Sei sulla pagina 1di 6

Dynamic Modeling of Losses in Electrical Machines

for Active Loss Control


Francesco Quattrone

Robert D. Lorenz

Student Member, IEEE


Leibniz Universitt Hannover
Institute for Drive Systems and Power Electronics
Hannover, Germany
francesco.quattrone@ial.uni-hannover.de

Fellow, IEEE
Univ. of Wisconsin-Madison
WEMPEC
Madison, WI 53706 USA
lorenz@engr.wisc.edu
such dynamic active loss control strategies. Firstly, when
dynamically controlling physical systems no future
magnetic states are known. Therefore, the area of hysteresis
loops and the trajectory of minor (inner) loops in the B-Hplane in particular, cannot be identified in advance. In
contrast, loss calculation methods used on finite element
software are dependent on that information [2-3]. Secondly,
loss calculation has to be performed within one sample
instant (one switching period), typically around 100s.
Numerical simulations can only be performed offline, but
not in real time. As a direct consequence of this, developing
an analytical loss model using input data available in the
controller, i.e. torque and flux linkage, is an inevitable
requirement for active loss control.
In this paper, a new approach of a dynamic core loss
model suitable for active loss control in DB-DTFC is
presented. This approach consists of two consecutive
working steps: First, numerical calculations are performed
in order to characterize core loss of the investigated machine
as a function of flux density. Secondly, this information is
used to calculate core loss as a function of stator flux
linkage analytically.

Abstract This paper presents a dynamic core loss model for


active loss control of permanent magnet synchronous
machines. Deadbeat-direct torque and flux control
(DBDTFC) enables direct control of torque and stator flux
linkage over each switching period. Total machine loss can be
minimized by selecting the optimal flux level in every sampled
time instant. Reliable models for dynamic copper and core loss
are an essential requirement for this optimizing task. This
paper presents an analytical dynamic core loss model in
switching period time-domain, using stator flux linkage as the
manipulated input variable. Combining numerical and
analytical calculation methods is used to create an easy and
practical approach to estimate core loss components based on
stator flux linkage without running FEA calculations for every
time instant. Additional effects like high frequency eddy
current loss in electrically conducting permanent magnets as
well as spatial harmonic content are taken into account.

I.

INTRODUCTION

Energy efficiency is an increasingly important field of


research. Particularly where available energy is limited, for
example in hybrid or fully electric vehicles, modern control
techniques like deadbeat-direct torque and flux control (DBDTFC) [1] have proven beneficial. Controlling machine flux
linkage on switching period level enables to minimize total
machine loss dynamically. Increased efficiency not only
improves mileage of electric vehicles, it also reduces the
thermal load of the electric drive, enabling higher power
ratings and compact machine designs. Reliable dynamic loss
models are the basis for loss minimizing control. However,
dynamic core loss modeling has hardly been investigated.
This paper focuses on replacing the classical
approaches to estimate core loss in permanent magnet
synchronous machines based on frequency by
comprehensive analytical, instantaneous loss models.
Calculating hysteresis and eddy current loss in the switching
period time domain could enable total loss minimization
under transient conditions. Transient core loss models have
to meet two main requirements in order to be applicable to

978-1-4673-0803-8/12/$31.00 2012 IEEE

II. ACTIVE LOSS CONTROL


DB-DTFC is a very promising control technique which
achieves commanded torque and flux changes within one
switching period. As shown in [1], there is an unlimited
r
number of Volt-second vectors vdqs(k) Ts which yield
deadbeat torque and flux change. The complex relationships
*
of reaching commanded torque Tem(k+1) in the next sample
instant k+1 can be visualized very easily in the volt-second*
plane. Figure 1 shows that commanded torque Tem(k+1) can
be represented by a linear function according to (1), called
torque line. Slope m and the y-intercept b are functions of
the current flux linkage, the desired torque change T and
speed.

47

(a)

Figure 1. Voltage hexagon and stator flux linkage vector dqs(k) in


volt-second-plane. Possible volt-second vector range vdqs(k)Ts and the
torque line Tem(k+1) are shown as well.
r

vqs(k) Ts = m vds(k)Ts + b

Figure 2. (a) Hysteresis curve and (b) equivalent elliptical loop


(EEL). Bsat is the induction where all magnetic domains are aligned.
Him specifies the width of the EEL and meets the requirement that the
area enclosed of both hysteresis curve and EEL are identical for a
sinusoidal flux waveform [8].

