IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 34, NO.

2, JUNE 2019 973

Challenges of Finite Element Analysis of Resolvers

Hamid Saneie , Student Member, IEEE, Ramin Alipour-Sarabi , Student Member, IEEE,
Zahra Nasiri-Gheidari , Member, IEEE, and Farid Tootoonchian , Member, IEEE

Abstract—Nowadays, position sensors are increasingly used in

many applications. One of the most commonly used sensors are re-
solvers. There are different types of resolvers including wound ro-
tor resolvers, variable air-gap length variable reluctance resolvers,
and sinusoidal rotor variable reluctance resolvers. Regardless of
the resolvers’ configurations, including linear, cylindrical and disk
type, finite element analysis (FEA) is inseparable part of design
and the performance evaluation of these magnetic sensors. The
scope of this study is to discuss challenges in the FEA of resolvers.
The influence of mesh quality, time steps, stop time, excitation
methods including voltage source and current source and signal
processing methods are investigated in detail. Then, a key question
in analysis of resolvers is answered. The question is Is total har-
monic distortion of induced voltages’ envelopes a suitable criterion
for the performance evaluation of resolvers? Finally, four proto-
types of different types of resolvers are used for experimental tests.
Good agreements are obtained between simulation and test results
indicating the accuracy of following discussions.
Index Terms—Position Sensor, resolver, total harmonic distor-
tion (THD), position error, time stepping finite element method
(TSFEM).

I. INTRODUCTION
ESOLVERS as a type of electromagnetic position sensors
R have gained much research attentions in recent years. The
reason is related to the increasing use of inverter-driven elec-
tric motors that needs to accurately measure the rotor position
for d-q transformation [1] and speed for torque estimation. The
main competitors of resolvers are optical encoders. Although
in normal conditions commercial encoders present higher accu-
racy (x-arc sec. in encoders with respect x-arc min. in resolvers),
their accuracy is questioned in industrial applications and where
the reliability is important. In many industrial applications re-
solvers are preferred due to their distinguished specifications
[1]–[4]. Some of these features are the ability to work in highly
vibrated, highly polluted environments with high temperature
variation [2]–[6]. Absolute position estimation is easily achieved
Fig. 1. Different type of resolvers, (a): VAGL-VR cylindrical type, (b): VAGL-
VR cylindrical type, (c): VAGL-VR disk type, (d): SR-VR cylindrical type,
Manuscript received April 10, 2018; revised September 10, 2018; accepted
(e): SR-VR disk type, (f): WR cylindrical type, and (g): WR disk type.
November 5, 2018. Date of publication November 16, 2018; date of current
version May 2, 2019. This work was supported in part by the research office of
Sharif University of Technology, in part by the Niroo Research Institute, and
in part by the Iran National Science Foundation. Paper no. TEC-00431-2018. by the resolvers, contrary to optical encoders. Furthermore, due
(Corresponding author: Zahra Nasiri-Gheidari.)
H. Saneie, R. Alipour-Sarabi, and Z. Nasiri-Gheidari are with the Elec- to frameless structure of resolver it has a simpler assembling
trical Engineering Department, Sharif University of Technology, Tehran process to be integrated with a motor. Finally, simple structure
11365-11155, Iran (e-mail:, hamid.saneie@gmail.com; ramin.alipoursarabi of copper wires and silicon–steel laminations ensure that higher
@gmail.com; znasiri@sharif.edu).
F. Tootoonchian is with the Electrical Engineering Department, Iran Uni- reliability can be achieved by using resolvers [7].
versity of Science and Technology, Tehran 16846-13114, Iran (e-mail:, There are two types of resolvers: Variable reluctance (VR) re-
Tootoonchian@iust.ac.ir). solvers, shown in Fig. 1(a)–(e) and wound rotor (WR) resolvers,
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. shown in Fig. 1(f) and (g). The former type that works based on
Digital Object Identifier 10.1109/TEC.2018.2881465 the sinusoidal variation of air-gap reluctance is divided into two
0885-8969 © 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
974 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 34, NO. 2, JUNE 2019

