Documente Academic
Documente Profesional
Documente Cultură
– 1 –
– 2 –
aaa Is ( s ) the symmetry of its track, we have ih = 0 when t0/2 has elapsed,
Fm ( s ) = (2) i.e. at the end of each half-cycle (3). A current signal sampled
Us ( s )
τ r s + 1 + jω r τ r at this time instant equals the fundamental current is1.
=
rσ (τ σ' s + 1 + jw sτ σ') (τ r s + 1 + jw r τ r ) − k1 (1 − jwτ r )
C. Inverter and sampling delay
Digital signal processing and the delay introduced by the
where k1 = kr lm/(rstr). PWM inverter are approximated by a transfer function of first-
order, referred to in synchronous coordinates
B. Carrier modulation
Carrier modulation allocates the six commutations of a six- Us ( s ) 1
Fd ( s ) = = , (5)
pulse inverter periodically within constant time intervals T0 = U* ( s ) τ d s + 1 + j ω sτ d
1/fs, where fs is the switching frequency [3, 4]. The resulting
where Us(s) and U *(s) are the Laplace transforms of the
fundamental output voltage u1 depends on the modulation
respective output and input signals us(t ) and u*(t ). The
index m = u1 /u1six-step ,
term jws td in the denominator of (5) results from the repre-
where u1six-step = 2 ud/π is
.4 sentation of the delay element in synchronous coordinates.
jℑm SVM the fundamental voltage
The normalized delay time td = 1.5 wsR/(2fs) [6], where 2fs is
at six-step operation, and
the sampling frequency and fs is the switching frequency.
SOM .2 ud is the dc link voltage.
The signal flow graph of the delay element (5) is shown as
A preferred variant of
part of Fig. 3.
carrier modulation is
– .2 .2 .4
space vector modulation
III. LINEAR CURRENT CONTROLLERS
ℜe
[5]. Among carrier mod-
ulation schemes, this Using space vector modulation permits the acquisition of
method produces the low- the instantaneous fundamental current component is1. This
est harmonic distortion favors the implementation of fast closed-loop stator current
– .4 [3], and also permits the control without the need of having a low-pass filter in the
instantaneous acquisition feedback path to reduce the current harmonics. Limiting the
Fig. 2 Trajectories of the steady- of the fundamental com- current reference vector in magnitude is a means for overload
state harmonic current ih at space ponent i s1 of the stator protection.
vector modulation (SVM) and at current vector is of an in-
synchronous optimal modulation A. PI current control
verter-fed machine.
(SOM); modulation index m = 0.85, The current control system operates in field coordinates.
switching frequency fs = 200 Hz
The modulator gener-
ates a sequence in time of Its signal flow graph is shown in Fig. 3. It is customary when
three switching state vectors ua, ub, u0 per half-cycle T0/2, designing the current controllers to neglect all signals that
were ua and ub are the two vectors in closest spatial neigh- multiply by imaginary gain coefficients. The respective sig-
borhood to the reference vector u*, and u0 stands for one of nal lines are shown in grey in Fig. 3. Given this approxima-
the zero vectors u0{+ + +} or u7{– – –}, respectivly. Two tion, the respective d- and q-component signals of the current
subsequent half-cycles are defined as loop result independent from each other. A PI controller is
assigned to each current component. The respective transfer
u0{t0/2} … ua{ta} … ub{tb} … u7{t0/2} (3a) functions are
u7{t0/2} … ub{tb} … ua{ta} … u0{t0/2} (3b) U*( s ) τ s +1
Fr ( s ) = = gc i
(6) ,
where ta, tb and t0 are the respective on-durations of the D Is ( s ) τ is
switching state vectors ua, ub and u0. The symmetric ar-
where ti is the integration time constant and gc is the propor-
rangement of switching states
in the two half-cycles (3)
makes the trajectory of the
harmonic current uir /rσ kr rstr
current
D is controller sampling
ih = is − is1 (4) delay
is* u* us 1 is yr
describe a closed pattern, gc,ti td rs tσ' lm tr
centered in the origin of the jtd jtσ' jtr
complex plane as shown by is
the red curve in Fig. 2. stator rotor
During the on-duration of ws wr
a zero vector, the trajectory ih
of the harmonic current pass- w
(≈ const.)
