Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

INSTITUTE OF PHYSICS PUBLISHING SUPERCONDUCTOR SCIENCE AND TECHNOLOGY

Supercond. Sci. Technol. 19 (2006) 1246–1252 doi:10.1088/0953-2048/19/12/004

Numerical solution of critical state in


superconductivity by finite element
software
Z Hong, A M Campbell and T A Coombs
Cambridge University Engineering Department, Trumpington Street, Cambridge CB2 1PZ,
UK

E-mail: zh223@cam.ac.uk

Received 24 August 2006, in final form 5 October 2006


Published 23 October 2006
Online at stacks.iop.org/SUST/19/1246
Abstract
A numerical method is proposed to analyse the electromagnetic behaviour of
systems including high-temperature superconductors (HTSCs) in
time-varying external fields and superconducting cables carrying AC
transport current. The E – J constitutive law together with an H-formulation
is used to calculate the current distribution and electromagnetic fields in
HTSCs, and the magnetization of HTSCs; then the forces in the interaction
between the electromagnet and the superconductor and the AC loss of the
superconducting cable can be obtained. This numerical method is based on
solving the partial differential equations time dependently and is adapted to
the commercial finite element software Comsol Multiphysics 3.2. The
advantage of this method is to make the modelling of the superconductivity
simple, flexible and extendable.
(Some figures in this article are in colour only in the electronic version)

1. Introduction processes which can be described with partial differential


equations (PDEs). Until recently, commercial software has had
High-temperature superconductors possess a number of unique difficulty with very nonlinear current voltage characteristics, so
properties that make them attractive for use in a range the magnetization measurement could not be modelled well.
of engineering applications. To optimize the design of
Thanks to the equation-based modelling capabilities of more
a superconducting system, it is necessary to predict the
recent commercial FEM packages, a solver of the critical state
electromagnetic behaviour of the superconductor in order to
find out the field distributions, critical current distributions, and based on a set of Maxwell equations can be easily set up
electromagnetic forces. and different magnetic field conditions can be achieved by
In the past, several numerical methods have been proposed defining different boundary conditions and initial conditions
to solve the critical state in superconductivity. Analytical of the PDEs. The sophisticated commercial FEM software
methods [1, 2] have been developed for simple geometries can solve the complex problems quickly and accurately. The
and uniform field conditions. For more complicated shapes ability of interactive post-processing and visualization make
and field conditions, numerical methods should be developed. the analysing of the results easier. The FEM software being
These numerical models usually use the finite element method used in this paper is Comsol multiphysics 3.2. The aim of
(FEM) or the finite difference method to solve Maxwell
this paper is to propose a general method to solve the critical
equations in two dimensions or three dimensions. These
methods can be classified by the main equations they are using: state in superconductivity in two dimensions by utilizing the
A–V formulations [3, 4], T – formulations [5, 6], and the FEM modelling package. It is a flexible, extendable numerical
unstrained H-formulations [7]. method without an algorithm and it can be simply implemented
Many commercial FEM modelling packages have been into most of the commercial FEM software to get an easy
developed and can be used for the simulation of many physical solution to many problems in complex geometries.

0953-2048/06/121246+07$30.00 © 2006 IOP Publishing Ltd Printed in the UK 1246


Numerical solution of critical state in superconductivity by finite element software

where Jc is the critical current density (defined with the


standard 1 μV cm−1 electric field criterion). E 0 = 1 ×
10−4 V m−1 and n = 21 are typical values for a type-II
superconductor. The critical current density Jc can either be
constant or field dependent [8]. If Jc is field dependent, then

Jc0 · B0
Jc (B) = , B = μ0 μr Hx2 + Hy2 ,
B0 + B
where Jc0 and B0 are constants determined by the property of
the material.
Ampere’s law, ∇ × H = J, can be used to describe the
Hx 
relationship between Jsc z and H y
; Jsc z has a component
only in the z -direction, so Ampere’s law in this case is
expressed as
∂ Hy ∂ Hx
Jsc z = − . (4)
∂x ∂y
Figure 1. Subdomains and boundaries; the magnetic field is in the
x – y -plane and the critical current density Jsc z and the induced Substituting (4) and (3) into (2) gives
electric field E sc z are in the z -direction. ⎡   ∂ H y ∂ Hx n ⎤