groups: Variable Air-gap Length Variable Reluctance (VAGL- types of resolvers are chosen as samples. These resolver shown
VR) resolvers (Fig. 1(a)–(c)), and Sinusoidal Rotor Variable in Fig. 1(a), (b), (d) and (g) and named as R4, R1, R2 and R3
Reluctance (SR-VR) resolvers (Fig. 1(d) and (e)). There are respectively. 3-D time stepping finite element method (TSFEM)
many works on VAGL-VR resolvers [6]–[9]. In [6]–[8], some is used for performance evaluation of the studied resolvers and
improved winding configurations have been proposed for this finally, experimental tests on the mentioned resolvers are con-
type of resolvers. The conventional windings of VAGL-VR re- ducted to validate the results of FEA. Applying all the discussed
solvers are variable turns, on-tooth, overlapping windings that challenges in the numerical simulation of the sensors, leads to
are replaced by a constant turns, on-tooth, overlapping windings the closest agreement between the experimental and the simula-
[7], [8], or constant turns, non-overlapping windings [6], [9]. In tion results. All simulations are carried out with Ansys-Maxwell
[9] an optimization is done on the rotor contour of the resolver 17.2.
to achieve the most accurate resolver. One of the challenges in
using VAGL-VR resolvers is their low accuracy in 2-pole config- II. THE ACCURACY FACTORS
uration that makes them unsuitable for absolute position sensing
Using 3-D time stepping finite element method (TSFEM)
[7]. However, SR-VR resolvers are mostly built in 2-pole con-
the complexity of the machine’s geometry and the exact ma-
figuration with non-overlapping winding [10]–[12]. Although
terial’s property can be modeled. Therefore, the most accurate
SR-VR resolvers are more reliable against eccentricities, their
method for performance evaluation of electrical machines and
accuracy is strongly affected by run-out error (the situations
electromagnetic sensors is 3-D TSFEM. However, in the case
when the rotor or stator is displaced along z-axis). Therefore,
of resolvers as position sensors there are some suggestions that
in [12] a new disk type configuration is proposed for SR-VR
must be considered in FEA to obtain reliable results. The most
resolvers that are robust against run-out fault. The other types of
important factors that can influence the accuracy of resolvers’
resolvers are WR resolvers that can be built in cylindrical, disk
simulation results are mesh quality, step time of simulation,
type or linear structures. Recently, WR disk type resolvers gain
simulation’s stop time, excitation method, and output signal
more attention due to their distinguished performance against
processing methods. It should be mentioned that the three first
static eccentricity and run-out faults [2]–[5]. In [5] examining a
factors are common in simulation of resolvers and any other
short circuit coil on the rotor of the resolver is proposed to in-
electrical machines. However, the other factors are dedicated to
crease the accuracy of the studied resolver under inclined rotor
resolvers. In continue, the effect of different factors is discussed
fault. In [2] a new analytical model based on winding function
in details.
method (WFM) is proposed for a disk type WR-resolver. Then,
effect of geometrical dimensions, winding pole numbers and
winding configuration have been studied [2]. In [4] a new con- A. The Mesh Quality and the Value of Step Time
figuration is proposed for the disk type, WR-resolvers that leads It is clear that using finer meshes and smaller time steps leads
to omit the primary and secondary cores of rotary transformer to more accurate results until numerical instability is reached
(RT). Therefore, extra empty spaces are added to the inner parts in the price of high computational time and storage memory.
of the stator and rotor cores that makes these type of resolvers Therefore, the question that arises is “what is the optimal mesh
suitable for position estimation of high torque machines with and time step size?”
large shaft diameter. The best mesh quality in WR resolvers is achieved when the
Despite all the mentioned researches on different type of re- sensor is meshed in “Magnetostatic-Mode” solver and the ap-
solvers that use finite element method (FEM) for design, perfor- plied mesh are imported to the “Transient-Mode” solver. How-
mance prediction or design evaluation of resolvers, there is no ever, for VR resolvers it is not as accurate as enough. In the
reported investigation on the challenges of FEA of these electro- Magnetostatic solver the meshes are only refined in one rotor
magnetic sensors. To evaluate resolvers’ performance by a FEM position due to absence of movement. While, in Transient solver
software, the size of time steps plays a major role. Generally the rotor is free to rotate and the meh operation must be refined
the smaller step time, the higher accuracy and the higher sim- during the rotation due to variation of the air-gap reluctance.
ulation time. Therefore, to acquire reliable and accurate results However, considering semi-closed slots for the stator and the
in a fast way, analysis of mesh quality as well as step time size rotor of WR resolvers, the air-gap permeance is relatively con-
is necessary. Accordingly, in this paper, by the aid of numerical stant. It means the optimal meshes for the VR resolver must be
experimentation the effect of time step size and mesh quality estimated in the Transient solver using try and error method and
on the performance evaluation of resolvers are studied. More- the imported mesh from Magnetostatic solver is not reliable. It
over, excitation methods, that are voltage and current types, are is worth noting that, in static solvers, mesh is refined to reach the
analyzed and proper solutions to gain reliable results are sug- intended accuracy, while transient solvers do not have this ca-
gested in following. Besides, Hilbert function is introduced to pability. Consequently, the meshes resulted from magnetostatic
find the envelope of resolver’s output signals. It is shown that solver is more accurate that the transient ones.
THDs of detected envelopes are not necessarily a good indica- To explain the importance of appropriate choosing time steps,
tor of estimated position error. Therefore, the primary idea of a simple problem is defined. As shown in Fig. 2, a simple cop-
this paper is to study the main challenges on numerical calcu- per coil is considered for simulation. The Magnetostatic-Mode
lations of the resolvers. Analytical discussions as well as FEM simulation is carried out for the coil and the best mesh quality
results are presented in the following sections. Four different is chosen for the Transient-Mode simulation. The outer and the
SANEIE et al.: CHALLENGES OF FINITE ELEMENT ANALYSIS OF RESOLVERS 975

