Professional Documents
Culture Documents
Abstract— Stator resistance adaptation is a crucial issue on the back-emf signal perturbing the q-axis current track-
in sensorless control of induction motor, particularly at low ing error. In [9] it has been shown that, under assumption
speed. In this paper, an improved indirect field oriented
controller based on high gain speed estimation, which was of known stator resistance, local asymptotic tracking of
previously proposed by the authors, is deeply revised in or- smooth speed and flux reference trajectories along with
der to add a stator resistance adaptation mechanism. Sim- asymptotic field orientation are obtained with unknown
plicity and easy-to-tune structure of the non-adaptive ver-
sion of the controller is preserved, while its robustness is sig-
load torque, if the speed observer is designed sufficiently
nificantly improved thanks to stator resistance estimation. fast and if persistency of excitation is guaranteed, namely
Experimental results are provided to show the effectiveness avoiding dc excitation.
of the proposed method. Similarly to other methods, using the solution proposed in
[9] it results that relevant uncertainties in stator resistance
I. Introduction value introduce speed estimation error, leading to uncor-
rect speed tracking and, in some cases, to instability.
Speed sensorless induction motor (IM) drives are becom- The main purpose of this paper is to revise the solution
ing wide spread solutions thanks to cost reduction and in- of [9] to add an adaptation mechanism which avoids the
creased reliability with respect to high performance sen- above-mentioned phenomenon, preserving the basic prop-
sorized IM drives. Many solutions have been presented erties of the original solution. Analyzing the effects of a
in the literature; owing to space limitation, an extensive wrong stator resistance value on the algorithm of [9], it
overview is omitted here, see [1], [2]. Sensorless control can be noted that a steady-state error arises in the current
techniques can be classified into two main areas: solutions component devoted to flux control (d-component), while
based on the fundamental IM model and standard control the stability property is lost only for very large stator re-
signals and methods exploiting IM non-ideality and/or sig- sistance error. Consequently, the main idea is to make ex-
nal injection for speed estimation. The former represents plicit the quasi-steady state d-axis current tracking error
an interesting approach, since it does not require ad hoc model to design a stator resistance adaptation mechanism.
IMs or extra hardware components. Nevertheless, this ap- A suitable and simple on-off supervisor has been added to
proach is critical when the excitation frequency is close to disable the stator resistance estimation mechanism in par-
zero (dc-excitation), due to reduction of back-emf signal for ticular working conditions where it is not useful.
speed estimation and high sensitivity to model parameters Main advantage of the proposed method is that the
and measurement errors [3]. physically-based structure of [9] is preserved leading to a
The most critical parameter affecting performances at tuning procedure similar to standard electrical drives with
low speed is stator resistance [4], which varies depending cascaded control loops.
on temperature and hence is required to be estimated on- The paper is organized as follows. In Section II the IM
line. Various techniques for sensorless control with stator model is presented and the control problem is formulated,
resistance estimation have been presented [5], [6], [7], [8]. in Section III the sensorless controller of [9] is deeply re-
In this paper a speed sensorless controller with a novel vised in order to add a suitably-designed stator resistance
on-line stator resistance adaptation is presented. The pro- estimator, in Section IV experimental results are presented
posed solution is based on the sensorless control algorithm in order to show the achievable performance. Finally, in
reported by the authors in [9], which has the following Section V conclusions are given.
properties: (a) an improved Indirect Field Oriented (IFO)
Control [10] is exploited to obtain decomposition into feed- II. Induction motor model and control problem
back interconnected electromechanical and electromagnetic statement
dynamics without flux estimation; (b) speed estimation is The equivalent two-phase model of the symmetrical IM,
performed with a high-gain adaptation mechanism based under assumptions of linear magnetic circuits and balanced
1264
r̃s ∗
From controller equations and error dynamics, it can be ω̃˙ = −kω ω̃ − T̃L + kω
βψ ∗ αβ ψ̃q − βω ψ̃d − σ iq +
noted that main features of the controller are:
• the controller shows a cascaded structure with inner cur- + µ(ψ̃d i∗q − ψ̃q id ) (12)
rent control loops (3) and outer flux (4) and speed (5) con-
T̃˙L = kωi ω̃ − βψ
kωi
∗ αβ ψ̃q − βω ψ̃d − r̃s ∗
σ iq
trol loops;
• the current loops are composed by proportional con-
¯ are evaluated with the quasi-steady state
where ω̄0 , ω̄2 , ω̂
troller and feed-forward actions based on model inversion;
• vector flux control is based on the improved IFOC struc- value of eω given in (10). Error dynamics (11), (12) is
ture, with additional stabilizing term ν0 and no flux esti- composed by the linearly-cascaded-interconnected flux and
mation required to obtain field orientation; speed subsystems. Under assumption of r̃s = 0, in [9] it has
• speed controller is composed by feed-forward terms and been shown that the origin of the overall error dynamics
a PI controller based on the estimated speed; is locally exponentially stable if a persistency of excitation
• speed estimation (6) is based on the back-emf signal in condition is satisfied.