∂ E 0 ∂xJc (B)∂ y
⎢ ⎥
⎢ ⎥  ∂ Hx 
⎢ ∂y ⎥
2. Numerical scheme ⎢ ⎥ ∂t
  ∂ H y ∂ H n ⎥ = −μ0 μr ∂ H y . (5)

⎢ ∂x − ∂ y
x
⎥ ∂t
⎣ ∂ E0 Jc (B) ⎦
The numerical scheme is to divide the space into several
subdomains, as shown in figure 1; Dsc , Dair and Dmag represent −
∂x
the superconducting region, dielectric region (normally air),
Equation (5) includes two PDEs and two dependent variables
and the region with magnetization respectively. A set of
PDEs is defined in each subdomain. All of these PDEs share
Hx and Hy ; combining its with suitable boundary conditions it
the same dependent variables. One subdomain interacts with can be solved with FEM software.
a conjoint subdomain according to the boundary conditions The equations for the non-superconducting subdomains
defined between them. By defining a suitable boundary are the same as the equations for the superconducting
condition on the interior boundary and the outer boundary, subdomain except that the E – J relationship of the non-
different kinds of magnetic field conditions can be set up, superconducting region is a linear Ohm’s law, E = ρ J , where
such as a superconductor in a uniform or non-uniform field, ρ is the resistivity of air in this paper.
a superconductor–magnet system and a superconducting cable
carrying transport current. The field-cooling and zero-field- 3. Boundary conditions and results
cooling problems can be distinguished by defining different
initial values for the PDEs. To model the critical state of a type-II superconductor
Let us consider a space with infinite length in the z - in the external magnetic field, some boundary conditions
direction which is normal to the plane of the paper (x – y - representing different magnetic field conditions need to be
plane), shown in figure 1. The magnetic flux is flowing applied. Dirichlet and Neumann boundary conditions are used
in the x – y -plane and the magnetic field is used as the to describe most of the electromagnetic situations and they can
dependent variable. In a two-dimensional problem there be easily set up by most commercial FEM software.
are two dependent variables, Hx and Hy , representing the
components of the magnetic field in the x - and y -directions. 3.1. Superconductor in uniform field
The critical current density Jsc z and the induced electric field
E sc z are in the z -direction. According to Faraday’s law, Consider the situation of a superconducting rod with infinite
length in the z -direction and with a uniform magnetic field
μ0 μr ∂ H applied in the y -direction. This situation can be modelled
∇×E=− . (1)
∂t by setting up a Dirichlet boundary condition on the outer
H  boundary of the dielectric subdomain (boundary 1 in figure 1)
Substitutinge H = x
Hy into (1), as
 ∂ E sc z   ∂ Hx  Hx = 0
∂y ∂t (6)
= −μ0 μr ∂ Hy
. (2) Hy = f (t)
− ∂ E∂xsc z ∂t
where f (t) is a function that describes how the external field
The electrical behaviour of the superconducting material in the y -direction varies with time. The boundary condition
is modelled by the E – J power law, between the superconducting subdomain and the dielectric
 subdomain (boundary 2 in figure 1) is continuity,
Jsc z n
E sc z = E 0 , (3)
Jc(B) n × (H1 − H2 ) = 0. (7)

1247
Z Hong et al

1.5e+5
(a) π
1.0e+5
• 1.1π • 0.9π

Magnetization (A/m)
5.0e+4
•0.6π
0.0 3π/2 π/2
• 1.6π
• 0.03π
–5.0e+4
1.9π • • 0.15π(Full penetration)
–1.0e+5
• 2.0π

–1.5e+5
–1.5e+6 –1.0e+6 –5.0e+5 0.0 5.0e+6 1.0e+6 1.5e+6
Applied magnetic field (A/m)

