Superconductor

Science and
Technology

PAPER

User defined elements in ANSYS for 2D multiphysics modeling of

superconducting magnets
To cite this article: Lucas Brouwer et al 2019 Supercond. Sci. Technol. 32 095011

Manuscript version: Accepted Manuscript

Accepted Manuscript is “the version of the article accepted for publication including all changes made as a result of the peer review process,
and which may also include the addition to the article by IOP Publishing of a header, an article ID, a cover sheet and/or an ‘Accepted
Manuscript’ watermark, but excluding any other editing, typesetting or other changes made by IOP Publishing and/or its licensors”

This Accepted Manuscript is© .

During the embargo period (the 12 month period from the publication of the Version of Record of this article), the Accepted Manuscript is fully
protected by copyright and cannot be reused or reposted elsewhere.

As the Version of Record of this article is going to be / has been published on a subscription basis, this Accepted Manuscript will be available for
reuse under a CC BY-NC-ND 4.0 licence after the 12 month embargo period.

After the embargo period, everyone is permitted to use copy and redistribute this article for non-commercial purposes only, provided that they
adhere to all the terms of the licence [Link]

Although reasonable endeavours have been taken to obtain all necessary permissions from third parties to include their copyrighted content
within this article, their full citation and copyright line may not be present in this Accepted Manuscript version. Before using any content from this
article, please refer to the Version of Record on IOPscience once published for full citation and copyright details, as permissions may be required.
All third party content is fully copyright protected, unless specifically stated otherwise in the figure caption in the Version of Record.

View the article online for updates and enhancements.

This content was downloaded from IP address [Link] on 30/07/2025 at 07:15

Page 1 of 24 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1

1
2
3
4
5
6
7
8
9
10
User defined elements in ANSYS for 2D
11 multiphysics modeling of superconducting magnets
12
13
14 Lucas Brouwer 1 , Diego Arbelaez 1 , Bernhard Auchmann 2,3
,
15
16
Lorenzo Bortot 3 , and Edvard Stubberud 3
1
17 Lawrence Berkeley National Laboratory, Berkeley, CA, 94720 USA
18 2
Paul Scherrer Institute, CH-5232 Villigen, Switzerland
19 3
CERN, CH-1211 Geneva 23, Switzerland
20
E-mail: lnbrouwer@[Link]
21
22
23 Abstract. Dynamic simulation of superconducting magnets is critical for the design
24 of quench protection systems to prevent potentially damaging temperatures and high
25 voltage from developing after magnet quench. Modeling these scenarios is challenging
26 due to the many multiscale phenomena which impact magnet behavior. These range
27 from conductor scale effects of quench and interfilament coupling currents up to
28 the behavior of the magnet in its powering and protection circuit. In addition, a
29
strong coupling between electromagnetic and thermal domains is required to capture
30
temperature and field dependent material properties and quench behavior. We present
31
a finite element approach which integrates the various effects into the commercial
32
33 software ANSYS by means of programming new element types. This is shown
34 capable of simulating the strongly coupled transient electromagnetic, thermal, and
35 circuit behavior of superconducting magnets required for quench protection studies.
36 A benchmarking study is presented which shows close agreement between the new
37 ANSYS elements and a COMSOL Multiphysics implementation developed at CERN
38 for dump resistor and Coupling Loss Induced Quench (CLIQ) based magnet protection
39 of a Nb3 Sn block dipole. Following this, the ANSYS implementation is shown
40 reproducing strongly coupled quench back behavior observed during the test of a Nb3 Sn
41 superconducting undulator prototype at Lawrence Berkeley National Laboratory.
42
43
44
45
46 Submitted to: Supercond. Sci. Technol.
47
48
49
50 Keywords: superconducting magnets, multiphysics modeling, finite element, quench
51 protection, superconducting undulators
52
53
54 1. Introduction
55
56 Transient behavior of superconducting magnets is frequently determined by multiscale
57 and multiphysics phenomena. A common example of this occurs when a quench
58
59 protection system (QPS) is activated in an attempt to safely bring down the magnet
60
 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1 Page 2 of 24

1
2
3 2
4
5 current. To design a system which prevents damage to the magnet, it is critical to be
6 able to accurately simulate the temperature and voltage rise during the current decay.
7
This requires modeling phenomena such as quench back due to eddy currents in the
8
9 conductor or structural material, current sharing and quench propagation within the
10 conductor, and inductive and resistive coupling of the various effects to the magnet’s
11 QPS circuit.
12
13 Previous work has focused on simulating these challenging multiphysics problems
14 using laboratory developed finite element codes [1, 2, 3, 4], lumped circuit element
15 models [5], customization of the commercial software COMSOL Multiphysics [6], and
16
17 the coupling of multiple codes or softwares using co-simulation [7, 8]. We present a
18 new approach using only the commercial finite element software ANSYS [9]. It has
19 previously been demonstrated that ANSYS has the capability to simulate some aspects
20
21 of the complete problem, such as quench propagation [10, 11, 12] and the effect of eddy
22 currents in mechanical structures on the current decay of a circuit coupled magnet [13].
23 We show user defined elements replicating these features and adding the additional
24
25 missing capabilities of: (1) modeling magnetization of the conductor due to coupling
26 currents and (2) combining all the effects into a single, coupled simulation with field
27 and temperature dependent material properties.
28
29 The ability for a user to define their own element type is a documented feature
30 of ANSYS, for which the authors are aware of two previous examples relevant to
31 electromagnetic applications [14, 15]. The creation of a user element is accomplished
32
33 by writing the Fortran code which defines the element’s properties and builds the finite
34 element matrices. A custom ANSYS executable is then compiled which allows for the use
35 of this element as if it were included in the standard distribution (making it compatible
36
37 with all geometry generation, meshing, solving, and post-processing features). User
38 programmed generation of the element matrices gives full control over the choice of
39 element shape functions, integration points, material properties, and FEM formulation.
40
41 Two elements are used to implement a custom FEM approach, with the first being
42 an electromagnetic element with optional coupling to an external circuit. An equivalent
43 magnetization term is included in the vector potential formulation to model interfilament
44
45 coupling loss (IFCL) within the conductor. A second, thermal element is used to
46 model the temperature rise due to quench induced ohmic heating and IFCL. These two
47 elements are coupled using the Multi-field Solver in ANSYS such that magnetic field,
48
49 temperature, and various loads are shared between them (see figure 1). We present an
50 initial benchmarking study with results for a Nb3 Sn dipole magnet compared between
51 this approach and a similar implementation in COMSOL Multiphysics [16]. Following
52
53 this, behavior predicted by the user elements is compared to data taken during the test
54 of a prototype Nb3 Sn undulator magnet at Lawrence Berkeley National Laboratory.
55
56
57
58
59
60
Page 3 of 24 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1