(1)

In a transient state no frequency can be identified, so


that loss equations in time domain have to be used instead.
There are only a few publications focusing on this topic,
such as [8]. The basic idea is to replace flux density and
frequency in (2) and (3) by the time derivative of flux
density. Using the time derivative is consistent to the
physical phenomenon, since core loss is caused by the
change of magnetic state.
Modeling hysteresis loss is far more complex than
eddy current loss, since hysteresis effects are dependent on
previous magnetic states. According to [8], the magnetic
field strength can be separated into a reversible component
Hr and an irreversible component Hi :

Every Volt-second vector within the Volt-second hexagon


which ends at the torque line can be chosen. The ability of
controlling flux can be used as an additional degree of
freedom. One possible solution is to choose that vector
which minimizes total machine loss in every sampled time
instant, i.e. minimizing loss dynamically on switching
period level. This enables a considerable potential of energy
savings. Both winding and core loss have to be taken into
account in order to find the loss-minimal flux level of the
machine. While transient winding loss (i2R loss) can easily
be described as a function of transient current, an
appropriate transient core loss model has to be found. In
addition, high frequency eddy current losses in electrically
conducive permanent magnets must often be included. In
more extreme cases, even the high frequency skin-effect
resistance of the motor winding may have to be included.
That second order issue is not discussed in this paper.

H = H r + Hi .

In the following, core loss is modeled as the


summation of hysteresis loss Ph in ferromagnetic materials
and classical eddy current loss Pc in both ferromagnetic and
permanent magnet materials. Core loss density is described
by the symbol (). In order to find an appropriate transient
model for core loss, classical frequency-based approaches of
Steinmetz and its extensions [2-7], considering elliptical
magnetization, minor hysteresis loops, and harmonics,
provide a basis for further investigation. Both loss
components are dependent on flux density B and frequency
f, as shown in (2) and (3). The loss coefficients kh and kc
represent the material properties and have to be measured,
considering the dependency on saturation and frequency [4].
2

(2)

2 2

(3)

pc'(f) = kc B f

(4)

Only the irreversible field strength is related to the


hysteresis effect and to the corresponding loss ph, while
reversible field strength is only related to reactive power.
Consequently, in [8], the real hysteresis curve is replaced by
an equivalent elliptic loop (EEL), as shown in Fig. 2. The
area of both hysteresis loop and EEL has to be identical, so
that the time-averaged loss (5) in one fundamental period T
is consistent with the frequency based steady-state loss
equation:
T
1 dB
2
(5)
dt = kh Bm f .
H
ph' =
T i dt
0
The ELL in the B-Hi-plane is described as follows:

III. TRANSIENT CORE LOSS MODELING USING FINITE


ELEMENT SOFTWARE

ph'(f) = kh B f

(b)

B = Bm sin() ,

Hi = Him cos().

(6)

The coefficient Bm determines the height of the elliptic loop


and is dependent on previous magnetization states. Here, its
magnitude is determined by the maximal flux density within
the last fundamental half-cycle. It represents the magnetic
record. Taking the condition (5) into consideration, the

48

instant k-1

coefficient Him can be derived, after substituting (6) in (5)


for a sinusoidal excitation and frequency f, as shown in (7).
2

ph' = kh Bm f
T
1
=
H cos(2f t) Bm 2f cos(2f t) dt
T im
0
T
2
1

= Him Bm 2f cos (2f t) dt = Him Bm f (7)


T
0
From this it follows
1
(8)
Him = kh Bm.

Therefore, the irreversible field strength Hi is dependent on


flux density, previous magnetic states, and material
properties. The resulting equation for instantaneous
hysteresis loss with the ellipse parameter according to Fig.
2 can be written as follows [8]:
dB 1
dB
(9)
= k B cos()
.
ph'(t) = Hi
dt h m
dt

instant k

Instantaneous eddy current loss pc' is dependent on the


square of flux density time derivative. It also has to meet
the requirement that time-averaged loss is consistent to the
frequency based steady-state loss equation (3). In (10) and
(11) the resulting basic equations for the instantaneous core
loss model are stated, which clearly show the similarity to
the steady-state models (2) and (3). Considering a periodic
flux waveform with time harmonics including local
minimums (minor hysteresis loops), the results of the
classical steady-state loss calculation method and the timeaveraged results of the new model match very well.
p'h(t) = Hi(t)

p'c(t) =

1
2

dB
dt
dB
dt

2 kc

[kW/m]
500

0
instant k+1

(10)
2

(11)