1.5e+5
(b)
f=10Hz
1.0e+5 f=5Hz
f=2.5Hz

Magnetization (A/m)
5.0e+4

Figure 2. The plot of the current density distribution and the 0.0
magnetic flux line. H0 = 1 × 106 A m−1 is the maximum value of
the applied field. The figures correspond to the different time steps.
–5.0e+4
( E – J power law n = 21, field-dependent Jc , Jc0 = 2 × 107 A m−1 ).
–1.0e+5

Figure 2 shows the simulation results of the superconduct- –1.5e+5


–1.5e+6 –1.0e+6 –5.0e+5 0.0 5.0e+5 1.0e+6 1.5e+6
ing rod with a rectangular cross section (0.12 m × 0.04 m) in Applied magnetic field (A/m)
a y -direction uniform external field. On the outer boundary,
Figure 3. (a) Magnetization loop for a full cycle of the applied field
equation (6) is defined as
in 5 Hz, obtained numerically from the model. The points on the
curve represent the time steps shown in figure 2. Full penetration
Hx = 0 occurs at ωt = 0.15π . ( E – J power law n = 21, field-dependent Jc ,
Jc0 = 2 × 107 A m−1 ). (b) Magnetization loops for a full cycle of the
Hy = H0 sin(ωt). applied field. The three curves illustrate the results obtained when the
frequency of the applied field is changed. ( E – J power law n = 21,
field-dependent Jc .)
Hy is varying as a sinusoidal wave for a full cycle in the
frequency of 5 Hz. The surface plot in figure 2 is the current
distribution and the streamline represents the total magnetic
field. The results are shown for eight moments from ωt = magnetization increases. As the applied field increases further,
0.03π to 2π . It can be clearly seen from the plots how the the magnitude of the current decreases because of the field
superconducting rod tries to oppose the penetration of the flux dependence of Jc and the absolute value of the magnetization
lines as the external field increases. When the external field tends to decrease although the current fills a larger volume of
decreases the superconducting rod also resists the motion of the sample. A similar shape of the magnetization loop of an
the flux lines and at the end of the cycle (ωt = 2π) the HTSC with field-dependent Jc was proposed by Sanchez [9].
superconducting rod has trapped some of the field and remains As the E – J power law is used in this model, the shape
magnetized. of the magnetization loop can be affected by the frequency
The magnetization of the sample in the y -direction can be of the applied field due to the flux creep effect [14]. The
calculated by magnetization loops for different frequencies are shown in
m0 figure 3(b) and only tiny differences can be found between
M= (8)
them. This result illustrates that, for this simulation, the flux
V
where m 0 is the total magnetic moment of the superconducting creep effect is insignificant around 5 Hz if the E – J power law
subdomain; it is expressed as is used with n = 21.

m 0 = 12 r × J dV. (9) 3.2. Superconductor in non-uniform field
Dsc
Sometimes it is required to model the superconductor in a
J only has the component in the z -direction, so r × non-uniform magnetic field such as the interaction between
J = x · Jz sc . The magnetization loop of a full cycle a superconductor and a magnet [10]. Consider a system
is plotted in figure 3(a). When the applied field starts to consisting of a type-II superconductor and an electromagnet.
increase from zero, the sample is in the diamagnetic state. The system is infinitely long in the z -direction and the
Then some currents penetrate into the superconductor and the symmetry allows the assumption that there are no field changes