1
2
3 3
4
5
6 Electromagnetic Field Thermal Field
7
8
EM Meshed Areas TH Meshed Areas
9
10 iron, insulation, structure, air temperature
iron, insulation, structure
11 • default element (PLANE53)
joule heating • default element (PLANE77)
• eddy currents in structure
12 • mat. prop. (temp)
• mat. prop. (temp)
13
conductor
14 • user defined EM element
temperature
conductor
15 • interfilament coupling loss joule heating
• user defined TH element
16 • quench
B, quench state • mat. prop. (temp, B, quench state)
17 • mat. prop. (temp, B)
18
19 coupled with stranded formulation
20
21 QPS: dump resistor, CLIQ, etc
Circuit • default elem. (CIRCU124)
22
23
24
25 Figure 1. An overview of coupled electromagnetic, circuit, and thermal simulation in
26 ANSYS with user defined elements is shown. Such an approach allows for simulating
27 the impact of interfilament coupling loss, quench, and structural eddy currents on
28
magnet behavior while including temperature and field dependent material properties.
29
The independently meshed electromagnetic and thermal domains are coupled using
30
31 the Multi-field Solver as described in section 3.
32
33
34 2. The finite element model
35
36 An electromagnetic and thermal model was developed for use in conductor regions
37 where superconducting effects are desired (as illustrated in figure 1). These models
38
39 were implemented in ANSYS by the creation of two user elements. This approach
40 integrates the desired effects at the point of element matrix generation, no longer
41 requiring manual updating of superconducting properties between a stop and restart of
42
43
the solver as implemented in previous work [11, 12]. The thermal model follows directly
44 from ANSYS (see documentation in [17]), with extended capabilities of programmable
45 material property fits and automatic quench checking and heat generation. The material
46
47
properties are homogenized during element matrix generation based on specified
48 fractions of conductor, superconductor, and insulation using a method similar to what
49 is described in [6]. The electromagnetic model is based on the vector potential approach
50
51
used in ANSYS, with modifications to the formulation made to include the effects of
52 quench and interfilament coupling loss. The following subsections describe this model.
53
54
55 2.1. Vector potential with equivalent magnetization for IFCL
56
57
The default approach to modeling eddy currents in ANSYS uses the A,V-A formulation
58 with both vector potential A and electric scalar potential V degrees of freedom in
59 conducting regions [18], requiring modeling and meshing of the conductive paths in
60
 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1 Page 4 of 24

1
2
3 4
4
5 which the induced currents flow. This is impractical for the simulation of interfilament
6 coupling currents in a 2D magnet cross section due to filament sizes on the order of 5 to
7
100 µm and the 3D nature of the induced current path. An alternative approach, which
8
9 does not require a mesh density on the same order of the induced currents, assumes a
10 predetermined current path producing an equivalent magnetization, with time constant
11 τ , of
12
13
14
2τ ∂B
Me = − . (1)
15 µ0 ∂t
16 This approach is found applied to the case of a twisted filamentary composite in a
17
18 uniform, changing transverse field (relevant for 2D simulations of multifilamentary
19 superconducting strands) in [19, 20] and has been implemented in several magnet
20 modeling codes [1, 5, 6]. With assumptions about the filament layout within the strand
21
22 and the resulting current loops, an interfilament coupling loss (IFCL) time constant of
23 2
24

µ0 L
25 τ= , (2)
2ρet 2π
26
27 is written in terms of an effective transverse resistivity of the strand matrix ρet and
28 filament twist pitch L along the length of the strand. Limitations of this approximation
29
30 and a more detailed approach can be found summarized in [21]. The induced currents
31 deposit energy as heat within the strand matrix with a power per unit volume of
32
33 ∂B
34 P e = Me · , (3)
∂t
35
36 which in many cases leads to IFCL being an effective quench back mechanism.
37 The equivalent magnetization approach includes the effects of eddy currents without
38 the need for an additional degree of freedom (DOF), and the finite element formulation
39
40 in 3D is derived from the vector potential only,
41
42 ∇ × ν∇ × A − ∇ × Me = Js , (4)
43
44 with Js as a source current density and ν = µ−1 . Considering the form of Me , the
45
46
differential equation to be solved using the FEM is
47
48 ∂A
∇ × ν∇ × A + ∇ × 2τ ν∇ × = Js . (5)
49 ∂t
50 Here it is seen the addition of the magnetization term introduces a damping matrix
51
52 of similar form (curl-curl) as the stiffness matrix used for a typical vector potential
53 element. To implement the FEM, the weak integral of equation 5 is taken with test
54 functions chosen to be the same as the shape functions carrying the DOF within the
55
56 element [18]. This leads to
57
58  AA   ∂
{A} + C AA {A} = {J s } ,

K (6)
59 ∂t
60
Page 5 of 24 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1

1
2
3 5
4    
5 from which the element stiffness matrix K AA , damping matrix C AA , and load vector
6 {J s } are extracted. With the vector potential within the element written in terms of
7
the shape function matrix [NA ] and the DOF values at the element nodes {Ae },
8
9
10 {A} = [NA ]T {Ae } , (7)
11
12 the resulting element matrices and load vector are
13
14
Z h
 AA  1
T
i
15 K = ∇ × [NA ]T ∇ × [NA ]T dVelem (8)
µ0
16
17  AA  2τ
Z h

T
i
18 C = ∇ × [NA ]T ∇ × [NA ]T dVelem (9)
19 µ0
20 Z
21
 S
J = {JS } [NA ]T dVelem . (10)
22
23
24
Here the equations are given in general 3D form in terms of the nodal potential (assuming
25 {A} = {Ax , Ay , Az }). For the 2D user element these simply reduce to a single component
26 Az , and are evaluated using Gaussian quadrature with similar shape functions and
27
28
integration points as used by ANSYS.
29
30 2.2. Coupling to an external circuit as a stranded conductor
31
32 Regions modeled as coils are coupled to an external circuit following the method
33
34
developed by ANSYS [22], with several modifications to account for IFCL, quench
35 effects, and separate effective lengths for coil resistance and inductance. A stranded
36 formulation is used which adds a current i and voltage e DOF to the vector potential
37
38
given in equation 5. Both these new DOF are constrained to be single values for a
39 modeled coil region, with i being the current per stranded coil turn, and e being the
40 voltage drop across the coil. The voltage drop is made up of both a resistive eR and
41
42
inductive eL contribution such that
43
44 e = eR + eL . (11)
45
46 For a general stranded coil of fixed cross section and resistivity ρ, the resistive voltage
47 is given by
48
49  2 Z
50 Nc
eR = i ρdV, (12)
51 Sc
52
53 where Nc is the number of coil turns and Sc is the modeled coil area. As will be
54 described in section 2.3, quench and current sharing effects are accounted for using
55 a single parameter If cu representing the fraction of current assumed to be flowing in
56
57 the stabilizing material compared with the superconductor. With this assumption, the
58 contribution to the resistive voltage is determined by the fractional area of stabilizer
59
60
 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1 Page 6 of 24