Fig. 2. The distribution of magnetic flux density on the proposed copper coil
for studying the effect of time steps. Fig. 3. The current of the studied coil: (a) analytically calculated current,
(b) the result of TSFEM, (c) the results of BEM with different time steps, and
(d) the result of RKM.

inner radius of the coil are chosen 24 and 20 mm, respectively

and the height of the coil is 4 mm. The calculated inductance netic sensors in TSFEM. Therefore, the next time integration
and the resistance of the coil are 68.795 mH and 148.95 μΩ, method -RKM- is employed for calculating the coil’s current:
respectively. Considering zero initial current for the coil and a
sinusoidal voltage as an excitation: vs1 + vs2 Δt L − R Δ2t
i2 = Δ t
+ i1 (5)
2 L+R 2 L + R Δ2t
vs (t) = Vm sin (ωt + ϕ) = 5cos (2π × 5000t) (1)
The current of the coil can be calculated as: As shown in Fig. 3-d, considering 12.5 μs as step size (16
   samples in each period), using RKM the results are in close
Vm ωL
i (t) = Ae− L t +  sin ωt + ϕ −tan−1
R
agreement.
R
(ωL)2 + R2 Therefore, it seems using RKM, even with large step times
(2) the results are reliable. However, RKM is not applicable in the
where 3-D analysis of resolvers (the option of RKM is not available in
    3-D analysis of Ansoft/Maxwell software) and the only method
Vm ωL
A=  sin tan−1 −ϕ (3) to achieve the accurate results is reducing the step size. In the
R
(ωL)2 + R2 analysis of electric machines that only a small number of periods
are needed and the steady state condition is achieved after a few
The calculated current is shown in Fig. 3-a. The mentioned periods, reducing the step size is applicable. While, in FEA
coil is also simulated using TSFEM. Considering the time step of resolvers that hundreds of periods of excitation signal are
of the Transient-Mode simulations to be equal to 12.5 μs (about required for performance evaluation such reduction in the step
6% of excitation period), the results of FEM is shown in Fig. 3-b. size is not possible. So, in the next section the methods to
Comparing the analytically calculated current and that of TS- eliminate the effect of time steps in the BEM are discussed
FEM shows a great difference. The reasons of the difference based on the excitation method.
between two results are related to the size of time steps in nu-
merical calculations of TSFEM. There are two commonly used
time integration methods in numerical calculations: Backward B. Excitation method
Euler Method (BEM), and second-order Runge-Kutta Method In the previously discussed parts the voltage source is used
(RKM). In the first method, the current and the voltage of each for excitation of coil. Because the value of resistance was much
step is used to calculate the values for the next step: lower than Lω, assuming the cosine voltage source there is
vs2 Δt + Li1 negligible transient response for the excitation current. However,
i2 = (4) for a resolver with a large time constant the DC flux linkage
L + RΔt
component may take a very long time to decay. Therefore, it
As shown in Fig. 3-c using this method for calculating the an- seems using a current source as an excitation the series resistance
alytical current, the obtained results from numerical calculation has no effect on the transient of the simulation results and the
is in closer agreement with the analytical current with respect results of each time step is independent from the previous step.
to the result of TSFEM. However, there is also a difference be- Considering the current source as:
tween two sets of the results. To reduce the difference between
results, the effect of reducing step size is examined, the results is (t) = Im sin (ωt) (6)
are shown in Fig. 3-c. It can be seen that using 1 μs as the time
step or using 200 samples in each period, the results of two The induced voltage is calculated analytically as:
methods are in the closest agreement. But, such reducing step
size is not practical for performance prediction of electromag- v (t) = ωLIm cos (ωt) (7)
976 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 34, NO. 2, JUNE 2019

Fig. 5. Frequency shifting method.

