load, as repeated kernel computations at each time advancement are not required. Several benchmark tests are conducted to validate the proposed TLSPH model with various initial particle distributions, demonstrating good accuracy and robustness while lowering computational load.
Article
Full-text available
The paper presents a new computational framework for the numerical simulation of fast large strain solid dynamics, with particular emphasis on the treatment of near incompressibility. A complete set of first order hyperbolic conservation equations expressed in terms of the linear momentum and the minors of the deformation (namely the deformation gradient, its co-factor and its Jacobian), in conjunction with a polyconvex nearly incompressible constitutive law, is presented. Taking advantage of this elegant formalism , alternative implementations in terms of entropy-conjugate variables are also possible, through suitable symmetrisation of the original system of conservation variables. From the spatial discretisation standpoint, modern Computational Fluid Dynamics code "OpenFOAM" [http://www.openfoam.com/] is here adapted to the field of solid mechanics, with the aim to bridge the gap between computational fluid and solid dynamics. A cell centred finite volume algorithm is employed and suitably adapted. Naturally, discontinuity of the conservation variables across control volume interfaces leads to a Riemann problem, whose resolution requires special attention when attempting to model materials with predominant nearly incompressible behaviour (κ/µ ≥ 500). For this reason, an acoustic Riemann solver combined with a preconditioning procedure is introduced. In addition, a global a posteriori angular momentum projection procedure proposed in [1] is also presented and adapted to a Total Lagrangian version of the nodal scheme of Kluth and Després [2] used in this paper for comparison purposes. Finally, a series of challenging numerical examples is examined in order to assess the robustness and applicability of the proposed methodology with an eye on large scale simulation in future works.
Article
Full-text available
This paper builds on recent work developed by the authors for the numerical analysis of large strain solid dynamics, by introducing an upwind cell centred hexahedral Finite Volume framework implemented within the open source code OpenFOAM [http://www.openfoam.com/http://www.openfoam.com/]. In Lee, Gil and Bonet [1], a first order hyperbolic system of conservation laws was introduced in terms of the linear momentum and the deformation gradient tensor of the system, leading to excellent behaviour in two dimensional bending dominated nearly incompressible scenarios. The main aim of this paper is the extension of this algorithm into three dimensions, its tailor-made implementation into OpenFOAM and the enhancement of the formulation with three key novelties. First, the introduction of two different strategies in order to ensure the satisfaction of the underlying involutions of the system, that is, that the deformation gradient tensor must be curl-free throughout the deformation process. Second, the use of a discrete angular momentum projection algorithm and a monolithic Total Variation Diminishing Runge-Kutta time integrator combined in order to guarantee the conservation of angular momentum. Third, and for comparison purposes, an adapted Total Lagrangian version of the Hyperelastic-GLACE nodal scheme of Kluth and Despres [2] is presented. A series of challenging numerical examples are examined in order to assess the robustness and accuracy of the proposed algorithm, benchmarking it against an ample spectrum of alternative numerical strategies developed by the authors in recent publications.
Article
Full-text available
In Part I of this series, Bonet et al. (2015) introduced a new computational framework for the analysis of large strain isothermal fast solid dynamics, where a mixed set of Total Lagrangian conservation laws was presented in terms of the linear momentum and an extended set of strain measures, namely the deformation gradient, its co-factor and its Jacobian. The main aim of this paper is to expand this formulation to the case of nearly incompressible and truly incompressible materials. The paper is further enhanced with three key novelties. First, the use of polyconvex nearly incompressible strain energy functionals enables the definition of generalised convex entropy functions and associated entropy fluxes. Two variants of the same formulation can then be obtained, namely, conservation-based and entropy-based, depending on the unknowns of the system. Crucially, the study of the eigenvalue structure of the system is carried out in order to demonstrate its hyperbolicity and, thus, obtain the correct time step bounds for explicit time integrators. Second, the development of a stabilised Petrov-Galerkin framework is presented for both systems of hyperbolic equations, that is, when expressed in terms of either conservation or entropy variables. Third, an adapted fractional step method, built upon the work presented in Gil et al. (2014), is presented to extend the range of applications towards the incompressibility limit. Finally, a series of numerical examples are presented in order to assess the applicability and robustness of the proposed formulation. The overall scheme shows excellent behaviour in compressible, nearly incompressible and truly incompressible scenarios, yielding equal order of convergence for velocities and stresses.
Article
A unified stability analysis of meshless methods with Eulerian and Lagrangian kernels is presented. Three types of instabilities were identified in one dimension: an instability due to rank deficiency, a tensile instability and a material instability which is also found in continua. The stability of particle methods with Eulerian and Lagrangian kernels is markedly different: Lagrangian kernels do not exhibit the tensile instability. In both kernels, the instability due to rank deficiency can be suppressed by stress points. In two dimensions the stabilizing effect of stress points is dependent on their locations. It was found that the best approach to stable particle discretizations is to use Lagrangian kernels with stress points. The stability of the least-squares stabilization was also studied. Copyright © 2000 John Wiley & Sons, Ltd.