1
2
3 6
4
5 and the amount of current assumed to be flowing in this region. Adjusting equation 12
6 to account for these assumptions, the resistive voltage drop is given by
7
8  2 Z
Nc If cu
9 eR = iLc ρst dS, (13)
10 Sc fcond (1 − fsc )
11 where Lc is introduced as an effective length chosen to best match the resistance of the
12
13 3D coil, fcond defines the fraction of the total coil area which is conductor, fsc defines
14 the fraction of superconductor within this conductor area, and ρst is the resistivity of
15 the stabilizing material. In case of multi-strand cables the factor Lc is also accounting
16
17 for the transposition pitch.
18 The inductive voltage is given by the time derivative of the linked flux
19
20
Z
∂Φ ∂
21 eL = = Nc (t̂ · A)dS, (14)
∂t ∂t
22
23 which for a 2D model with current restricted to flow only in the z direction is given by
24
25
Z
Nc ∂Az
26 eL = Li t dS. (15)
Sc ∂t
27
28 Here the variable t is positive or negative one based on the direction of the current, and
29 Li is introduced as an effective length scaling of the inductance chosen to match the 3D
30
31
magnet.
32 Circuit coupling is accomplished by the addition of equation 11 to the original
33 vector potential formulation in equation 6 (with the source current density now derived
34
35
from the current per turn DOF using the proper winding function relations in the K Ai
36 matrix). This leads to the coupled equations
37
38  ∂
C AA {A} + K AA {A} + K Ai {i} = {0}
    
39 (16)
∂t
40
41  eA  ∂
{A} + [K ee ] {e} + K ei {i} = {0} ,
 
42 C (17)
43 ∂t
44 where the matrices shown are matched to the DOF dependence of each term. With
45 respect to the original ANSYS model, a new matrix C AA is included for interfilament
46
47 coupling loss, and the voltage balance matrices K ei and C eA are derived from equations
48 13 and 15 to include the changes mentioned. The discrete form for the circuit coupled
49 element (similar to equations 8-10), and the method by which the e and i DOF’s are
50
51 coupled to other external circuit elements can be found in the appendix.
52
53 2.3. Current sharing and quench
54
55 The stranded formulation assumes a uniform current density and therefore does not solve
56
57 for the distribution of current within the superconductor or stabilizing regions which
58 changes based on the superconducting properties and quench behavior (this could be
59 done, for example, with A − V or H formulations). The choice of a stranded approach
60
Page 7 of 24 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1

1
2
3 7
4
5 is motivated by modeling of multifilamentary conductors typical of Nb-Ti and Nb3 Sn
6 magnets. In order to capture the impact of the effect of quench on ohmic loss and
7
resistance growth, a single scaling parameter If cu is used to represent the fraction of
8
9 current assumed to be in the stabilizing material. With this definition, three distinct
10 quench states can be defined as shown in table 1.
11
12
13 Table 1. Quench States
14 Regime Fraction of Current in Stabilizer
15
16 Fully Superconducting If cu = 0
17 Current Sharing 0 < If cu < 1
18 Fully Quenched If cu = 1
19
20
21
22
23 The conductor is assumed to have a critical surface parametrized by magnetic field,
24 temperature, and transport current density beneath which it is fully superconducting.
25
26 Strain dependence of this critical surface is neglected at this time. Above the critical
27 surface, some (current sharing) or all (fully quenched) of the current is assumed to
28 move from the superconductor into the stabilizing material, leading to resistive loss.
29
30 Accurate assumptions for the variation of If cu during current sharing is dependent on the
31 superconducting material and the significance of this quench state to magnet behavior
32 being simulated. An exhaustive review of this topic is found in Chapter 18 of [23]. A
33
34 common assumption for the current sharing regime of Nb3 Sn and Nb-Ti conductor is a
35 linear variation of If cu with temperature T
36
37 TcB − T
38 If cu = 1 − (18)
TcB − Tcs
39
40 from the point at which the critical current equals the transport current Tcs up to the
41 temperature TcB at which the transport current is equal to zero in the superconductor
42 [24].
43
44
45 2.4. Homogenized joule heating
46
47 The two sources of element heating are resistive loss due to quench and IFCL. These
48 losses are homogenized to account for fill factors of conductor and superconductor within
49
50 the modeled coil region. The parameter fcond is used to define the fraction of the total
51 coil area which is conductor, and fsc to define the fraction of superconductor within this
52 conductor area. Quench induced loss is assumed to occur only within the stabilizer of
53
54 the conductor, with the magnitude dependent on the quench state (see table 1) and the
55 resistivity of the stabilizing material. If Je is the element current density, the power per
56 unit volume of modeled conductor is then given by
57
58
(If cu Je )2
59 Pres = ρst (19)
60 fcond (1 − fsc )
 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1 Page 8 of 24

1
2
3 8
4
5 for quench based loss, and
6
2
7 fcond dB
8 Pe = 2τ (20)
µ0 dt
9
10 for losses resulting from interfilament coupling currents (see equation 3). For a given
11
quench state, the temperature and field dependence of these losses is driven by the
12
13 variation of the stabilizer’s resistivity ρst . For Nb-Ti and Nb3 Sn this stabilizer is typically
14 a high RRR copper, resulting in ρst (and then also τ for which ρet in equation 2 also
15 depends on ρst ) showing strong temperature dependence and magnetoresistive effects.
16
17
18 3. Coupled simulation using the Multi-field Solver
19
20
21
The ANSYS Multi-field Solver allows for solving of sequentially coupled problems with
22 independent meshes. A unique, meshed region is generated for each physics field and
23 load coupling interfaces for which loads will be passed between them are specified. Each
24
25
region is solved independently with its own time stepping and solution options. The
26 solver transfers the loads across the defined interfaces (even with dissimilar meshes),
27 and iterates between each physics field in sequence until the transfer of loads converges
28
29
for a user defined “stagger” time step as shown in figure 2.
30
31
32 ts Stagger Loop: ts to tf tf
33
ϕ𝐻𝐺𝐸𝑁
𝑖 = ϕ𝐻𝐺𝐸𝑁 𝐻𝐺𝐸𝑁
𝑖−1 + α 𝐻𝐺𝐸𝑁𝑖 − ϕ𝑖−1
34 ϕ𝑇𝐸𝑀𝑃
𝑖
35 Thermal
Electromagnetic
36 ts tf ts tf 𝑖 =𝑖+1