Fig. 4. The effect of using current source as an excitation using BEM for
numerical calculation: (a) the initial results, and (b) the results after phase-
As it can be seen from (16)–(18), it can be concluded that
shifting. by using current source for excitation winding only the space
harmonics can be seen in the output characteristic of the re-
And using the BEM the induced voltage can be obtained as: solver. While, using voltage source the current of excitation
winding is dependent to the inductance of the excitation wind-
is2 − is1
L = v2 (8) ing. Therefore, beside the space harmonics, time harmonics are
Δt also effective in the output characteristic when voltage source
Both voltages of (7)–(8) are shown in Fig. 4-a. Comparing is taken into account.
the results of analytical method and that of BEM, using current Finally, it can be concluded using current source for the ex-
source shows there is a phase shift equal to half of the step size citation of the resolver with large time steps leads to reliable
in the output voltage. So, as shown in Fig. 4-b, after a phase-shift results only if the harmonics of the self-inductance be very
the results are in close agreement. Finally, it seems using current small. If the harmonics of the self-inductance are not negligible,
source as an excitation is much suitable for performance pre- the position error will be considerable. Furthermore, using volt-
diction. However, it must be clarified that what is the influence age source as the excitation of resolver with large time steps,
of current/voltage source excitation on the output characteristic results in inaccurate prediction of excitation current. However,
of a resolver. To do that, the flux linkage-current equation of the resolver’s output voltages are in close agreement with the
resolver is considered as: real voltages. To explain that, it should be considered that as
⎡ ⎤ ⎡ ⎤⎡ ⎤ shown in Fig. 3-c, the difference between the calculated current
λe Le (θ) Mes (θ) Mec (θ) ie
⎣ λs ⎦ = ⎣ Mse (θ) Ls (θ) Msc (θ) ⎦ ⎣ is ⎦ (9) using BEM with large time steps and the real current is related
λc Mce (θ) Mcs (θ) Lc (θ) ic to a DC component. In performance prediction of the resolver
due to absence of saturation and linear behavior of the magnetic
where λγ , iγ , Lγ , and Mγ α are flux linkage, current, self- circuit, the DC component in the current and consequently in
inductance and the mutual inductance between two windings, the flux has no effect in the output induced voltages. Therefore,
respectively. The subscripts γ = e, s, and c stand for the exci- using voltage source as the excitation of the resolver even with
tation, sine, and cosine windings. In practical applications the large time steps leads to accurate outputs.
output of resolver is connected to a high input resistance resolver
to digital converter (RDC).
Therefore, the nominal current of signal windings is almost C. Envelope Detection Method
zero [2]–[4]. Then, (9) can be rewritten as: The induced voltages in the signal windings of the resolver
are amplitude modulated voltages. So, at first step envelope of
λe = Le (θ) ie (10)
these voltages should be calculated. Then arctangent of envelope
λs = Mse (θ) ie (11) ratio leads to calculate the position and also the total harmonic
distortion (THD) of envelopes is a factor for judgment about
λc = Mce (θ) ie (12)
the resolver accuracy. To calculate the THD of output voltages’
where envelope and retrieve the position of the rotor, two steps should
 be taken. First, the excitation signals should be separated from
Le (θ) = Leo + len cos (nθ − ϕn ) (13) the output signals. This process is called demodulation or en-
 velope detection. The second step is to calculate the THD of
Mse (θ) = Mse0 sinP θ + msn sin (nθ − ϕsn ) (14)
the induced voltages’ envelope and to estimate the angle of the
 rotor [13]. Two common methods of extracting the envelope of
Mce (θ) = Mce0 cosP θ + mcn cos (nθ − ϕcn ) (15)
output signals are frequency shifting and peak detection [14].
where P is number of rotor pole pairs and for VAGL-VR re- In frequency shifting method that is widely used in communi-
solvers it is equal to number of rotor saliencies. Assuming cation systems, output signals are multiplied by the excitation
Mse0 = Mce0 = M and msn = mcn = mn the flux linkages signal and passed through a low pass filter. In this method that
can be rewritten as: is briefly shown in Fig. 5, presence of synchronized excita-
 tion signal is necessary [14]. Resistance of the rotor winding
λe = Leo + len cos (nP θ − ϕn ) ie (16) leads to phase shift error. Detection, measurement and com-
 pensation of the phase shift error make it difficult to achieve
λs = M sinP θ + mn sin (nP θ − ϕsn ) ie (17) synchronized signal. In peak detection method, the output sig-
 nal is sampled when the modulated high frequency excitation
λc = M cosP θ + mn cos (nP θ − ϕcn ) ie (18)
signal reaches into its maximum point. Therefore, neglecting the
SANEIE et al.: CHALLENGES OF FINITE ELEMENT ANALYSIS OF RESOLVERS 977

error in one mechanical cycle of the sensor, especially when

on-tooth winding is used for the signal windings. In fact, taking
one mechanical cycle into account the effect of sub-harmonics
will be considered in the performance evaluation while in one
electrical period of the signal they are ignored. So, the stop time
of simulations should be set to obtain one mechanical period of
envelopes.