1248
Numerical solution of critical state in superconductivity by finite element software

of the induced current flowing inward in the x – y -plane is the


Electromagnet same as that flowing outward.
Figure 5 shows the current distribution and the magnetic
flux line for the model in position 1. The applied field
increases from zero to the maximum and then decreases to
Y
zero. The superconductor has a circular cross section with a
HTSC radius of 2 cm and it is symmetric about the y -axis with the
electromagnet.
X Figure 6 shows the current distribution and the magnetic
flux line for the situation where the HTSC is placed in position
3. In this case the central axis of the superconductor is not
aligned with that of the electromagnet and the current density
Figure 4. The cross-section in the X –Y -plane of the HTSC and the
electromagnet in different positions. is not distributed symmetrically in the superconducting region;
hence the lateral force between the superconductor and the
electromagnet is not zero. The boundary condition Itransport = 0
along the z -axis. A series of arrangements according to the constrains the total induced current inside the superconducting
region to be zero.
different positions of the HTSC is modelled separately, as
The Lorentz force method was chosen to calculate
shown in figure 4. The central axis of the HTSC in the y -
the magnetic force between the superconductor and the
direction is aligned with that of the electromagnet in position 1
electromagnet:
but not aligned in the other positions.

In this model the system is divided into three subdomains:
a superconducting region, a dielectric region and a region of Fx = Jsc z × μ0 Hy ds
Dsc
magnetization. The region of magnetization is modelled as a  (15)
subdomain in which the magnetic field is uniform. To model Fy = Jsc z × μ0 Hx ds.
an electromagnet generating a magnetic field in the y -direction Dsc
the equations of this subdomain are
The vertical forces ( y -direction) between the supercon-
Hx = 0 ductor and the electromagnet are plotted in figure 7. The sim-
∀t = 0 ∼ t0 (10) ulated result of the force shows a hysteretic property which
Hy = H0 t
has been proposed both numerically and experimentally in the
for the increasing field and literature [11, 12]. As expected, the maximum vertical force
decreases as the HTSC is placed further away from the elec-
Hx = 0 tromagnet. The penetration depth also decreases, which results
∀t = t0 ∼ 2t0 (11)
Hy = H0 (2t0 − t) in a decrease in the hysteresis of the loop. That means that
the curves become more reversible as the arrangement changes
for the decreasing field. The magnitude of the field in the y - from position 1 to position 4.
direction varies linearly. The lateral forces (x -direction) between the superconduc-
The boundary condition of the superconducting subdo- tor and the electromagnet for positions 2, 3 and 4 are plotted
main and the subdomain of magnetization is continuity. Two in figure 8. The lateral force for position 1 is zero. In this
Dirichlet boundary conditions need to be applied to the outer case, as the HTSC is placed further away from the electromag-
boundary of the dielectric subdomain (boundary 1 in figure 1). net the lateral force increases initially and then decreases; as
The first one is magnetic insulation: can be seen from the figure, the lateral force in position 3 is
 larger than that in position 2. This characteristic has also been
Hx proposed in [12].
n· = 0. (12)
Hy

The second is that the transport current in the z -direction is 3.3. Superconducting cable carrying transport current
zero,  In this section the behaviour of a superconducting cable
J = Itransport = 0, ∀i = 1, 2, 3. (13) carrying AC transport current is modelled in order to calculate
Dι the AC loss of the cable in self-field. The cable consists of
According to Ampere’s law, five superconducting wires connected in parallel. Each wire
 has a cross-sectional area of 1 × 10−5 m2 (10 mm × 1 mm).
The model is made up of five superconducting subdomains and
Itransport = Ht dl = 0 (14)
Boundary 1 one dielectric subdomain. Two boundary conditions need to
be applied to the outer boundary of the dielectric subdomain.
where Ht is the tangential component of the magnetic field The first one is magnetic insulation as shown in (12); the other
on the boundary of the dielectric subdomain. Note that is that the line integration of the tangential component of the
the boundary condition in (13) can be neglected if the magnetic field on the boundary equals the forced transport
electromagnet and the superconductor are symmetric about the current: 
y -axis (position 1 in figure 4). For all other arrangements this Itransport = Ht dl. (16)
boundary condition has to be held to make sure that the amount Boundary 1

1249
Z Hong et al

Figure 5. Plot of the current density distribution and the magnetic flux line. H0 = 1 × 106 A m−1 is the maximum value of the applied field.
The figures correspond to the different values of the applied field. The fourth figure is plotted at the moment the applied field is half of H0
when it decreases. ( E – J power law n = 21, constant Jc = 2 × 107 A m−1 , position 1.)