37
𝑖 =𝑖+1
38
39 ϕ𝑇𝐸𝑀𝑃
𝑖+1 = ϕ𝑇𝐸𝑀𝑃
𝑖 + α 𝑇𝐸𝑀𝑃𝑖 − ϕ𝑇𝐸𝑀𝑃
𝑖 no convergence of yes ϕ𝑇𝐸𝑀𝑃
𝑖+1 = ϕ𝑇𝐸𝑀𝑃
𝑖 +

40 ϕ𝐻𝐺𝐸𝑁
𝑖 , ϕ𝑇𝐸𝑀𝑃
𝑖
α(𝑇𝐸𝑀𝑃𝑖 − ϕ𝑇𝐸𝑀𝑃
𝑖 )

41
42
43
44
45 Figure 2. A stagger loop within the Multi-field Solver is shown for coupled
46 electromagnetic and thermal fields (see figure 1 for an example of how such a simulation
47 is set up with the user elements). In this example, the loads transferred between fields
48 are heat generation φHGEN and temperature φT EM P . This approach loops over the
49 stagger time step (from ts to tf ) with a relaxation factor α applied to the load transfer
50 until convergence of the loads is achieved. Separation of the problem into sequentially
51 defined stagger steps is used to simulate over the entire time domain.
52
53
54 This solver has been successfully used for fully coupled simulations including the
55 user elements (for example see the verification study in section 4). To do this, two
56
57 physics fields are created which are shown labeled as “electromagnetic” and “thermal”
58 in figure 1. A load transfer interface is specified between meshed coil regions and any
59 structural regions with eddy currents. This allows for passing Joule heat loads from
60
Page 9 of 24 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1

1
2
3 9
4
5 the electromagnetic region to the thermal region, and passing temperature back. Both
6 temperature and Joule heating are standard loads which may be transferred with the
7
Multi-field Solver. To allow for thermal material properties to also vary with magnetic
8
9 field and quench state, a workaround using a shared module was implemented to pass
10 these two variables in a non-standard use of the solver.
11
12
13 4. Benchmarking with COMSOL
14
15 A verification study was completed comparing results from the ANSYS user elements
16
17
to a similar 2D FEM implementation in COMSOL developed at CERN [6] within the
18 STEAM project [25]. The results from the full study are found in [26]. Effects such as
19 quench resistance, yoke saturation, IFCL, and structural eddy currents were compared
20
21
across several models with good agreement found. We present the results of one such
22 study which focuses on a simplified Nb3 Sn dipole magnet protected with a dump resistor
23 and Coupling Loss Induced Quench (CLIQ) [27].
24
25
26 4.1. The Nb3 Sn dipole model
27
28 A dipole model was designed to allow for comparison of results in a regime representative
29 of realistic Nb3 Sn accelerator magnets. This model is shown in figure 3, and a list of
30
31
high level parameters are given in table 2.
32
33
Table 2. Overview of the Nb3 Sn Dipole Model
34
35 Parameter Value Unit
36
37 inner layer turns (per quadrant) 14
38 outer layer turns (per quadrant) 18
39 turn width 1.5 mm
40 turn height 15 mm
41 strands (per turn) 40
42 strand diameter 0.75 mm
43 filament twist pitch 14 mm
44 fef f with ρet = fef f ρst 1.0
45 Cu RRR 200
46 non-Cu fraction 0.4
47
Nb3 Sn Jc (4.5 K, 12 T) 2040 A/mm2
48
short-sample current (4.5 K) 15.37 kA
49
50 short-sample cond. field (4.5 K) 11.7 T
51 Lc: effective res. coil length 10.11 m
52 Li: effective ind. coil length 9.2 m
53
54
55
56
57
58
59
60
 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1 Page 10 of 24

1
2
3 10
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27 Figure 3. The two conductor layers and iron yoke of the Nb3 Sn dipole ANSYS model
28 are shown with the dump resistor circuit (a) and the CLIQ circuit (b). The DOF of
29 the meshed conductor regions of these models are coupled to their respective circuits,
30 allowing for consistent simulation of both electromagnetic and circuit effects. A similar
31 model consisting of only the conductor region is used for the thermal domain during
32 the Multi-field solution
33
34
35 4.2. Protection with a dump resistor
36
37 A first comparison between ANSYS and COMSOL was performed for a dump resistor
38
39
extraction exhibiting strongly coupled electromagnetic and thermal behavior. A simple
40 circuit was built using CIRCU124 elements consisting of a resistor, voltage source, and
41 a stranded coil element coupled in e and i to the coil region in the meshed model (see
42
figure 3). A dump resistor value of 30 mΩ was used for all tests. The simulation begins
43
44 with the magnet operating in a static condition at 4.5 K and 13.8 kA. This operating
45 point is slightly less than 90 % of the magnet’s short-sample limit. At 5.0 ms the voltage
46
source is ramped down to zero over 0.1 ms to effectively put the magnet in series with the
47
48 dump resistor only. The full details of this simulation including the material property
49 fits are found in [26].
50
The behavior of the magnet was studied with increasing levels of detail as outlined
51
52 in table 3. A first simulation was performed with no IFCL or quench losses, making the
53 current decay dependent only on the magnet’s inductance. A second simulation added
54
IFCL which influences the current decay by changing the differential inductance of the
55
56 magnet. For this case, coupling to a thermal model was also included to capture effects
57 on τ due to changes in material properties from heating of the conductor. For the final
58
simulation, current sharing and quench were added to the previous case, allowing for
59
60
Page 11 of 24 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1

1
2
3 11
4
5 IFCL induced quench resistance growth.
6
7
8 Table 3. Dump Resistor Verification Tests
9 Yoke Saturation IFCL Thermal Coupling Quench
10
11 No Loss x
12 IFCL Only x x x
13 IFCL w/Quench x x x x
14
15
16
17
18 Agreement between ANSYS and COMSOL results for the magnet current decay is
19
shown in figure 4 for the final simulation case outlined in table 3. This magnet exhibits
20
21 strong quench back behavior with IFCL heating the coil to quench, after which the
22 coil resistance growth rapidly increases rate of magnet current decay. This is further
23
illustrated by figure 5 where the rapid growth of hotspot temperature and coil resistance
24
25 in the final, fully coupled case (IFCL w/Quench) is compared between codes. The energy
26 loss for each mechanism is compared to the total change in energy of the system in table
27
4.
28
29
30 Table 4. Energy Loss Comparison for IFCL w/Quench Case
31
32 Location ANSYS COMSOL
33
E (kJ) % E (kJ) %
34
Dump Res. 505.96 34.45 497.25 33.80
35
36 IFCL 14.32 0.98 14.24 0.97
37 Coil Res. 948.24 64.56 959.55 65.23
38
39 % energy deposition is based on a total energy change between 5 and 500 ms.
40
41
42
43
44 4.3. Protection with CLIQ
45
46 A second study compares results for a Coupling Loss Induced Quench (CLIQ) discharge
47
48 [27] in a layer-layer configuration as shown in figure 3. CLIQ is a protection scheme
49 which seeks to quench large portions of the magnet coil in a distributed fashion in
50 order to rapidly bring down the magnet current using coil quench resistance. At the
51
52 detection of a quench, a capacitor bank is discharged across one or more sections of
53 the magnet coil which generates an oscillation of current in the coil sections about
54 the nominal magnet transport current (see figure 6). The resulting field oscillations
55
56
induce interfilament coupling currents which heat portions of the coil to quench. This
57 approach is typically considered for large inductance accelerator magnets where dump
58 resistor based protection is no longer possible due to voltage and hotspot temperature
59
60
constraints. CLIQ protection scales well to large inductance magnets because the current
 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1 Page 12 of 24