the iq current dynamics. In fact, according to the q-axis
current controller (3) and assuming, as preliminary step, B. Stator resistance adaptation
r̃s = ψ̃d = ψ̃q = 0, the back-emf signal βpωψd is compen- In order to preserve the stability properties of the solu-
sated by βpω̂ψ ∗ exploiting only estimated and reference sig- tion presented in [9], the rs adaptation mechanism is de-
nals. Hence, uncorrect speed estimation leads to non-null signed to be slower than error dynamics (11), (12). In
q-axis current tracking error in (9), which can be exploited particular, it is based on the steady-state of that dynam-
through the integral term (6) in the speed observer. By the ics. This is admissible only if the stability of (11), (12)
way, in the following analysis the flux and stator resistance is preserved even when r̃s perturbation is present. This is
estimation errors will be considered. conceivable according to robustness of exponentially sta-
Error dynamics can be decomposed into three feedback ble systems with respect to bounded disturbances and if r̃s
interconnected subsystems: electromagnetic (7), mechani- remains small thanks to a suitably-designed on-line adap-
cal (8) and estimation dynamics (9). In particular, select- tation.
k2
ing kω , kωi such that kωi = 2ω the isolated mechanical dy- Under steady-state operating conditions, i.e. with con-
namics (8) can be rendered exponentially stable. Similarly, stant ω ∗ , ψ ∗ , TL , solving (11), (12) with time derivatives
k2 equal to zero, the dependence on the stator resistance esti-
selecting kiq , kio such that kio = 2iq the isolated estimation
dynamics (9) can be made exponentially stable, too. In mation error r̃s of the steady-state d-axis current tracking
addition, as underlined in [9], defining = kkiq ω
and impos- error ĩd and the speed tracking error ω̃ can be analytically
determined. The analytical dependence of ĩd on r̃s is the
ing it sufficiently small, time-scale separation between the
following
estimation subsystem (fast) and the speed and flux track-
ing error dynamics (slow) can be achieved. Hence, under (t)
ĩd = − ndrr(t) r̃s
this particular choice of gains, the estimation error dynam-
ics (9) can be approximated by its quasi-steady state given nr (t) = αi∗q + (ω0 − pω̂)id + (pω̂ − pω)id
by (13)
i∗
¯ĩ = 0 dr (t) = σ kid ω0 + α pω̂ + αLm ψq∗ +
q
(10) +(kid − αLm β)(pω̂ − pω)) .
1
ēω = ω̃ − βpψ ∗ αβ ψ̃q − βpω ψ̃d − r̃σs i∗q .
This result is similar to [9] with the additional r̃s -dependent In Fig. 1 these relations are represented graphically for the
term, hence it follows that the estimated speed is affected IM used in the testing bench considered in Section IV. It
by flux tracking errors and stator resistance estimation follows that, at constant r̃s , the effect on ω̃ and ĩd depends
error. Perturbation due to flux error is compensated by on the IM operating condition, i.e. on reference speed and
proper design of ν0 and γ1 , in order to impose asymptotic load torque. Major sensitivity is present at low speed and
stability of the electromagnetic subsystem (7), while stator in regenerating mode, when instability may arise.
resistance mismatching will be compensated by a proper From (13) and Fig. 1 it follows that ĩd is proportional to r̃s
adaptation law. with larger gain at higher torques, i.e. with high |i∗q | and
Thanks to time-scale separation between control and es- |ω2 | = |ω0 − pω|, and at lower excitation frequency |ω0 |.