E. The Accuracy Index

A well-known index to evaluate the accuracy of detected po-
sition is the THD of induced voltages’ envelopes. The lower
THDs, the more accurate estimated position is. In no-error con-
ditions when the output coils contain identical mesh quality,
the output signals’ THDs should be equal. In practice when the
Transient-mode solution is used, achieving identical mesh for
different coils is a field of trouble. In Transient-mode, in each
time-step the meshes are revised and changed by the software.
Even though the same initial meshes, the final meshes may
Fig. 6. Hilbert transfer function.
be different. Different mesh schemes lead to different results.
Therefore, the THD of SIN and COS windings are not neces-
transformation ratio, the output signals are just the SIN and COS sarily equal, which results in imprecise estimated position.
of the rotor position. This approach enjoys several advantages Beside the finite element software deficiencies, THD has
over frequency shifting method. First, at peak instances the sig- some limitations in describing estimated position error; so that
nal to noise ratio is at its maximum value, so the output signals signals with equal THDs will result in different position errors.
are less sensitive to the noise. Second, the speed voltages as a These limitations are listed as follows:
disturbance to the precise position detection are in their mini- 1) THD, as an indicator of harmonic contents of a signal,
mum value at these instances [14]: is insensitive to the phase and order of harmonics. This
ωm ω e t=k π leads to equal THDs for different harmonic contents. Eq.
Vsin 1 = V1 sin (θ1 ) cos (ωe t) + V1 cos (θ1 ) sin (ωe t) → (21) shows the estimated position resulted from output
   ωe
tr an sf or m er v oltag e
   signals:
speed v oltag e  
Vsin
Vsin 1 = V1 sin (θ1 ) (19) θ = arctang (21)
Vcos
ωm ω e t=k π
Vcos1 = V1 cos (θ1 ) cos (ωe t) − V1 sin (θ1 ) sin (ωe t) → where:
   ωe ∞
tr an sf or m er v oltag e
   
speed v oltag e Vsin = Vh sin (θh ) (22)
h=1
Vcos1 = V1 cos (θ1 ) (20)
∞

Several proposals are suggested in literatures to implement Vcos = Vh cos (θh ) (23)
the peak detection method [14]–[16]. Catching the right instance h=1
of peak times, especially when the noise and disturbance exist, Therefore:
is a field of concern. Moreover, when the excitation frequency  ∞
1 
is low the quality of estimated envelope will fall down. To θ = arctang Vh sin (θh )
overcome these problems, Hilbert transformation is used. By 2
h=1
the aid of Hilbert transformer, the envelope detection process  
∞ 2  ∞ 
will be insensitive to the noise and errors that may exist in the + Vh sin (θh ) +4 Vh cos (θh )−1
h=1 h=1
excitation signal. Fig. 6 shows the envelope detection method
by Hilbert transformation.  ∞
The method based on Hilbert transformer, eliminates the mod- 1 
+ arctang Vh sin (θh )
ulated excitation signal from output voltages regardless of the 2
h=1
phase shift that may occur due to the rotor resistance.  
∞ 2  ∞ 
− Vh sin (θh ) +4 Vh cos (θh ) − 1
h=1 h=1
D. Stop Time
Although in 2-pole resolvers one electrical period of the out- (24)
put signals is equal to one mechanical period, for multi-pole re- Eq. (24) shows that if two pair of signals share similar THDs
solvers they are different from each other. The important point and first harmonics but different higher order ones; the resultant
in performance analysis of the resolver is to see the position estimated position would not be equal necessarily. Fig. 7 shows
978 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 34, NO. 2, JUNE 2019

TABLE I
THE PARAMETERS OF STUDIED RESOLVERS

Fig. 7. Estimated position error for different harmonic contents, but same THD
= 2.2361% C ase1 : sin(ωt) + 0.02sin(3ωt) + 0.01sin(9ωt), C ase2 :
sin(ωt) + 0.02sin (5ωt) + 0.01sin (9ωt), C ase3 : sin (ωt) + 0.02sin
(3ωt + π/5) + 0.01sin(9ωt + 3π/5).