1
2
3 12
4
16
5
ANSYS No Loss
6 13.8 kA
ANSYS IFCL Only
7 ANSYS IFCL w/Quench
8 COMSOL IFCL w/Quench
9
10 12
11
12 5 ms 50 ms
13
14
15
I (kA)

16 8
17
18
19
20
21
22 4
23
24
25
26
27
0
28 0 125 250 375 500
29
Time (ms)
30
31 Figure 4. ANSYS and COMSOL results are compared for a dump resistor extraction
32 exhibiting strongly coupled electromagnetic and thermal behavior. Effects are built up
33
from no losses, to only IFCL, and finally to a fully coupled case where quench and IFCL
34
are considered. A slight difference in the initial current decay between the “No Loss”
35
36 and other two cases is seen up to about 20 ms. Here the coil is not yet quenched, but
37 the influence of coupling currents on the magnet’s differential inductance causes the
38 current to decay faster. At 20 ms, quench resistance growth begins to play a dominant
39 role as “IFCL w/Quench” quickly decays away from “IFCL Only”. This is an example
40 of quench back, where the rapid field change initiated by the dump resistor induces
41 IFCL which heats the coil to quench. The resistance rise due to quench drives the
42 current down much faster than a case where the effects of IFCL and quench are not
43 considered.
44
45
46 oscillations are induced between sections of the magnet, and the relevant inductance for
47
48
inducing these currents is much lower than total magnet inductance.
49 In this comparison, the dump resistor in the main circuit is set to zero, and a
50 CLIQ resistor of 15 mΩ and capacitor of 35 mF are added. For the results shown, the
51
capacitor was charged to 350 V. The CLIQ system is activated between 5.00 and 5.01
52
53 ms with the magnet previously set up in a static condition at 4.5 K and 13.8 kA (90%
54 of short-sample). The CLIQ discharge induces current oscillation between the magnet
55
layers on a similar scale of the IFCL time constant (see the no loss case in figure 6).
56
57 Figures 7 and 8 show how the IFCL heating induced by this oscillation drives the coil
58 to quench, with the resulting coil resistance growth bringing down the magnet current.
59
Comparison between the ANSYS and COMSOL results in these figures show excellent
60
Page 13 of 24 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1

1
2
3 13
4 200 160
5
6 180
140
7
8 160
120
9 140
10

Hotspot Temp. (K)

100
11
Rcoil (mOhm)

120
12
13 100 80

14 80
15 60
16 60
17 40
18 40

19 ANSYS Coil Resistance 20

20
20 ANSYS Hotspot Temp
COMSOL
21 0 0
22 0 100 200 300 400 500 600
Time (ms)
23
24 Figure 5. The rise in coil resistance and peak coil temperature is compared between
25
ANSYS and COMSOL for the final, fully coupled simulation (IFCL w/Quench).
26
27
28
29 agreement for this fully coupled electromagnetic, thermal, and circuit simulation.
30
31 ANSYS I1
32 15 ANSYS I2
33
34
I (kA)

14
35
36
37 13
38
39 12
40 0 5 10 15 20 25 30 35 40 45 50
41 Time (ms)
42
43 Figure 6. The induced current oscillation in the two coil sections about the starting
44 current of 13.8 kA from a CLIQ discharge is shown for a no loss case. The oscillation
45 period is similar to the IFCL time constant in the high field region.
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1 Page 14 of 24

1
2
3 14
4 16
5 ANSYS I1
6 14
ANSYS I2
7 ANSYS ICLIQ
COMSOL
8 12
9
10 10
11
12 8
13
I (kA)

14
6
15
16
4
17
18
2
19
20
0
21
22
-2
23 0 20 40 60 80 100 120 140 160 180 200
24 Time (ms)
25
26 Figure 7. The currents in the different coil sections following a CLIQ discharge are
27 compared between the ANSYS and COMSOL models. The induced field oscillations
28 generate IFCL heating which leads to quenching of a large portion of the coil (also see
29 figure 8).
30
31
32
33
34
35
36
37
38 500 150
39 ANSYS Coil Resistance
40 450 ANSYS Hotspot Temp
COMSOL
41
400
42
43 350
44 100
Hotspot Temp. (K)

45
Rcoil (mOhm)

300
46
47 250

48 200
49
50
50 150
51
52 100

53 50
54
55 0 0
56 0 50 100 150 200 250 300 350
Time (ms)
57
58 Figure 8. The rise in coil resistance and peak coil temperature is compared between
59 ANSYS and COMSOL following the CLIQ discharge.
60
Page 15 of 24 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1

1
2
3 15
4
5 5. Comparison to Nb3 Sn undulator test data
6
7 A series of Nb3 Sn superconducting undulators (SCUs) were built and tested at Lawrence
8 Berkeley National Laboratory as part of a R&D program for LCLS-II and future free
9
10 electron lasers [28]. These magnets consist of two ferromagnetic iron cores with machined
11 pockets along the length. The pockets are wound with superconducting strand such that
12 the current polarity changes from one pocket to the next along the length. The two
13
14 cores are assembled together with a gap left between them for an electron beam. The
15 interaction of the beam with the alternating fields generated by the SCU serves as the
16 radiation source in free electron lasers.
17
18 These magnets are well suited for benchmarking the user elements due to: (1) the
19 use of single strands eliminating cable based coupling currents found in many other
20 Nb3 Sn magnets and (2) exhibiting strong quench back behavior changing as a function
21
22 of field level. The test of a short prototype SCU was selected for a first comparison
23 due to existing data for a series of pre-quench, dump resistor extractions at increasing
24 levels of initial magnet current. This allows for benchmarking of the user elements with
25
26 test data over a wide range quench back behavior. An overview of the properties of this
27 magnet is found in table 5 and a picture of the cross section can be seen in figure 9.
28 This prototype corresponds to a short, 80 cm, length single core, whereas the final SCU
29
30 magnet includes two cores of 1.4 m length assembled together.
31
32 Table 5. Overview of the Nb3 Sn SCU Short Model
33
34 Parameter Value Unit
35
period length 19 mm
36
37 pocket width 6.32 mm
38 pocket height 4.67 mm
39 powered turns per pocket (left to right in sym.) 7-35-56-56
40 strand architecture 132/169 RRP R
41 strand diameter 0.6 mm
42 filament twist pitch 12 mm
43 Cu RRR 250
44 non-Cu fraction 0.45
45 fef f with ρet = fef f ρst 4.0
46 effective resistive coil length (Lc ) 97.1 mm
47 effective inductive coil length (Li ) 90 mm
48
Nb3 Sn Jc (4.5 K, 10 T) 2880 A/mm2
49
short-sample current (4.5 K) 965 A
50
51 short-sample cond. field (4.5 K) 5.2 T
52
53
54
55
56
The ANSYS model was matched to the test configuration of a single, short
57 prototype magnet as seen in figure 9. The effective lengths Lc and Li , which scale
58 resistive and inductive effects from 2D to 3D, were chosen to match the physical length
59
60
of the coil and the linked flux of a 3D ANSYS model. The use of a large scaling factor
 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1 Page 16 of 24