80

Loop of the vertical force (position 1)


60 Loop of the vertical force (position 2)
Loop of the vertical force (position 3)
Loop of the vertical force (position 4)
Force (N)

40

20

0.0 2.0e+5 4.0e+5 6.0e+5 8.0e+5 1.0e+6 1.2e+6


Applied magnetic field (A/m)

Figure 7. The vertical force between the electromagnet and the


HTSC against the magnetic field generated by the electromagnet.
The four curves in the plot correspond to the positions 1, 2, 3 and 4.

Figure 6. The plot of the current density distribution and the


magnetic flux line. H0 = 1 × 106 A m−1 is the maximum value of
simulated. The current distribution and the magnetic flux line
the applied field. The figures correspond to the different values of the
applied field. The fourth figure is plotted at the moment the applied for one cycle are shown in figure 9. As can be seen from the
field is half of H0 when it decreases. ( E – J power law n = 21, figures, the cross-section of the superconducting wire carrying
constant JC = 2 × 107 A m−1 , position 3.) AC transport current is divided into three regions: a region with
positive current, a region with reverse current, and a current-
free region. This distribution matches the analytical result
A 50 Hz sinusoidal current with amplitude of 800 A is fed proposed by Norris [13] in 1969 and it is also proved by the
into the cable. Two cycles with 1000 time steps per cycle are result obtained numerically [11].

1250
Numerical solution of critical state in superconductivity by finite element software

Figure 10. The plot of values of the AC loss against the transport
Figure 8. The lateral force between the electromagnet and the HTSC current (normalized). Ic is the maximum DC lossless current that the
against the magnetic field generated by the electromagnet. The three cable can carry. Wc is the RMS value of the AC loss when the cable
curves in the plot correspond to the positions 2, 3 and 4. carries Ic .

Table 1. Computational time, mesh vertex and time steps.


Total Computational
Mesh Time computational time per step
vertex steps time (s) (s)
HTSC in 6763 2500 622 0.25
uniform field
HTSC in non- 11210 400 422 1.05
uniform field
Cable 14810 2000 3428 1.71

Table 2. The speed and the stability of the solver.

Figure 9. The plot of the current density distribution and the Speed Stability
magnetic flux line of the superconducting cable carrying AC current Refine the mesh Much slower Better
I = 800 sin(ωt)A. The cross-sectional area of each wire is Refine the time steps Slower Much better
1 × 10−5 m2 . The figures correspond to the different time steps in Use a smaller n Faster Better
one cycle. ( E – J power law n = 21, constant Jc = 2 × 107 A m−1 ). of E – J power law

The AC loss is evaluated by the equation computational time for the cases which have been investigated
 in the previous sections.
Q acloss = Jsc z · E sc z ds. (17) Another important characteristic of this numerical method
Dsc
is the ability of the solver to converge to a unique solution.
In figure 10, the RMS value of the AC loss of this Sometimes it is difficult to converge; the level of difficulty
superconducting cable is plotted against different values of the depends on the characteristics of the PDEs. If the equations or
transport current. The transport current is normalized to Ic , the boundary conditions are highly nonlinear, certain methods
which is the maximum DC lossless current that the cable can can be used to improve the convergence and help reach an
carry. The value of the AC loss is normalized to Wc , which accurate solution. Table 2 shows how the specifications of the
is the RMS value of the AC loss when the cable carries the solver affect the speed and the stability.
current Ic . The result is compared with the analytical model Under a particular specification, the speed and the stability
proposed by Norris [13] and shows a good match. of the solver can also be affected by the condition of the
external field and the forced current. Generally speaking, a
4. Conclusion large magnitude in the applied field and transport current or a
high frequency in the AC field and current will slow down the
The speed of the numerical method proposed in this paper solver and cause instability. It is not suggested that a smaller
depends on many factors such as the mesh quality, the number value of the power n of the E – J power law should be used
of time steps of the simulation and the number n used in the to improve the stability, because this will cause a significant
E – J power law. The simulations in this paper were run on an flux creep effect [14]. Both refining the mesh and refining
Intel Centrino (1.73 GHz) PC. Table 1 gives a summary of the the time step can result in a smoother convergence but will