In order to calculate instantaneous machine loss, a


transient field solution of finite element simulation is
required. It provides the spatial distribution of flux density.
Its time derivative is replaced by the average rate of change
calculated by the difference of flux density between two
samples divided by the sampling time Ts in the discrete time
domain. Both radial (r) and peripheral () components are
calculated to account for rotating magnetization [5],
according to (12) and (13).
dBr Br(k+1) - Br(k)

= Br(k+1)
dt
Ts

(12)

dB B(k+1) - B(k)

= B(k+1)
dt
Ts

(13)

Figure 3. Spatial distribution of core loss for different time instants


whereby between the instants k-1 and k the d-axis flux is changed.

Figure 3 illustrates the resulting instantaneous loss


distribution in an interior permanent magnet machine, after
implementing the presented loss equations in transient
Finite-Element-Method (FEM) software. Every change of
the flux level (here between the instants k-1 and k) results in
a temporarily dramatic increase of core loss (evaluated at
instant k), which is consistent to the physical phenomenon.
While for steady-state operation at instant k-1 only stator
loss due to the rotating field and rotor surface loss due to
spatial harmonics of air gap induction can be identified, at
instant k the d-axis flux change causes a dramatic increase
of stator loss, rotor loss, and magnet loss in particular.

49

Capturing transient effects is a basic requirement for


dynamic loss control. Steady-state loss models do not
provide this information in the frequency domain.
The presented FEM transient loss model inherently
captures effects of permanent magnet and spatial harmonic
loss. Since all loss equations are based on the time
derivative of flux density without the need of separating the
frequency components (Fourier analysis), also time
harmonic loss is captured. In FEM, flux change and the
corresponding loss is evaluated in every finite element
separately. The origin of flux change, due to time harmonics
or spatial harmonics, is not relevant for the calculation.
As shown in section IV, evaluating the loss
components depending on their origin is an important
requirement for the analytical core loss model.

p'c(t) =

| |

ds ds
r
=
- el qs
dt
dt

(16)

qs qs
r
=
+ el ds
dt
dt

The numerical core loss calculation method presented


in section III cannot be directly applied to an active loss
controller, since spatial distribution of flux density is
unknown. In this section, an analytical model is proposed
using stator flux linkage as input variable. Numerical
results (offline calculations) are used to derive machine
characteristics. It has to be noted here that deriving
analytical equations always involves simplifications. The
purpose of the analytical model is not to accurately capture
every loss component in the machine (that is hardly
possible), instead its purpose is to be a practical approach
with sufficient accuracy for active loss control.
The new analytical model differentiates between five
loss components: Hysteresis loss in the stator and rotor,
classical eddy current loss in the stator and rotor, as well as
magnet loss.
In contrast to the FEM-based approach in the last
section, the origin of loss has to be identified in the
analytical model. This includes the difference between loss
caused by d- and q-axis flux linkage, the effect of permanent
magnets on loss in the stator teeth, the effect of rotating
magnetization, spatial harmonics, and leakage flux.
Every loss component and origin has to be described
mathematically in terms of an own loss equation. The
appropriate separation of the loss components and origins
(based on physical phenomena) is the key aspect of the
analytical core loss model. The resulting basic loss
equations are derived from the corresponding flux density
dependent loss equations (10) and (11), as shown in (14)
and (15). rec represents the maximum flux linkage within
the last electrical fundamental period, it is the corresponding
quantity to Bm. PWM caused time harmonics are modelinherent, since loss is calculated instantaneously.
1
dB
k cos() Bm
dt
h
dqs
ph(t) = ah rec
dt

dB
dt

2 kc

(15)
2

dqs

pc(t) = ac
dt
Depending on the location of calculated loss, flux
linkage is described in the stator (superscript s) or rotor
(superscript r) reference frame, respectively. Describing the
flux linkage derivative vector in stator reference frame in
terms of flux in rotor reference frame and electrical
frequency el, are shown in (16) and (17).