1
2
3 16
4
5 (a) (b)
6
7
8
9
10
11
12
13
14
15
16
17 Figure 9. (a) The electromagnetic ANSYS model is shown in a dump resistor circuit.
18 The meshed regions consist of ferromagnetic low carbon steel (light blue), un-powered
19 corrector turns (magenta), powered turns (teal), glass filled epoxy (orange), and air
20
which extends beyond the top and left boarders for far field boundary conditions (dark
21
blue). This is coupled to a similar thermal region (without air) using the Multi-field
22
23 Solver to simulate IFCL induced quench back in an external dump resistor protection
24 circuit. (b) A cross-section of the short prototype Nb3 Sn undulator with the symmetric
25 region marked.
26
27
28 fef f between the matrix resistivity and an effective resistivity for all coupling losses is
29 based on measurements made on a similar Nb3 Sn strand [29, 30]. For each case, a single
30
31
static solution was first solved with the voltage source set to produce the initial current
32 matching the test. This source was then ramped to zero over 0.1 ms to effectively
33 place the magnet in series with a fixed resistance dump resistor for current extraction.
34
35
The solution proceeds with transient effects using the Multi-field Solver for coupling of
36 independently meshed electromagnetic and thermal domains as described in section 3.
37 To match data taken during the test, current extractions with no initial quench
38
39
were simulated from 400 to 800 A using a dump resistor of 48.1 mΩ. Figure 10 shows
40 the current decay curves from these simulations including cases with and without IFCL
41 and quench effects. When normalized to peak current, the “No Loss” cases show only
42
43
a small variation with initial current which is the result of non-linear magnetization of
44 the iron core with field level. At the 400 A level, the “IFCL w/Quench” case remains
45 superconducting due to IFCL heating not being able to overcome the large margin to
46
quench. This results in a current decay which shows little deviation from the no loss
47
48 case. As the initial current increases up to 800 A, the margin is reduced and IFCL
49 heating grows. At a certain level this begins to induce quench, adding resistance to
50
the circuit and driving the current down faster (quench back). As expected, the degree
51
52 of quench back is seen increasing with initial current (this is particularly clear when
53 comparing the normalized current decay).
54
A sensitivity study of the ANSYS results to input parameters and material property
55
56 fits was performed to allow for a more detailed comparison to test data. Table 6
57 summarizes this study and the results. The deviation from the nominal case was
58
evaluated on a parameter by parameter basis by comparing both the quench integral and
59
60
Page 17 of 24 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1

1
2
3 17
4 a) b)
5 800 800A IFCL w/Quench 1 800A IFCL w/Quench
6 700A IFCL w/Quench 700A IFCL w/Quench
0.9
7 700 600A IFCL w/Quench
500A IFCL w/Quench
600A IFCL w/Quench
500A IFCL w/Quench
8 600
400A IFCL w/Quench 0.8 400A IFCL w/Quench
800A No Loss 800A No Loss
9 700A No Loss
0.7
700A No Loss
10 500 600A No Loss 0.6 600A No Loss

I/Ipeak
I (A)

500A No Loss 500A No Loss

11 400 400A No Loss 0.5 400A No Loss

12 0.4
300
13 0.3
14 200
0.2
15 100
0.1
16
17 0
0 10 20 30 40 50
0
0 10 20 30 40 50
18 Time (ms) Time (ms)
19
20 Figure 10. Magnet current decay curves from ANSYS are compared with and without
21 the effects of IFCL and quench for absolute (a) and normalized current (b). The impact
22 of quench back is seen increasing with initial current level. The large thermal margin
23 and smaller rate of field change (driving IFCL) at the 400 A current level results in
24 the magnet staying superconducting and showing minimal deviation from the no loss
25 case. As the initial current increases up to 800 A, IFCL induced quench plays a larger
26 role, with coil resistance growth driving the current down more quickly than the no
27 loss case.
28
29
30 peak coil hotspot temperature for each initial current level. The quench integral is the
31
32 time integral of the square of magnet current from the start to end of the decay. This
33 is a material property independent measure of the total energy deposited during the
34 decay which would generate joule heating at a quench location. The quench integral is
35
36
typically used for an estimation of quench location hotspot temperature in the adiabatic
37 limit with the material properties considered [19], and in this case serves as a metric
38 for the degree of quench back when compared to a no loss case. In the future, a more
39
40
detailed metric could be used along with an exploration of combined effects due to
41 multi-parameter deviation from the nominal case.
42 The results show the most sensitive fits and parameters are those associated with
43
44
the resistivity of the copper matrix (fef f , Cu RRR, and Cu resistivity fit). This is not
45 unexpected, as these impact the IFCL time constant and coil resistance. A range of
46 behavior about the nominal case due to changes in Cu resistivity was created to visualize
47
48
this sensitivity when comparing to test data. This range is bounded by two curves which
49 represent the extremes of the material property fits and RRR. One curve corresponds
50 to a RRR of 150 and the fit from CUDI, and the other a RRR of 350 and the fit from
51
52
MATPRO. Figure 11 compares this range and the nominal case to test data. Current
53 decay curves are shown from 400 to 800 A, along with the quench integral of these decays
54 from 5 to 50 ms. The ANSYS simulations are seen reproducing the trend seen in the
55
measured data of larger deviation from the no loss case at higher current. The source of
56
57 the remaining difference between the ANSYS predictions and test data could originate
58 from the 2D elements not including longitudinal quench propagation or 3D effects on
59
peak field, inductance, and coupling loss. In addition, further study of the accuracy
60
 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1 Page 18 of 24