for high precision estimations. Instead, evaluation of position

error by the aid of arc tangent method is more realistic.
estimated position error for three different cases. In all cases the
profiles of estimated position error have changed with respect III. FINITE ELEMENT SIMULATIONS
to the phase and order of harmonics.
2) Quadrature error is one of prevalent errors in resolvers. To evaluate the points that are mentioned for finite element
It refers to a situation where the output signals are not simulation of the resolvers, four different resolvers are selected:
exactly perpendicular [17]: two VAGL-resolver (R1, R4), one SR-resolver (R2), and one
WR-resolver (R3). R1 and R2 have the same 2-pole stator and
Vsin 1 = V1 sin (θ1 ) cos (ωe t) (25) the only difference between two resolvers is related to the struc-
Vcos1 = V1 cos (θ1 + β) cos (ωe t) (26) ture of the rotor. But, R3 is a disk type wound rotor resolver
with 6-pole winding. Finally, R4 is a 10-pole variable reluc-
where β shows the amount of imperfect quadrature. The tance resolver. Despite, R1, R2, and R3 that need 3-D finite
output signals’ THDs are independent of β. Therefore, element simulation, R4 can be evaluated using 2-D simulations.
THD will wrongly predict higher accuracy than what is The geometrical dimensions and the excitation information of
represented by the actual position error. the resolvers are presented in Table I. To clarify the effect of
3) Investigations show that estimated position error is not a presented accuracy factors on the performance of the resolvers,
monotonic function of output signals’ THDs [2]. In WR some simulations are carried out on R4. The effect of excita-
resolvers when the slot numbers are fixed, as the num- tion methods on maximum position error, and THD of induced
ber of pole pairs increases, the THDs of output signals voltages’ envelopes, are shown in Fig. 8 for R4 with different
increase too. It is mainly due to the lower slot/pole/phase number of stator teeth. It can be seen for all simulations the value
in this circumstance. While, the estimated position er- of THD considering voltage source is higher than that of cur-
ror is inversely related to the pole pairs. Therefore, the rent source. It means considering voltage source as excitation,
slot/pole/phase and the pole pairs differently affect the the harmonics due to self-inductance of excitation windings are
accuracy of estimated position. In other words, estimated significant. While, the value of maximum position error using
position error does not follow the THD necessarily. two mentioned excitation method does not show a predictable
To sum up, however, the THD of output signals is a good format. For example, in the case of 16-slot resolver, the esti-
indicator for quality of output signals, but it is not reliable index mated position errors of two excitation methods are almost the
SANEIE et al.: CHALLENGES OF FINITE ELEMENT ANALYSIS OF RESOLVERS 979

Fig. 8. The influence of excitation method on the performance of VAGL

resolver (R4) considering different stator slots: (a) maximum position error, and
(b) THD of induced voltages’ envelope.

Fig. 10. The influence of time step size on the performance of resolver:
(a) THD of envelopes, (b) Maximum position error, (c) The average of absolute
position error and (d) The average of the excitation current.

same. While, the THD of output voltages are significantly dif-

ferent from each other. Furthermore, comparing the results of
Fig. 8-a with those of Fig. 8-b shows that the worst resolver con-
sidering the maximum position error is 14-slots resolver while
based on THD, the highest THD is related to resolver with
18-slot.
To clarify the effect of mesh size on the accuracy of the
simulation, R4 with different mesh setting is simulated. THD of
envelopes, maximum and the average of absolute position error
and the average of the excitation current versus number of mesh
elements are presented in Figs. 9-a through 9-d. It can be seen
that refining the mesh size leads to a constant value for all the
performance characteristics of the resolver.
The next accuracy factor is step time of the numerical calcu-
lation. Assuming the optimal mesh size, as shown in Figs. 10-a
through 10-d, the average of excitation current increases by in-
creasing the time step. However, other parameters are almost
constant.
Finally the effect of stop time is studied for R4. Since R4 is
a 10-pole resolver, 5-electrical cycle of position error leads to
one mechanical cycle of error. The position error of the studied
Fig. 9. The influence of mesh size on the performance of resolver: (a) THD
of envelopes, (b) Maximum position error, (c) The average of absolute position (R4) resolver in one mechanical rotation of rotor is shown in
error and (d) The average of the excitation current. Fig. 10-a. It can be seen the error in each electrical period (72
980 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 34, NO. 2, JUNE 2019

Fig. 12. The schematic of meshes and magnetic flux density on the studied
resolvers: (a) R1, (b) R2 (c) R3, and (d) R4.

Neglecting value of even harmonics in harmonic content of

healthy resolvers, clarifies the proper selection of time step and
mesh size. Fig. 14-b shows the THD of voltages’ envelope for
the studied resolvers. It can be seen the lowest THD is related to
the R4 and the highest THD belongs to the R1. For the sake of
brevity, the position error of the studied resolvers resulted from
FEA is presented in the next section along with experimental
results.

