Beruflich Dokumente
Kultur Dokumente
Abstract—This paper proposed a super-twisting algorithm ba- area due to their simplicity, robustness against parameter vari-
sed sliding mode observer (STA-SMO) for permanent magnet sy- ations and fast response, they suffer from chattering problem
nchronous machine (PMSM) sensorless control, in which voltage- [7]. In order to reduce the chattering problem, a sigmoid
source inverter (VSI) nonlinearity is taken into consideration.
A lookup table for online tunning sliding mode coefficients function is used to replace the traditional sign function in [8].
of STA-SMO is obtained based on existing stable conditions. Besides, an adaptive gain tunning method for first-order SMO
Since mismatch between reference and actual voltage due to is also proposed to extend the sensorless operation range and
VSI nonlinearity may lead not only to estimation error but obtains good performance. However, chattering problem can’t
also to system instability, especially at low speed range. Voltage be eliminated completely due to the inherent characteristic
distortion due to VSI nonlinearity is compensated to improve
the estimation accuracy. A high-frequency harmonic components of first-order SMO. High-order SMO is an effective method
extraction method is used to compensate the VSI nonlinearity. for chattering reduction. Super-twisting algorithm (STA) is
Thus, a STA-SMO with suitable sliding mode coefficients and widely employed in observers and controllers due to its
accurate voltage parameters is obtained. The effectiveness of the chattering-free characteristic [2], [5], [6], [9]. In [2], a STA
proposed sensorless control method with VSI nonlinearity com- based SMO (STA-SMO) is employed for sensorless control
pensation is verified on a surface-mounted PMSM by simulations
and experiments. of inductance machine to eliminate chattering. However, fixed
Index Terms—permanent magnet synchronous machine; sen- sliding mode coefficients are adopted, making it difficult to
sorless control; super-twisting algorithm; sliding mode observer; extend the sensorless operation range. In [5] and [6], possibi-
voltage-source inverter nonlinearity compensation lity of operating PMSM based on STA-SMO is investigated
by simulations. Compared to conventional first-order SMO,
I. I NTRODUCTION chattering is reduced greatly using STA-SMO based sensorless
technique. However, fixed sliding mode coefficients are also
In the last decade, permanent magnet synchronous machine adopted. Furthermore, mismatch between reference and actual
(PMSM) has been widely used in industry area. High perfor- voltage is ignored in [5] and [6], which may bring large
mance field oriented control (FOC) of PMSM needs accurate estimation error at low-speed range.
rotor position information which is usually obtained by resol- In this paper, a STA-SMO (STA-SMO) for PMSM sen-
ver or encoder. Sensorless control is becoming more and more sorless control is proposed where voltage source inverter
popular due to its compactness, low cost, high reliability, etc. (VSI) nonlinearity is taken into consideration. A STA-SMO
Sensorless control methods can be divided into two groups: is designed to estimate stator currents. At the same time, a
1) magnetic saliency based methods [1], 2) machine model lookup table is used for online tunning of the sliding mode
based methods [2]. In the first category, inductance variations coefficients. The proposed lookup table satisfies the existing
due to saturation or saliency of PMSM are used to estimate stable conditions, which means the stability of self-tunning
rotor position, which is suitable for standstill or low speed algorithm can be proved by the same method. Furthermore,
sensorless operation. Therefore, it is difficult to use magnetic VSI nonlinearity compensation technique is also employed to
saliency based methods for surface-mounted PMSM sensorless reduce the mismatch between reference and actual voltage.
control. On the other hand, machine model based sensorless With the proposed signal process method, the 6th harmonic
control methods, including model reference adaptive system components caused by VSI nonlinearity can be extracted wit-
(MRAS) [3], extended Kalman filter (EKF) [4] and sliding hout attenuation. Then, the extracted 6th harmonic components
mode observer (SMO) [2], [5], [6], are suitable for medium are injected into the reference voltage to compensate VSI
or high speed sensorless operation, where back electromotive nonlinearity.
force (back-EMF) is easier to be estimated.