1
2
3 18
4
5 Table 6. Sensitivity Study of ANSYS Results for the SCU
6
7 Variation of Results a : ∆Qint (%), ∆Thot (K)
b
8 Parameter/Fit Nominal Changed 400 A 500 A 600 A 700 A 800 A
9
2.0 -4.6, 3.0 -6.4, 2.1 -4.7, 0.7 -3.0, 0.1 -1.1, 0.0
10 fef f 4.0
6.0 2.3, -1.3 4.5, -5.9 5.6, -1.3 3.8, -0.5 2.4,-0.2
11
12 150 1.9, -1.0 3.2, -0.8 0.4, 1.3 -2.3, 1.7 -4.2, 1.8
Cu RRR 250
13 350 -1.1, 0.4 -1.2, 0.0 0.0, -0.6 1.1, -0.8 2.1,-0.9
14 CUDI 0.9, -0.5 1.9, -0.8 0.4, 0.7 -1.6, 1.4 -3.7, 2.2
Cu resistivity NIST
15 MATPRO -0.1, 0.0 0.0, -0.2 0.5, -0.6 1.4, -1.0 2.4, -1.4
16 Cu heat capacity NIST CUDI 0.0, 0.0 0.0, 0.0 0.0, -0.1 0.0, -0.1 0.1, -0.2
17 Cu thermal cond. NIST CUDI 0.0, 0.0 0.0, 0.0 0.0, 0.0 0.0, 0.0 0.0, 0.0
18 Nb3 Sn heat capacity NIST CUDI 0.0, -0.4 1.3, -1.0 1.9, 0.0 1.8, 0.5 1.7, 1.4
19 G10 heat capacity c NIST Fermilab 0.0, 0.6 -2.6, 1.5 -3.4, 0.8 -3.7, 0.4 -3.8, 0.0
20 G10⊥ therm. cond. CryoComp NIST 0.0, -0.4 0.9, -1.1 0.9, -0.9 0.9, -0.7 1.0, -0.7
21
22
a
23 based on deviation of the results from the nominal case for the variation of the single listed
24 parameter. ∆Qint is the change in quench integral and ∆Thot is the change in peak hotspot
25 temperature of the coil; b all material property fits can be found in [31]; c G10 properties (normal) are
26 used for the epoxy impregnated glass fiber region between strands
27
28
29
30 of equivalent magnetization models for coupling currents in this application may be
31 revealing. Despite room for further study, the level of agreement between ANSYS and
32
33 test data is a promising sign the user elements can be used to understand and predict
34 quench back effects in Nb3 Sn SCU’s.
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Page 19 of 24 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1

1
2
3 19
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26 Figure 11. Comparison of ANSYS results with test data shows both exhibiting similar
27 quench back behavior over a range of initial currents. Deviation from the no loss cases
28 is minimal at low current and increases with current level as quench back becomes
29 more prominent. To visualize the sensitivity of the ANSYS model to the parameters
30 and material fits affecting Cu resistivity, a range is plotted around the nominal case
31 corresponding to the limiting behavior found in the sensitivity study (as described in
32 table 6 and section 5).
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1 Page 20 of 24

1
2
3 20
4
5 6. Conclusion
6
7 User defined elements in ANSYS allow for programing the generation of the finite
8 element matrices, providing the opportunity to implement new formulations and
9
10 material property fits. Two user defined elements were created as a new method for
11 simulating coupled, multiphysics behavior of superconducting magnets. These elements
12 extend the capabilities of ANSYS to now include superconducting specific phenomenon
13
14 of quench and interfilament coupling loss, while maintaining the meshing, solving, and
15 post-processing capabilities of the standard distribution. These new elements were
16 shown benchmarked against existing codes for a Nb3 Sn dipole and compared to test
17
18 data for a Nb3 Sn prototype undulator. In both cases the user elements were shown
19 predicting IFCL induced quench back, demonstrating for the first time that ANSYS
20 can be used to simulate this strongly coupled behavior required for accurate modeling
21
22 of many superconducting magnets.
23 This work is part of a larger effort within the US Magnet Development Program
24 and the Berkeley Center for Magnet Technology to advance analysis and modeling
25
26 capabilities for superconducting magnets [32]. It is our goal that this work becomes
27 a tool usable by the magnet design community. The effort to make these elements
28 available is underway, and interested parties are encouraged to contact the authors for
29
30 more information. Future work is focused on the extension of this approach to 3D
31 and towards simulation of HTS coated conductors and bulk superconducting devices by
32 implementing the E − J power law model within the A − V formulation.
33
34
35 7. Acknowledgments
36
37
Lucas Brouwer would like to thank Emmanuele Ravaioli of CERN (previously LBNL)
38
39 for many interesting and helpful discussions about the simulation of dynamic effects in
40 superconducting magnets and CLIQ based magnet protection. This work was supported
41
by the Director, Office of Science, High Energy Physics, and U.S. Department of Energy
42
43 under contract No. DE-AC02-05CH11231 in the context of broader collaboration with
44 the US Magnet Development Program.
45
46
47 Appendix
48
49 The matrices used for the 2D electromagnetic element with circuit coupling found in
50
51
equations 16 and 17 are the result the following steps. The K AA and C AA matrices in 2D
52 are found by simply reducing the general form already given in Equations 8 and 9. For
53 the stranded conductor, the source term is now supplied by the circuit and determines
54
55
the form of K Ai . Considering the weak integral of the source term in 2D (J is now a
56 scalar Jz ), and that the current density of the stranded conductor is derived from the
57 current per turn i using Jz = NScc ti leads to
58
59
60
Page 21 of 24 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1

1
2
3 21
4
5 Z Z
6
Nc
{N }Jz dS = {N } t{N }T {i}dS, (A.1)
7 Sc
8
where {N } is a vector of shape functions carrying the DOF across the element based
9
10 on the values at the element nodes (i = {N }T {ie } for example). Here clearly the K Ai
11 matrix is given by
12
13
Z
 Ai  tNc
14 K =− {N }{N }T dSelem . (A.2)
15
Sc
16 The 2D matrices in the voltage balance are derived from Equations 11, 13, and 15.
17 The e term is re-written as an area integral with the shape functions added, such that
18
19
20 Z Z
Nc ∂ 1
21 t Li {N } T
{Az}dS − {N }T {e}dS (A.3)
22 Sc ∂t Sc
23
 2 Z
Nc If cu
24 +Lc ρst {N }T {i}dS = 0.
25 Sc fcond (1 − fsc )
26
27 ANSYS assembles the three DOF using the following form
28        
29 C AA 0 0 ∂A
K AA
0 K Ai
A 0
30  eA   ∂t   ee ei   
C 0 0  0  +  0 K K   e  = 0 . (A.4)
 