es through the origin of the Fig. 3 Current controlled induction motor, signal flow graph. The signals
complex plane, Fig. 2. Given represented by shaded lines are conventionally neglected in the controller design
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
– 3 –
80 A 80 A
aaa
60 60
jℑm{is} jℑm{is}
i q1 40 i q1 40
iq1 iq1
is1 is1
20 t 20 t
0 0
20 A 20 A
id1
0 0
id1
i d1 – 20 i d1 – 20
a) b) a) b)
– 40 – 40
0 20 40 60 ms – 40 0 40 A 0 20 40 60 ms – 40 0 40 A
t ℜe{is} t ℜe{is}
Fig. 4 q-current step response, PI-controller; operation at rated Fig. 6 q-current step response, PI-controller with feedforward com-
speed and 200 Hz switching frequency, (a) fundamental current pensation; operation at rated speed and 200 Hz switching frequen-
components, (b) trajectory of the fundamental current space vector, cy, (a) fundamental current components, (b) trajectory of the funda-
marked by circles at every 5 ms mental current space vector, marked by circles at every 5 ms
tional controller gain. The open loop transfer function be- – 170% deviation from its constant reference value. The ef-
comes fect leads to a temporary demagnetization of the machine. Such
undesired behavior is the consequence of using an unsuited
gc τ i s + 1 1 1 controller and neglecting the imaginary gain coefficients in
Fo ( s ) = . (7)
rs τ i s τ d s + 1 τ σ' s + 1 its design.
Letting ti = ts' eliminates the pole at – 1/ts'. The gain gc =
B. PI current control with feedforward compensation
ts'/2td is adjusted for critical damping which yields the closed
loop transfer function The dynamics of the current control loop can be improved
by compensating the motion induced voltage jwsts' is in the
F0 ( s ) I (s) 1 stator winding through a feedforward signal. Fig. 5 shows the
Fc ( s ) = = s* = . (8)
F0 ( s ) +1 I ( s ) 2τ 2 s 2 + 2τ s + 1 signal flow diagram. The signals represented by shaded lines
s d d
are again neglected in the controller design. The feedforward
This design procedure, although being based on an approx- signal is chosen as jws l̂s is which, including the gain factor
imated transfer function, gives satisfactory results if the sam- 1/r s at the input of the stator delay element, amounts to
jwsts' is. In the foregoing equation, l̂s is the
estimated total leakage inductance. Not con-
sampling uir /rσ sidered in this compensation is the sampling
delay delay. The delay is negligible if t d << ts'
is* gc
u* us 1 is holds, which is only valid at higher switch-
ti td rs t'σ ing frequency. The larger delay at low switch-
jtd jtσ'
ing frequency leads to an incomplete com-
is jlˆs pensation. It is seen in the signal flow graph
PI controller stator winding
Fig. 3 that the compensation signal gets de-
ws layed in the process of digital sampling be-
fore it is used for cross-coupling compensa-
Fig. 5 Current control with feedforward compensation
tion. The compensation is therefore not fully
effective. Parameter errors in l̂s may deteri-
pling delay is negligible, t d << ts'. This applies for switching orate the situation.
frequencies of several kHz. The sampling delay is then small, The performance of feedforward compensation at low
permitting the controller to override the machine dynamics switching frequency is illustrated in Fig. 6. Cross-coupling
and to independently force the stator current to its command- still exists with a temporary deviation of the d-current by
ed value. to – 125%. Conditions would turn worse if an error existed in
The eigenbehavior of the machine cannot be overturned if the estimated leakage inductance l̂ s. Parameter sensitivity is
t d and ts' have the same order of magnitude. This is demon- a common demerit of feedforward control.
strated in the oscillogram Fig. 4, showing a step response of The deficiencies of this approach persist even at increased
the q-current component at very low switching frequency, fs switching frequency (e.g. 1.1 kHz, [7]). These shortcomings
= 200 Hz. Cross-coupling between the q- and d-current dom- indicate that PI current control is not a viable solution when
inates the response. The peak value of the d-current exhibits the inverter operates at low switching frequency.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
– 4 –
C. Current controller with complex eigenvalues The transient response Fig. 8 shows that cross-coupling is
The complex coefficients in the transfer functions of the almost eliminated.