II. STA-SMO
For machine model based sensorless techniques, MRAS
may have difficulty in the model parameters adaptation and A. Mathematic Model of Surface-Mounted PMSM
EKF needs heavy online matrix computing. Although SMO Assuming that the inductance of surface-mounted PMSM
based sensorless techniques are widely employed in industrial in αβ stationary reference frame are the same, the voltage
978-1-5090-4281-4/17/$31.00 2017
c IEEE.
equations of PMSM in αβ-axis can be expressed as By subtracting (2) from (5), the state equations of current
errors are derived
diα R 1 1
= − i α + uα − e α (1a) dīα Rs 1
dt L L L =− īα + (k1 |īα |1/2 sign(īα )
dt L
s L s
diβ R 1 1 + k2 sign(īα )dt + eα )
= − i β + uβ − e β (1b)
dt L L L (6)
dīβ Rs 1
where iαβ , R, L, uαβ , eαβ are stator currents, stator resistance, =− īβ + (k1 |īβ |1/2 sign(īβ )
dt L
s L s
stator inductance, stator voltage and back electromagnetic
force (back-EMF), respectively. Stator inductance in αβ-axis + k2 sign(īβ )dt + eβ )
are assumed to be the same in this paper.
When STA-SMO reaches to the equilibrium point, it means
B. Basic Principle of STA īαβ = 0. Then, the estimated back-EMF can be obtained as
follows
STA is a well-known second-order SMO proposed by A.
1/2
Levant in [10] to reduce chattering. A strong Lyapunov êα = −k1 |īα | sign(īα ) − k2 sign(īα )dt (7a)
function is proposed in [11] to obtain stable conditions of
STA. The basic form of STA can be expressed as êβ = −k1 |īβ |1/2 sign(īβ ) − k2 sign(īβ )dt (7b)
dx1
= −k1 |x̄1 |1/2 sign(x̄1 ) + x2 + ρ1 (x1 , t) The estimated rotor position and speed can be obtained by
dt (2) following equations
dx2
= −k2 sign(x̄1 ) + ρ2 (x2 , t) θ̂ = arctan(−êα /êβ )
dt
(8)
where xi , x̄i , ki , ρi and sign() are state variables, error bet- ω̂ = ê2α + ê2β /ψf
ween estimated and actual state variables, sliding coefficients,
D. Proposed Self-Tunning STA-SMO Based on Lookup Table
perturbation terms and sign function, respectively.
According to the stability analysis presented in [11], sliding From the above analysis, it can be seen that stable conditions
mode coefficients of the above STA-SMO should be chosen for STA-SMO is not explicit. It seem that different sets of
as follow parameters could lead to similar finite time convergence of
5δ1 k1 + 4δ12 STA-SMO. In practice, although sliding mode coefficients
k1 > 2δ1 , k2 > k1 (3) selected based on (3) guarantee stable operation of the system,
2(k1 − 2δ1 )
the actual performance is different. Take [5] forexample,
and the perturbation terms should be globally bounded by sliding mode coefficients are set to k1 = 1.5 f + and
k2 = 1.1f + , where f + is the upper bound of perturbation. Sli-
|ρ1 | δ1 |x1 |1/2 , ρ2 = 0 (4) ding mode coefficients that satisfy the largest perturbation are
selected. However, large sliding mode coefficients may lead to
where δ1 is a positive constant. If the STA-SMO satisfies the chattering in sensorless control system, especially at low-speed
above conditions, it converges in finite time. range where perturbation is much smaller. Therefore, online
tunning of sliding mode coefficients under different conditions
C. STA-SMO for PMSM are necessary.
In this paper, a lookup table for STA-SMO based on existing
The STA based stator current observers for PMSM sensor-
stable conditions is used. From (5), it can be seen that the
less control are constructed as follow
perturbation terms can be written as
dîα R 1 1 R 1
= − îα + uα + k1 |īα |1/2 sign(īα ) (5a) ρ1 = − îαβ + uαβ (9)
dt L
L L L L
1
+ k2 sign(īα )dt It is obvious that the perturbation changes under different
L operation conditions and it is mainly affected by stator voltage.
Large perturbation with small sliding mode coefficients may
dîβ R 1 1 cause system instability, small perturbation with large sliding
= − îβ + uβ + k1 |īβ |1/2 sign(īβ ) (5b) mode coefficients may cause chattering. Therefore, it is ne-
dt L
L L
1 cessary to tune the sliding mode coefficients online. In this
+ k2 sign(īβ )dt paper, a lookup table is used for online tunning the sliding
L
mode coefficients of STA-SMO. The sliding mode coefficients
where ˆ denotes estimated value. Comparing with (1) and (2), are obtained by experiments.
1 1
it can be noticed that − R
L iα + L uα and − L iβ + L uβ in PMSM
R
The block diagram of STA-SMO with lookup table is shown
voltage equations are regarded as perturbation terms. in Fig. 1.
Fig. 1. Waveforms of Dd and Dq under id = 0 control and ω = 52.4 rad/s.