Fig. 11. The influence of stop time step on the performance of resolver: IV. EXPERIMENTAL EVALUATION
(a) Position error in one mechanical rotation of rotor, (b) THD of envelopes,
(c) Maximum position error, and (d) The average of absolute position error. To validate the finite element simulations, prototypes of stud-
ied resolvers are experimentally tested. Fig. 15 shows the stator
mech. Deg.) is not periodic. The variation of THD, average of and the rotor of prototypes.
absolute position error and maximum position error with stop Fig. 16-a, shows the experimental test circuit for the stud-
time of simulation are shown in Figs. 11-b through 11-d. ied resolvers (R1 is shown in the Fig. 16-a). An electric motor
It can be seen due to aperiodic behavior of the induced volt- equipped with an absolute optical encoder is used to rotate the
ages in five electrical cycle of envelopes (one mechanical cycle), sensors. The motor is driven in desired speed using a shielded
the stop time of simulation must be set in such a way that the inverter. The coupling between the resolver and the motor is
performance of the resolver can be evaluated based on the results done using a self-centering three-jaw chuck and a z-axis table
of one mechanical rotation of rotor. to ensure the rotational axis and the symmetrical one of the mo-
Finally, based on the above-mentioned analysis the optimal tor are coincide with those of resolver. A digitally synthesized
setting is considered for finite element simulation of R1, R2, function generator with the resolution of 0.1 Hz is used to sup-
R3, and R4. Fig. 12 shows the schematic of meshes and mag- ply the excitation winding. The amplitude of excitation voltage
netic flux density on the studied resolvers. For R1, R2 and is adjusted using an automatic gain control circuit. When the
R4 the mesh size is refined in the transient simulation and for motor rotates in a constant speed, the output induced voltages
the R3 the mesh of Magnetostatic solver is imported to the of the resolver are saved using a digital oscilloscope, as shown
transient solution. It can be seen the highest magnetic flux in Figs. 16-b for R1. The induced voltages for R2, R3 and R4
density is less than 60 mT. Therefore, there is no concern are shown in Figs. 16-c, 16-d and 16-e, respectively. Calculating
about the influence of core saturation on the accuracy of de- position error of the studied resolvers shows the most accurate
tected position. Considering the optimal mesh and time step resolvers are R3 and R4 while R3 has the highest value of THD.
size with voltage source as excitation, the simulations are car- It means as mentioned earlier THD of induced voltages’ enve-
ried out for more than one mechanical rotation of the rotor. The lope is not a good indicator for the accuracy of resolver. The
flowchart of finite element simulation of resolvers is presented average of absolute position error and the maximum position er-
in Fig. 13. ror of the studied resolvers are presented in Figs. 17-a, and 17-b,
Harmonic content of induced voltages’ envelope for R1, R2, respectively. It can be seen due to considering the points that
R3 and R4 are shown in Fig. 14-a. it can be seen that the third are mentioned in section II and Fig. 13, there is close agree-
harmonic is the dominant harmonic for all studied resolvers. ment (less than 6%) between the results of TSFEM and those of
SANEIE et al.: CHALLENGES OF FINITE ELEMENT ANALYSIS OF RESOLVERS 981

Fig. 13. The flowchart of finite element simulation of resolvers.

982 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 34, NO. 2, JUNE 2019

Fig. 14. The results of 3-D TSFEM for the studied resolvers: (a) harmonic
content of induced voltages’ envelope, and (b) THD of induced voltages’ enve-
lope.

Fig. 16. The experimental Results: (a) the test circuit, (b) The induced voltages
for R1, (c) the induced voltages for R2 (d) the induced voltages for R3 and
(e) the induced voltages for R4.

Fig. 15. The stator and the rotor of three prototypes: (a) R1, (b) R2 (c) R3, Fig. 17. Position error of the studied resolvers: (a) the average of absolute
and (d) R4. position error, (b) the maximum position error.
SANEIE et al.: CHALLENGES OF FINITE ELEMENT ANALYSIS OF RESOLVERS 983