Article
This paper presents a new Smooth Particle Hydrodynamics (SPH) computational framework for explicit fast solid dynamics. The proposed methodology explores the use of the Streamline Upwind Petrov Galerkin (SUPG) stabilisation methodology as an alternative to the Jameson-Schmidt-Turkel (JST) stabilisation recently presented by the authors in [1] in the context of a conservation law formulation of fast solid dynamics. The work introduced in this paper puts forward three advantageous features over the recent JST-SPH framework. First, the variationally consistent nature of the SUPG stabilisation allows for the introduction of a locally preserving angular momentum procedure which can be solved in a monolithic manner in conjunction with the rest of the system equations. This differs from the JST-SPH framework, where an a posteriori projection procedure was required to ensure global angular momentum preservation. Second, evaluation of expensive harmonic and bi-harmonic operators , necessary for the JST stabilisation, is circumvented in the new SUPG-SPH framework. Third, the SUPG-SPH framework is more accurate (for the same number of degrees of freedom) than its JST-SPH counterpart and its accuracy is comparable to that of the robust (but computationally more demanding) Petrov Galerkin Finite Element Method (PG-FEM) technique explored by the authors in [2–5], as shown in the numerical examples included. A series of numerical examples are analysed in order to benchmark and assess the robustness and effectiveness of the proposed algorithm. The resulting SUPG-SPH framework is therefore accurate, robust and computationally efficient, three key desired features that will allow the authors in forthcoming publications to explore its applicability in large scale simulations.
Article
This paper presents a new Smooth Particle Hydrodynamics (SPH) computational framework for large strain explicit solid dynamics. A mixed-based set of Total Lagrangian conservation laws is presented in terms of the linear momentum and an extended set of geometric strain measures, comprised of the deformation gradient, its co-factor and the Jacobian. Taking advantage of this representation, the main aim of this paper is the adaptation of the very efficient Jameson-Schmidt-Turkel (JST) algorithm, extensively used in computational fluid dynamics, to a SPH based discretisation of the mixed-based set of conservation laws, with three key distinct novelties. First, a conservative JST-based SPH computational framework is presented with emphasis in nearly incompressible materials. Second, the suppression of numerical instabilities associated with the non-physical zero-energy modes is addressed through a well-established stabilisation procedure. Third, the use of a discrete angular momentum projection algorithm is presented in conjunction with a monolithic Total Variation Diminishing Runge-Kutta time integrator in order to guarantee the global conservation of angular momentum. For completeness, exact enforcement of essential boundary conditions is incorporated through the use of a Lagrange multiplier projection technique. A series of challenging numerical examples (e.g. in the near incompressibility regime) are examined in order to assess the robustness and accuracy of the proposed algorithm. The obtained results are benchmarked against a wide spectrum of alternative numerical strategies.
Article
Shape-memory polymers (SMPs) represent a class of smart materials able to store a temporary shape and to recover the original shape upon an external stimulus, such as temperature. The present paper proposes a three-dimensional finite-strain phenomenological model for thermo-responsive SMPs, which distinguishes between two material phases presenting different properties and is based on a rule of mixtures. The proposed model is motivated by the earlier work of Reese et al. (2010) and it considers several significant material features that had not been addressed in previous phenomenological approaches. Specifically, the model reproduces both heating-stretching-cooling and cold drawing shape-fixing procedures and it takes into account the non-ideal behavior of realistic SMPs (i.e. imperfect shape-fixing and incomplete shape-recovery). Several numerical tests are reported to assess model performances, from simple uniaxial and biaxial tests to complex simulations of biomedical devices. Comparisons with experimental data taken from the literature are also provided to validate the model. The proposed improvements increase the model applicability over a wide range of polymer types and operating conditions.
Article
This paper describes in detail the formulation of large strain solid mechanics based on the tensor cross product, originally presented by R. de Boer, Vektor- und Tensorrechnung für Ingenieure, Springer-Verlag, 1982., page 76, and recently re-introduced by Bonet et al. in J. Bonet, A. J. Gil, R. Ortigosa, A computational framework for polyconvex large strain elasticity, Computer Methods in Applied Mechanics and Engineering 283 (2015) 1061 – 1094., and J. Bonet, A. J. Gil, C. H. Lee, M. Aguirre, R. Ortigosa, A first order hyperbolic framework for large strain computational solid dynamics. Part I: Total Lagrangian isothermal elasticity, Computer Methods in Applied Mechanics and Engineering 283 (2015) 689 – 732. The paper shows how the tensor cross product facilitates the algebra associated with the area and volume maps between reference and final configurations. These maps, together with the fibre map, make up the fundamental kinematic variables in polyconvex elasticity. The algebra proposed leads to novel expressions for the tangent elastic operator which neatly separates material from geometrical dependencies. The paper derives new formulas for the spatial and material stress and their corresponding elasticity tensors. These are applied to the simple case of a Mooney-Rivlin material model. The extension to transversely isotropic material models is also considered.
Article
Accurate stress computation is an essential step of the efficient Galerkin meshfree formulation with the stabilized conforming nodal integration (SCNI). In this work particular emphasis is placed on the stress computations of SCNI-based Galerkin meshfree methods. The requirement for variational consistency of the SCNI formulation is discussed. It is shown that the discrete SCNI formulation is consistent in the variational sense if a constant stress field based on the smoothing strain is employed in the nodal representative or smoothing domain. Subsequently three stress computation approaches, i.e., direct nodal stress evaluation (DNS), consistent nodal stress evaluation (CNS), and consistent centroid stress evaluation (CCS), are presented. It turns out that both CNS and CCS satisfy the condition of variational consistence whereas DNS does not. A comprehensive numerical comparison of nodal strain energy error reveals that the variational consistence does not necessarily lead to more accurate results, while the proposed CCS approach is uniformly confirmed to yield the most favorable results.