IV. ANALYTICAL LOSS MODEL FOR ACTIVE LOSS CONTROL

p'h(t) =

(17)

The transient (first term) and steady-state (second term) loss


components can be easily separated. The new defined
coefficients ah and ac are used to simplify (14) and (15).
They contain the material loss coefficients kh and kh as well
as several constants. Since stator flux linkage is an integral
quantity of the machine, resulting machine loss in (14) and
(15) also represents total machine loss p and not loss density
p.
The new defined coefficients a represent the individual
loss characteristics of the analyzed machine (including
geometry) and are derived from the numerical loss model in
section III. The loss coefficients ai are evaluated for every
loss location (stator / rotor) and every effect (spatial
harmonics (), permanent magnet, and leakage flux). In
total, 21 loss coefficients ai are identified for the loss model.
In Table 1, all coefficients are listed. For every loss
coefficient, a corresponding analytical loss equation exists.
In (18), the equation for eddy current loss in the stator is
shown. The first two terms represent loss caused by the
stator linked flux, while the third term represents loss
caused by the permanent magnet leakage flux pm.
2

dsr
qsr
r
r

pcs = acsd
- el qs + acsq
+ el ds
dt

dt

+ acspm

(18)

r 2
el pm

Permanent magnet loss is described by (19), where the first


two terms are related to the fundamental flux. In contrast to
(18) el is not included in this term since loss is described in
rotor reference frame (fundamental field does not cause any
flux change in the rotor at steady-state operation). The
remaining terms in (19) represent the effect of spatial
harmonics and permanent magnet flux.

(14)

50

designed for active loss control, assuming constant


coefficients is expected to be acceptable.

TABLE 1. CORE-LOSS-COEFFICIENT-ARRAY (CCA) AND ITS ELEMENTS


WHERE ROWS REPRESENT LOSS COMPONENTS AND THE COLUMNS LOSS
ORIGINS

core loss
component

spatial
fundamental
d
q

Hysteresis stator
(hs)
Hysteresis rotor
(hr)
Eddy current
stator (cs)
Eddy current
rotor (cr)
Eddy current
magnet (cm)

spatial harmonics and


PM leakage
pm
d
q

ahsd

ahsq

ahspm

---

---

ahrd

ahrq

ahrpm

ahrd

ahrq

acsd

acsq

acspm

---

---

acrd

acrq

acrpm

acsd

acsq

acmd

acmq

acmpm

acmd

acmq

Figure 4. Workflow of dynamic loss calculation in active loss control


whereby numerical simulations are performed offline in order to derive
the machine loss characteristics needed

+ acmd el (ds-pm)

)2 + acmq(el qsr)2

Stator eddy current loss

Stator hysteresis loss

instants

instants

core loss (in W)

dsr
qsr
pcm = acmd + acmq
dt
dt

(19)

)2

+ acmpm el pm

Numerical loss model Analytical loss model

The eddy current loss equation for eddy current loss in the
rotor lamination is identical to (19) with the corresponding
coefficients.
Stator hysteresis loss equation is given in (20).
Hysteresis rotor loss is calculated accordingly.
phs =

)2

(
(ddqss /dt)2

ahsd dds /dt + ahsq dqs /dt

)2

| |

Figure 5. Example comparison between results of numerical loss


model and new analytical model. The simulation is performed for a
surface permanent magnet machine at constant speed. The flux linkage
vector is changed randomly every 15 instants. Transient states are
characterized by loss peaks.

dqs
rec dt

After evaluating the characteristic coefficients ai no more


numerical calculations are required. Using the analytical
equations, core loss can be evaluated in real-time. The
corresponding workflow is illustrated in Fig. 4.
In the example shown in Fig. 5 the flux density vector
is changed randomly. It illustrates a very good match
between the results of the time consuming numerical
calculation, which takes many minutes or even hours, and
the fast analytical calculation, which can be performed
within milliseconds or even faster. The loss peaks are
caused by change in flux level, and they represent the
transient loss states as shown in Fig. 3 at instant k.

(20)

+ ahspm rec |el| pm


For a given machine, the coefficients are derived from
numerical calculations by defining specific excitation
functions which allow for differentiating between core loss
caused by d- and q-axis flux change. This task is
comparable to a classical identification problem. These
coefficients are considered constant and represent the
machine characteristics for the online loss calculation using
the analytical model. Considering constant loss coefficients
is a valid simplification. The offline identification of the
coefficients is based on FEM simulations for a typical
machine operation (within the specified operating range).
Therefore, machine saturation is taken into account for this
specific operating range. Operating points with higher
saturation automatically increase magnetizing current and
consequently winding and core loss. Using DB-DTFC loss
minimizing control, the algorithm inherently avoids high
saturation, so that it can be expected that the saturation level
is not changed dramatically. Since this loss model is

V.