experimental tests that clarifies the success of the finite element [5] Z. Nasiri-Gheidari and F. Tootoonchian, “Axial flux resolver design tech-
simulations. niques for minimizing position error due to static eccentricities,” IEEE
Sensors J., vol. 15, no. 7, pp. 4027–4034, Jul. 2015.
[6] X. Ge, Z. Q. Zhu, R. Ren, and J. T. Chen, “A novel variable reluctance
V. CONCLUSION resolver for HEV/EV applications,” IEEE Trans. Ind. Appl., vol. 52, no. 4,
pp. 2872–2880, Jul./Aug. 2016.
In this study the challenges and the intricacies of numerical [7] X. Ge, Z. Q. Zhu, R. Ren, and J. T. Chen, “A novel variable reluctance
calculations for performance prediction of resolvers were dis- resolver with nonoverlapping tooth–coil windings,” IEEE Trans. Energy
Convers., vol. 30, no. 2, pp. 784–794, Jun. 2015.
cussed. It was clarified that the common numerical method is [8] X. Ge, Z. Q. Zhu, R. Ren, and J. T. Chen, “Analysis of windings in variable
time stepping finite element method that its results are very sen- reluctance resolver,” IEEE Trans. Magn., vol. 51, no. 5, pp. 1–10, May
sitive to proper selection of time steps, size of mesh elements, 2015.
[9] X. Ge and Z. Q. Zhu, “A novel design of rotor contour for variable
and simulation’s stop time. Furthermore, the influence of sup- reluctance resolver by injecting auxiliary air-gap permeance harmonics,”
plying the excitation winding with voltage/current source was IEEE Trans. Energy Convers., vol. 31, no. 1, pp. 345–353, Mar. 2016.
discussed. [10] J. Shang, H. Wang, M. Chen, N. Cong, Y. Li, and C. Liu, “The effects
of stator and rotor eccentricities on measurement accuracy of axial flux
It was shown the most reliable supply for the excitation wind- variable-reluctance resolver with sinusoidal rotor,” in Proc. Int. Conf.
ing is voltage source. Then, two common methods for envelope Elect. Mach. Syst., Hangzhou, China, Oct. 22–25, 2014, pp. 1206–1209.
detection were discussed and a new method based on the Hilbert [11] S. Jing, W. Hao, and W. Weiqiang, “The analysis of multi pole axial flux
reluctance resolver with sinusoidal rotor,” in Proc. Int. Power Electron.
transformation was introduced. Finally, the validity of THD as Motion Control Conf., Harbin, China, 2012, pp. 1206–1209.
accuracy index was called into question. It was clarified that [12] Z. Nasiri-Gheidari and F. Tootoonchian, and F. Zare, “Design oriented
the only reliable accuracy index for resolvers is position error. technique for mitigating position error due to shaft run-out in sinusoidal-
rotor variable reluctance resolvers,” IET Elect. Power Appl., vol. 11, no. 1,
To verify the discussed points, four different resolvers included pp. 132–141, 2017.
two VAGL-VR, one SR-VR, and one WR-resolver were sim- [13] O. A. Tolstykh, A. P. Balkovoi, M. G. Tiapkin, and A. S. Markov, “Re-
ulated using finite element method and experimentally tested. search and development of the 4X-variable reluctance resolver,” in Proc.
IEEE NW Russia Young Res. Electrical Electron. Eng. Conf., St. Peters-
Taking the points that were discussed in the paper into account, burg, Russia, 2016, pp. 704–709.
good agreements were obtained between the simulation and the [14] N. L. H. Aung, C. Bi, A. Al Mamun, C. S. Soh, and Y. YinQuan, “A
experimental results. demodulation technique for spindle rotor position detection with resolver,”
IEEE Trans. Magn., vol. 49, no. 6, pp. 2614–2619, Jun. 2013.
[15] D. A. Khaburi, “Software-based resolver-to-digital converter for DSP-
REFERENCES based drives using an improved angle-tracking observer,” IEEE Trans.
Instrum. Meas., vol. 61, no. 4, pp. 922–929, Apr. 2012.
[1] R. Alipour-Sarabi, Z. Nasiri-Gheidari, F. Tootoonchian, and H. Oraee, [16] S. Sarma, V. K. Agrawal, and S. Udupa, “Software-based resolver-to-
“Performance analysis of concentrated wound-rotor resolver for its appli- digital conversion using a DSP,” IEEE Trans. Ind. Electron., vol. 55,
cations in high pole number permanent magnet motors,” IEEE Sensors J., no. 1, pp. 371–379, Jan. 2008.
vol. 17, no. 23, pp. 7877–7885, Dec. 1, 2017. [17] D. C. Hanselman, “Resolver signal requirements for high accuracy
[2] R. Alipour-Sarabi, Z. Nasiri-Gheidari, F. Tootoonchian, and H. Oraee, resolver-to-digital conversion,” IEEE Trans. Ind. Electron., vol. 37, no. 6,
“Effects of physical parameters on the accuracy of axial flux resolvers,” pp. 556–561, Dec 1990.
IEEE Trans. Magn., vol. 53, no. 4, pp. 1–11, Apr. 2017.
[3] F. Tootoonchian, “Design, performance, and testing of a brushless axial
flux resolver without rotor windings,” IEEE Sensors J., vol. 16, no. 20,
pp. 7464–7471, Oct. 15, 2016.
[4] Z. Nasiri-Gheidari, “Design, analysis, and prototyping of a new wound-
rotor axial flux brushless resolver,” IEEE Trans. Energy Convers., vol. 32,
no. 1, pp. 276–283, Mar. 2017. Authors’ photographs and biographies not available at the time of publication.