induction motor (2) and the sampling delay element (5) sug-
gest the design of a current controller having complex eigen- IV. TRAJECTORY TRACKING CONTROL
values [7]. The procedure starts from the transfer function of The primary objective is to operate a medium voltage in-
the plant verter at very low switching frequency such that the switch-
Is ( s ) 1 ing losses reduce and the power output increases. The related
Fp ( s ) = = (9) dynamic problems with respect to response time and cross-
U*( s ) τ d s + 1 + jωsτ d
coupling can be solved using a current controller having com-
τ r s + 1 + jω r τ r
⋅ plex eigenvalues, described in Section III-C. A remaining
rσ (τ σ' s + 1 + jw sτ σ') (τ r s + 1 + jω rτ r ) − k1 (1 − jwτ r ) problem is the high harmonic content of the machine currents.
which includes the delay (5) and the machine (2). Based on Its solution requires a different approach.
this equation, the current controller is obtained as
A. Synchronous optimal pulsewidth modulation
U* ( s ) τ s + 1 + jω sτ d Carrier modulation produces a fixed number of inverter
Fr ( s ) = = gc rσ ⋅ d (10)
D Is ( s ) (τ d s + 1) commutations per sampling interval; optimal modulation tech-
(τ σ' s + 1 + jwsτ σ') (τ r s + 1 + jω rτ r ) − k1 (1 − jwτ r )
⋅ 80 A
τ i s (τ r s + 1 + jω r τ r ) is1
60
The controller is designed such that the poles and the ze- jℑm{is}
i q1 40 iq1
roes of the plant are cancelled, and an integrating term is add-
ed for steady-state accuracy. The signal flow graph Fig. 7 il- 20
lustrates that a compensating element is provided in the con- t
troller for every delay element of the plant. 0
The open-loop transfer function Fo(s) = Fr(s) . Fp(s), obtained 20 A
from (9) and (10), results as a second-order system with real
0
coefficients:
i d1 – 20 id1
1 a) b)
Fo( s ) = Fr ( s ) Fp( s ) = gc . (11)
τ i s (τ d s + 1) – 40
0 20 40 60 ms – 40 0 40 A
This equation is identical to (7) and hence also (8) applies t ℜe{is}
here. There is a difference, though, since no approximations
have been made in the controller design. Since the state Fig. 8 q-current step response, Current controller with complex
eigenvalues; operation at rated speed and 200 Hz switching fre-
variables in (8) are of complex nature, the two eigenvalues of quency, (a) fundamental current components, (b) trajectory of the
the characteristic equation (8) are single complex poles [7]. fundamental current space vector, marked by circles at every 5 ms
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
– 5 –
modulator
pattern pattern
select modifier = optimal pulse sequence uss(I), and since the initial
u* ~ value ys(I)(t1) in (12)
wsss
ω st1
yss
dˆ(t) Dys DP is y s(I) (t1 ) = ∫ uss(I) (α )dα − ys(I)(α = 0) , (13a)
optimal trajectory 0
trajectory controller
us 2π 2π
wsss yˆs
hybrid
M
3~
y s(I) (α = 0) = ∫ ∫ uss(I) (α ) dα 2 dα1 (13b)
w 0 0
observer
is defined such that it is located on the optimal flux
trajectory. Equation (13b) determines the initial value
Fig. 9 Self-controlled drive system with synchronous optimal pulsewidth in (13a) by taking the average over a full fundamental
modulation and flux trajectory tracking control period of the optimal flux trajectory.
It is now assumed that the pulse pattern changes at t2
> t1. The new pattern is P (II); it generates the voltage wave-
niques do not have this restriction. Instead, the commutations
form uss(II). The flux trajectory in interval II develops as
are distributed over a fundamental period in a defined manner
such that an optimum criterion is satisfied. Minimizing the t
total harmonic distortion is considered here [8]. y s (t ) = ∫ uss(II) (t ) dt + y s(I) (t2 ) .