31
32 0 0 0 0 0 0 0 i 0
33
34
To match the sizing of the submatrices here, the vectors in Equation A.3 are expanded
35 to square matrices using the outer product with an identity vector {I}. This leads to
36 2D element matrices of
37
38  eA  Nc
Z
39 C = t Li {I}{N }T dSelem (A.5)
40 Sc
41 Z
42 ee 1
[K ] = − {I}{N }T dSelem (A.6)
43 Sc
44
 2
45
Z
 ei  Nc If cu
46 K = Lc ρst {I}{N }T dSelem (A.7)
47
Sc fcond (1 − fsc )
48 The user element is coupled to an external circuit as voltage source using the
49
standard distribution, circuit element CIRCU124 [22]. With key option one set to select
50
51 a stranded coil, this element consists of three nodes labeled i, j, and k. The first two
52 nodes are connected to adjacent circuit elements and each carry a single voltage DOF.
53
The third node carries both a current and voltage drop DOF, and is chosen as one of
54
55 the nodes in the meshed coil region to make it part of the coupled set. The stiffness
56 matrix for the stranded coil CIRCU124 element is given by
57
58
59
60
 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1 Page 22 of 24

1
2
3 22
4
5     
6 0 0 1 0 Vi 0
7 0

0 −1 0 Vj  0
   
8   =  , (A.8)
−1 1 0 s  ik  0

9
10 0 0 0 0 ek 0
11
12
where s is a factor to account for modeling of a symmetric region. This couples the
13 stranded coil into an external circuit which may be made up of additional coil regions
14 or generic circuit elements selected using other key options for CIRCU124.
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Page 23 of 24 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1

1
2
3 23
4
5 References
6
7 [1] ROXIE: a Computer Code for the Integrated Design of Accelerator Magnets URL https:
8 //[Link]/roxie
9 [2] Guo X L, Wang L and Green M A 2012 Cryogenics 52 420
10 [3] Marinozzi V, Sorbi M, Manfreda G, Bellina F, Bajas H and Chlachidze G 2015 Physical Review
11 Special Topics - Accelerators and Beams 18 032401
12 [4] Breschi M, Cavallucci L, Ribani P L, Gavrilin A V and Weijers H W 2017 IEEE Trans. Appl.
13 Supercond. 27 4301013
14 [5] Ravaioli E, Auchmann B, Maciejewski M, ten Kate H H J and Verweij A P 2016 Cryogenics 80
15
346
16
[6] Bortot L, Auchmann B, Cortez-Garcia I, Fernandez-Navarro A M, Maciejewski M, Prioli M, Schops
17
18 S and Verweij A 2018 IEEE Trans. Appl. Supercond. 54 7000404
19 [7] Schops S, Gersem H D and Bartel A 2010 IEEE Trans. Magn. 46 3233
20 [8] Bortot L, Auchmann B, Cortez-Garcia I, Fernandez-Navarro A M, Maciejewski M, Mentink M,
21 Prioli M, Ravaioli E, Schops S and Verweij A 2018 IEEE Trans. Appl. Supercond. 28 4900706
22 [9] ANSYS URL [Link]
23 [10] Caspi S, Chiesa L, Ferracin P, Gourlay S A, Hafalia R, Hinkins R, Lietzke A F and Prestemon S
24 2003 IEEE Trans. Appl. Supercond. 13 1714
25 [11] Yamada R, Marscin E, Lee A and Wake M 2004 IEEE Trans. Appl. Supercond. 14 291
26 [12] Ferradas-Troitino J, Ambrosio G, Bajas H, Bordini B, Ferracin P, Fleiter J, Izquierdo-Bermudez
27 S, Vicente-Lorenzo-Gomez J, Carlos-Perezand J, Vallone G and Senatore C 2019 IEEE Trans.
28 Appl. Supercond. 29 4701306
29
[13] Brouwer L, Arbelaez D, Caspi S, Marchevsky M and Prestemon S 2018 IEEE Trans. Appl.
30
Supercond. 28 4001006
31
32 [14] Hauser A 1997 IEEE Trans. Magn. 33 1572
33 [15] Testoni P 2003 Implemenation in the ANSYS finite element code of the electric vector potential T-
34 Ω,Ω formulation and its validation with the magnetic vector potential A-V,A formulation Ph.D.
35 thesis University of Cagliari, Italy
36 [16] COMSOL Multiphysics URL [Link]
37 [17] 2016 ANSYS Mechanical APDL Theory Reference, Release 17.1
38 [18] Biro O and Preis K 1989 IEEE Trans. Magn. 25 3145
39 [19] Wilson M 1983 Superconducting Magnets (Oxford University Press) ISBN 0198548052
40 [20] Morgan G 1970 Journal of Applied Physics 41 3673
41 [21] Louzguiti L, Zani L, Ciazynski D, Turck B and Topin F 2016 IEEE Trans. Appl. Supercond. 26
42
4700905
43
[22] Wang J S 1996 IEEE Trans. Magn. 32 1071
44
45 [23] Russenschuck S 2010 Field Computation for Accelerator Magnets: Analytical and Numerical
46 Methods for Electromagnetic Design and Optimization (Wiley-VCH) ISBN 3527407693
47 [24] Stekly Z J J and Zar J L 1965 IEEE Transactions on Nuclear Science 12 367
48 [25] STEAM: Simulation of Transient Effects in Accelerator Magnets URL [Link]
49 ch/steam/
50 [26] Brouwer L, Auchmann B, Bortot L and Stubberud E 2019 LBNL Eng. Note: SU-1010-4841, R1.0
51 URL [Link]
52 [27] Ravaioli E 2015 CLIQ. A new quench protection technology for superconducting magnets Ph.D.
53 thesis University of Twente, Netherlands
54 [28] Emma P, Holtkamp N, Nuhn H, Arbelaez D, Corlett J, Myers S, Prestemon S, Schlueter R, Doose
55 C, Fuerst J, Hasse Q, Ivanyushenkov Y, Kasa M, Pile G, Trakhtenberg E and Gluskin E 2014
56
A plan for the development of superconducting undulator prototypes for LCLS-II and future
57
FELs FEL 2014 Conference Proceedings, Basel, Switzerland p THA03
58
59 [29] Zhou C, van Lanen E, Veldhuis D, ten Kate H, Dhalle M and Nihuis A 2011 IEEE Trans. Appl.
60
 AUTHOR SUBMITTED MANUSCRIPT - SUST-103276.R1 Page 24 of 24

1
2
3 24
4
5 Supercond. 21 2501–2504
6 [30] Zhou C, Miyoshi Y, van Lanen E, Dhalle M and Nihuis A 2012 Supercond. Sci. and Technol. 25
7 065018
8 [31] Manfreda G 2011 CERN Internal Note: EDMS NR 1178007
9 [32] Gourlay S, Prestemon S, Zlobin A, Cooley L and Larbalestier D 2016 The U.S. Magnet
10 Development Program Plan URL [Link]
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60