III. VSI N ONLINEARITY C OMPENSATION Fig. 2. Waveforms of Dd and Dq under id = 0 control and ω = 52.4 rad/s.
VSI nonlinearity may cause mismatch between reference
and actual voltage, which will enlarge estimation error. Voltage
distortion in dq rotational reference frame caused by VSI
nonlinearity can be expressed as [12]
Vd,dead Vdead · Dd
=
Vq,dead Vdead · Dq
= Vdead × 2
⎡ ⎤ Fig. 3. High-frequency harmonics components extraction algorithm.
2π 2π
⎢ cos(θ) cos(θ − ) cos(θ + ) ⎥
×⎣ 3 3
2π 2π ⎦ TABLE I
−sin(θ) −sin(θ − ) −sin(θ + ) M AIN PARAMETERS OF PMSM AND C ONTROL S YSTEM
3 3
(10a)
⎡ ⎤ Items Value Items Value
sign(ias ) Stator resistance (Ω) 0.57 Rated phase current (A) 10.6
× ⎣ sign(ibs ) ⎦ Rotor flux linkage (Wb) 0.1385 DC bus voltage (V) 150
sign(ics )
Stator inductance (mH) 10.24 Rated speed (rpm) 500
1, i 0 Pole pairs 10 Rated torque (N·m) 20
sign(i) = (10b)
−1, i < 0 Rated power (kW) 10
Tdead + Ton − Tof f
Vdead = Vdc (10c)
Ts
where Tdead , Ton , Tof f , Ts , Vdc and Dd, Dq are dead-time ,
turn-on delay, turn-off delay, sampling period, DC bus voltage
and distorted dq-axis voltage, respectively. The simulated
waveforms of Dd and Dq under id = 0 and ω = 52.4 rad/s is
shown in Fig. 2. It is obvious that distorted voltage in dq-axis
are mainly composed by 6th harmonic components. It is repor-
ted in [12] and [13] that distorted voltage can be compensated
by injecting 6th harmonic components into reference voltage.
In [13], first-order low-pass filter (LPF) is used to obtain high- Fig. 4. System diagram of PMSM sensorless control based on proposed self-
frequency components. Under ideal situation, a first-order LPF tunning STA-SMO and VSI nonlinearity compensation.
is enough for extracting high-frequency components. However,
offset of current sensors may change, which could bring 1st
harmonic components. Traditional high past filter (HPF) can’t and 7th harmonic components are transfered to 6th harmonic
filter out high-frequency components completely, because it is components by inverse park transformation. The 6th harmonic
difficult to choose a suitable cut-off frequency to distinguish components can be obtained with negligible phase delay and
1st and 6th harmonic components. gain attenuation. The 6th harmonic components due to VSI
In this paper, a simple high-frequency components ex- nonlinearity can be minimized by injecting the extraction sig-
traction method considering the 1st harmonic components is nals into reference voltage. Compared to LPF based method,
proposed. The algorithm for high-frequency components ex- dq-axis currents still compose the 1st harmonic components
traction is shown in Fig. 3. First, park transformation presented by using the proposed method. Otherwise, the 1st harmonic
in (11) is used to transfer 1st and 6th harmonic components components caused by offset current error cannot be modified
to dc and 5th, 7th harmonic components, respectively. And by adjusting the offset of current sensors to normal level [14].
then, the dc component is filtered by a first-order LPF, whose
transfer function is shown in (12). Thus, the dc components u∗dc,5,7th cos(θ) sin(θ) u∗d
= (11)
can be filtered out by the LPF completely. Finally, the 5th u∗dc,5,7th − sin(θ) cos(θ) u∗q
Fig. 5. Simulation results of STA-SMO with lookup table.
(b) (b)
(c) (c)
(d)
Fig. 7. Simulation results of proposed VSI nonlinearity compensation method.
(d)
The PMSM is operated under a speed command of 50 rpm
and a 10 N·m load. The dead time of the inverter is 4 us. In
Fig. 7, it can be seen that the VSI nonlinearity has a great
influence on reference voltages and currents. Sixth harmonic
components become dominated at low speed. These high
frequency components may cause large position estimation
error. However, the proposed VSI nonlinearity compensation
method can reduce the 6th harmonic components greatly as
shown in Fig. 8. (e)
Fig. 8. Simulation results of proposed VSI nonlinearity compensation method.
V. E XPERIMENTAL R ESULTS
Since the SMO based sensorless control method can not In order to verify the proposed self-tunning STA-SMO, the
operate at zero speed, the PMSM is started up in open loop. speed command of a surface-mounted PMSM is decelerated
The startup process is shown in Fig. 9. from 450 rpm to 30 rpm with a 5 Nm load. The experimental
Fig. 9. Startup process STA-SMO.