(II)
(14)
The switching angles ak, k ∈ 1…N, are precalculated off- t2
line under assumed steady-state conditions. The angles are
stored in a memory table as pulse patterns P(m, N). The pat- This is not the optimal flux trajectory since its initial value
(I)(t ) at the time t of pattern change is not computed as in
terns are functions of the modulation index m and the pulse y s 2 2
number N = fs/f1. As N is integer, the pulse patterns result (13); it is rather (II)
carried over from the flux trajectory in
(I)
synchronized with the fundamental voltage and its subhar- interval I: ys (t2) = ys (t2) must hold since the stator flux
monic content is eliminated [9]. vector cannot instantaneously change. More general: It is not
The signal flow graph of a synchronous optimal pulsewidth possible to continue on the optimal trajectory after a pattern
modulator is shown in the upper portion of Fig. 9. The con- change, since any two steady-state flux trajectories that asso-
trolling input is the reference voltage vector u*. Its amplitude ciate to two different pulse patterns do not have the same
u* ∝ m selects the appropriate pulse pattern P(m, N) which is value at a given time instant. Switching from one pulse
then passed on to the modulator. The phase angle arg(u*) pattern to another therefore entails a displacement from the
determines the actual pulse location within the fundamental optimal flux trajectory.
period of the pattern. The resulting switching angles a k are
translated to switching instants tk = a k/w s ss, where w s ss is a 200 A
smoothed signal that averages the stator frequency w s, Sec- ia
tion IV-E. The output of the modulator is a sequence of dis- ia
0 a)
crete switching state vectors uk that control the inverter.
The comparison with space vector modulation, Fig. 2 and
in Fig. 10, shows that synchronous optimal modulation pro- – 200
duces much lower harmonic content of the stator currents. [10]. 0 10 20 30 40 50 ms
The difference between both methods reduces as m reduces. t
It is therefore preferred using space vector modulation in the 200 A
lower modulation range, e.g. at m < 0.3, [11]. ia
ia
B. The dynamic modulation error 0 b)
Steady-state operation with optimal pulse patterns is as-
sumed in the following discussion. A time interval I is con- – 200
sidered first in which the stator voltage is continuously de- 0 10 20 30 40 50 ms
rived from only one particular pulse pattern P (I). The trajec- t
tory of the stator flux linkage vector develops according to Fig. 10 Measured waveforms of the phase a stator current in a
t three-level inverter at 200 Hz switching frequency and 33.5 Hz
fundamental frequency; (a) space vector modulation, (b) synchro-
y ss(I) (t ) = ∫ uss(I) (t ) dt +y s(I) (t1 ) , (12) nous optimal modulation
t1
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
– 6 –
3 d∑
1
wsss D da = u sai D t si , (17)
∠x i =1
1
(S) and similarly for phases b and c. The result is the possible
k ryˆr yˆs1 rs iˆs1
rs /sls change
wsss n
u* yˆs1 yˆs1 u*'
6 d∑
1
ts' ts' j Dd = u [(2 sai Dtai – sbi Dtbi − sci Dtci ) (18)
i =1
rotor + j 3 (sbi Dt bi – sci Dt ci )]
Dys e jd kr
Gs(w) of the complex dynamic modulation error
yˆs [d ] yrd within the sampling interval. A required
stator tr change Dd * (16) may not be executable
is
sls i s(F)
e jd lm tr since the switching transitions can be only
ŵ s
D
ŵ r d̂ displaced within the time limits of the cur-
[q ] 1 rent sampling interval. A remaining error
w N
w must be left for compensation in the subse-
quent interval.