1251
Z Hong et al

Table 3. The computational time required and the stabilities of the [3] Barnes G, McCulloch M and Dew-Hughes D 1999 Computer
model in figure 9 under different numbers of mesh vertices and time modelling of type II superconductors in applications
steps. Supercond. Sci. Technol. 12 518–22
[4] Coombs T A, Campbell A M, Murphy A and Emmens M 2001
Computational time A fast algorithm for calculating the critical state in
and stability 7328 mesh vertices 14810 mesh vertices superconductors COMPEL-The Int. J. Comput. Math. Electr.
1000 time steps 169 s 1221 s Electron. Eng. 20 240–52
Less stable Stable [5] Amemiya N, Murasawa S, Banno N and Miyamoto K 1998
2000 time steps 324 s 3428 s Numerical modelings of superconducting wires for AC loss
Stable Very stable calculations Phys. Rev. C 310 16–29
[6] Meunier G, Le Floch Y and Guérin C 2003 A non-linear circuit
coupled T –T0 – formulation for solid conductors IEEE
Trans. Magn. 39 1729–32
reduce the speed of the solver. Refining the time step is a [7] Pecher R, McCulloch M D, Chapman S J, Prigozhin L and
better approach because in this case the solver slows down Elliottk C M 2003 3D-modelling of bulk type-II
linearly as the number of time steps increases, as shown in superconductors using unconstrained H-formulation EUCAS
table 3. To refine the mesh will cause a dramatic increase in 2003: 6th European Conf. on Applied Superconductivity
[8] Kim Y B, Hempstead C F and Strnad A R 1962 Critical
the computational time. persistent currents in hard superconductors Phys. Rev. Lett.
So far, we have used commercial finite element software 9 306–9
to solve the critical state in superconductivity. The method [9] Sanchez A and Navau C 2001 Magnetic properties of finite
allows the simulation of the superconductor under uniform superconducting cylinders. I. Uniform applied field, II.
or non-uniform external fields and forced transport currents. Nonuniform applied field and levitation force Phys. Rev. B
64 214506
The results of this numerical method have been compared [10] Qin M J, Li G, Liu H K, Dou S X and Brandt E H 2002
with other critical state models and they show a good match. Calculation of the hysteretic force between a superconductor
The performance of the solver has been discussed and the and a magnet Phys. Rev. B 66 024516
relationship between the computational time and the stability [11] Ruiz-Alonso D, Coombs T A and Campbell A M 2004
of the solver has been investigated. The numerical method Numerical analysis of high-temperature superconductors
with the critical-state model IEEE Trans. Appl. Supercond.
proposed in this paper is compatible to a two-dimensional axial 14 2053–63
symmetry problem and has the potential to be extended to a [12] Ueda H and Ishiyama A 2004 Dynamic characteristics and
three-dimensional model of the critical state. finite element analysis of a magnetic levitation system using
a YBCO bulk superconductor Supercond. Sci. Technol.
17 S170–5
References [13] Norris W T 1970 Calculation of hysteresis losses in hard
[1] Prigozhin L 1997 Analysis of critical-state problems in type-II superconductors carrying ac: isolated conductors and edges
superconductivity IEEE Trans. Appl. Supercond. 7 3866–73 of thin sheets J. Phys. D: Appl. Phys. 3 489–507
[2] Brandt E H 1996 Superconductors of finite thickness in a [14] Feigelman M V, Geshkenbein V B, Larkin A I and
perpendicular magnetic field: strips and slabs Phys. Rev. B Vinokur V M 1989 Theory of collective flux creep Phys.
54 4246–64 Rev. Lett. 63 2303

1252

You might also like