EXPERIMENTAL RESULTS

The experimental results in Fig. 6 demonstrate the


significance of transient loss in electrical machines. In this
test, two flux density levels (represented by point 1 and 2)
causing the same total machine loss are identified. After
that, flux linkage is changed between these points every
sample instant at 10 kHz using DB-DTFC. The loss level
does not reach the minimum between the two reference
points, instead, it increases due to transient loss.

51

controllers. Effects like high frequency eddy current loss in


electrically conducting permanent magnets as well as spatial
harmonic content, rotational magnetization, and leakage
flux are taken into account.
ACKNOWLEDGMENT
The authors wish to acknowledge the motivation
provided by the Wisconsin Electric Machines and Power
Electronics Consortium (WEMPEC) of the University of
Wisconsin-Madison and the Leibniz University of
Hannover, Germany.
REFERENCES

Description

Pin
[W]

Pmech
[W]

Pw
[W]

Pcore
[W]

[1]

Point 1 at steady-state
Point 2 at steady-state
Switching between 1 and 2

55.9
55.9
58.9

50
50
50

4.69
5.02
7.38

1.21
0.88
1.52

[2]
[3]

Figure 6. Experimental results using an interior permanent magnet


machine. Input power Pin, mechanical power Pmech, winding loss Pw,
and core loss Pcore.

VI.

[4]

CONCLUSIONS

[5]

The experimental evaluation has shown the existence


and significance of transient machine loss. Consequently,
transient core loss modeling seems to be a promising field
of research to enable active loss control in order to increase
energy efficiency and to reduce thermal load in electrical
machines. Using the proposed dynamic loss model,
instantaneous core loss can be calculated on switching
period level based on stator flux linkage as input variable
for both surface as well as interior permanent magnet
machines. This makes it possible to predict total machine
loss for different Volt-second vector selections in DBDTFC and to select the loss minimal vector dynamically.
Combining numerical and analytical approaches has
proven to be a promising way to estimate dynamic loss in

[6]

[7]

[8]

52

J. S. Lee, C. Choi, J. Seok, and R. D. Lorenz, "Deadbeat Direct


Torque & Flux Control for Interior Permanent Magnet Synchronous
Machines with Discrete Time Stator Current and Stator Flux Linkage
Observer," IEEE Trans Industry Applications, Vol. 47, pp. 17491758, May 2011.
G. Bertotti, "General Properties of Power Losses in Soft," IEEE
Trans. Magn., vol. 24, pp. 621-630, Jan. 1988.
J. D. Lavers, P. P. Biringer, and H. Hollitscher, "A Simple Method of
Estimating the Minor Loop Hysteresis Loss in Thin Laminations,"
IEEE Trans. Magn., vol. 14, pp. 386-388, Sept. 1978.
D. M. Ionel, M. Popescu, S. J. Dellinger, T. J. E. Miller, R. J.
Heideman, and M. I. McGilp, "On the Variation of the Core Loss
Coefficients in Electrical Machines with Flux and Frequency," IEEE
Trans. Ind. Applicat., vol. 42, pp. 658-667, June 2006.
C. A. Hernandez-Aramburo, T. C. Green, and A. C. Smith,
"Estimating Rotational Iron Losses in an Induction Machine," IEEE
Tran. Magn., vol. 39, pp. 3527-3533, Nov. 2003.
R. Liu, C. C. Mi, and D. W. Gao, "Modeling of Eddy-Current Loss of
Electrical Machines and Transformers Operated by PulsewidthModulated Inverters," IEEE Trans. Magn., vol. 44, pp. 2021-2028,
Aug. 2008.
D. M. Ionel, M. Popescu, M. I. McGilp, T. J. E. Miller, S. J.
Dellinger, and R. J. Heidemann, "Computation of Core Losses in
Electrical Machines Using Improved Models for Laminated Steel",
IEEE Trans. Ind. Applicat., vol. 43, pp. 1554-1564, Dec. 2007.
D. Lin, P. Zhou, W. N. Fu, Z. Badics,and Z. J. Cendes, "A Dynamic
Core Loss Model for Soft Ferromagnetic and Power Ferrite Materials
in Transient Finite Element Analysis", IEEE Trans. Magn, vol. 40,
pp. 1318-1321, Mar. 2004.

Potrebbero piacerti anche