In (16), 1, a and a2 are the respective uni-
Fig. 12 Hybrid observer for the estimation of fundamental machine quantities ty vectors in the directions of the phase axes
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
– 7 –
aaa parameter the machine generate the reference voltage vector u* for the
compensator
modulator. The machine then controls itself, as shown in the
e jd wst0 signal flow graph Fig. 9. This implies that any given steady-
ud state operation is perpetuated, unless external disturbances
u*' d̂ u* sync.-opt. uk change. It also implies that the drive system is not controlla-
ejj
=
PWM ~ ble. Controllability is established by separate means which
will be discussed in Section IV-F.
rotor flux The full-order observer Fig. 12 that generates the required
yr* us steady-state signals is implemented in a hybrid design. The
stator is modelled in stationary coordinates
ys* Dys Dd *
yˆ r d yˆ s1 ˆ
speed e jd τ s' + y s1 = k r yˆ r + Gs (ω )(yˆ s1 − yˆ s )+ τ s' u* (19a)
dτ
w* d̂
is and the rotor in field coordinates
hybrid
observer d yrd
yˆs1 τr + yrd = l m isd (19b)
M dτ
w
w 3~
lm iq
δ =∫ + ω dτ (19c)
τ r yrd
Fig. 13 The superimposed control system. The synchronous opti-
mal pulsewidth modulator is detailed in Fig. 9. The transformations where ts' = sls/rs is the transient time constant of the stator
between the respective coordinate systems are performed by the and d is the rotor field angle.
field angle d̂ . The stator model receives the reference voltage u* of the
a, b and c. modulator as an input. This is a signal without harmonic con-
The trajectory controller in Fig. 9 uses the optimal trajecto- tent. The model therefore generates the fundamental compo-
ry yss as a the target, also during a dynamic process. The con- nent ŷs1 of the stator flux vector. A correcting signal ŷs1 – ŷs
troller then performs a pattern optimization in real-time. is added to its input through a gain tensor Gs(w), [13, 14], to
account for model parameter mismatch.
D. The steady-state requirement The rotor model (19b, 19c) estimates the rotor field angle d̂
Ideal steady-state operation is difficult to establish in a drive and the rotor flux vector ŷr. The latter is transformed to sta-
system. Signal noise and external disturbances inevitably gen- tionary coordinates to serve as an input to the stator model
erate minor, but perpetual excursions of the control signals (19a), and to generate the stator flux vector
around the operating point. Using a noisy reference vector u* yˆs = k ryˆr + sls i s . (20)
to control the optimal modulator would create repeated pat-
tern changes. In consequence, dynamic modulation errors that forms part of the correcting signal ŷs1 – ŷs in (19a).
would occur in every sampling interval. Although the trajec- The output of the hybrid observer
tory controller is designed to counteract the deviations, such
yˆ − k yˆ
situation is undesirable for the following reasons: u*' = jω s ss yˆs1 + rs s1 r r (21)
sls
• The volt-second error caused by a pattern change can be
substantial even when the noise-induced variations of the is a first estimate u*' of the reference voltage vector u*. In
reference voltage vector u* are small. This is because (21), the second term represents the resistive voltage rs is.
the functions ak(m) of the optimal angles are discontin- The first term ws ss = arg(ŷr) is the steady-state stator fre-
uous, Fig. 11. A small deviation of m may lead to large quency. This smooth signal is obtained using the rotor equa-
displacements in time of the PWM pulses. A full com- tion (19b), dominated by the large rotor time constant tr, as a
pensation is then only possible if favored by the particu- filter. The signal flow is illustrated by the graph Fig. 12.
lar locations of pulse transitions in the respective sam-
pling interval. F. Decoupled control of torque and flux
• Since dynamic modulation errors may appear in a rapid The superimposed control system Fig. 13 consists of the
sequence, the resulting deviations may accumulate to respective controllers for rotor flux and speed. Their output
large magnitudes before the trajectory controller can signals are joined to form the complex reference signal ys*.
counteract. This signal is compared with the actual fundamental flux vec-
tor ŷs1 obtained from (19a). The result is the commanded
It is therefore expedient to maintain the modulator in an change Dys* of the stator flux vector, defined in rotor field
approximate steady-state, so as to meet the condition at which coordinates. Such signal would conventionally act as an input
the pattern optimization was performed (Section IV-A). to the pulsewidth modulator.
E. The self-controlled machine This is not viable when operating with synchronous opti-
mal pulse patterns. The modulator then requires a smooth ref-
The requirement of near-steady-state operation of the mod-
erence input u*, a signal that must not be distorted by com-
ulator is met in an approach in which the terminal voltages of
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
– 8 –