timation dynamics, the flux-speed error dynamics can be Based on (13), the following stator resistance adaptation
approximated by its quasi-steady state version, obtained law is designed:
substituting (10) into (7)-(8). It is expressed as
r̂˙s = −fs ηs n̂r (t)sign(dˆr (t))ĩd
ĩ˙ d = −kid ĩd + αβ ψ̃d + βpω ψ̃q − r̃σs id
n̂r (t) = αi∗q + (ω0 − pω̂)id (14)
˙
ψ̃d = −αψ̃d + ω̄2 ψ̃q + αLm ĩd i∗
∗ dˆr (t) = σ kid ω0 + α pω̂ + αLm ψq∗
˙ ¯ + αLm iq∗ ĩd + γ1 pω̂
ψ̃q = −ω̄0 ψ̃d − β1 r̃σs i∗q − β1 pω̂ ψ
¯ ĩd
(11) where ηs is a positive constant tuning gain and fs (·) is the
1265
10 the r̂s updating with small values of ĩd and i∗q , when sensi-
Speed error [rad/s]
90%rs
tivity to r̃s of the proposed sensorless solution is actually
5 95%r
s negligible. Differently, the saturation is added to guaran-
0 tee that the rate of convergence of r̂s is slower than the
−5 speed-flux dynamics.
−10
It is worth noting that to guarantee the time scale sepa-
rations and to track the stator resistance thermal variation,
−150 −100 −50 0 50 100 150
a slow convergence rate is required. On the other hand, the
on-line estimation must be sufficiently fast to track the sta-
d−axis current error [A]
0.1 90%rs tor resistance variation occurred when the adaptation law
0.05 95%rs is disabled. Hence, a suitable trade-off in designing ηs and
fM is required.
0
Guidelines in selecting parameters are the following. The
−0.05
dead-zone amplitude fdz is basically related to current mea-
−0.1 surement noise. The saturation limit fM is set to impose
−150 −100 −50 0 50 100 150 a sufficiently small variation of the stator resistance along
Reference speed [rad/s] a time interval comparable to the rotor time constant in
every working condition, hence the following rule can be
Fig. 1. Dependence of ω̃ and ĩd on r̃s at steady-state, for different
values of ω ∗ , with TL = 7 Nm and ψ ∗ = 0.86 Wb for r̂s = 90%rs given: fαM < 5%rsN . Thresholds i∗qM , ω0M are selected in
and r̂s = 95%rs . For ω0 close to zero, errors are not shown since order to guarantee that whenever the adaptation law is en-
equilibrium point is unstable. abled, for a conceivable maximum speed estimation error
it holds |nr − n̂r | < 12 |n̂r | and |dr − dˆr | < 12 |dˆr |; in this
following saturation function with dead-zone: way, nr , n̂r and dr , dˆr have the same sign as underlined
before. The adaptation gain ηs is selected following this
0 if |y| ≤ fdz rule derived from (13), (14): ηs < ασ(α+k id )ω0M
, in order
4α2 i∗2
fs (y) = min(y − fdz , fM )
qM AX
if y > fdz to obtain a r̃s dynamics slower than the flux dynamics in
max(y + fdz , −fM ) if y < −fdz typical working conditions.
and fdz , fM are the dead-zone amplitude and the satura- IV. Experimental results
tion limit respectively. The adaptation law (14) can be eas-
ily derived adopting the Lyapunov-like function V = 12 r̃s2 The proposed speed sensorless algorithm has been tested
applying standard Lyapunov design arguments. Note that using a 1.1 kW induction motor, whose data are reported
the adopted V takes into account the r̃s state only, since in Table I. The experimental set-up is equipped with a
the steady-state behavior of (11), (12) is considered accord- rapid prototyping station including:
ing to the hypothesis of slow stator resistance adaptation. • a 50 A/380 V three phase inverter operated with a sym-
From (13), (14) the time derivative of V is metrical PWM technique with 10kHz switching frequency,
dead-times set to 1.5 µs and compensated in the current
V̇ = −r̃s fs ηs n̂r sign(dˆr ) ndrr r̃s (15) control loop;
• a custom floating-point digital signal processor (DSP)
which is negative semidefinite if sign(n̂r ) = sign(nr ) and board, based on TMS320C32 and installed on a PC used
sign(dˆr ) = sign(dr ). It is worth noting that nr , dr in (13) as operator interface. The DSP board performs data ac-
are replaced by their estimations since they are not com- quisition (two stator currents, dc-link voltage and rotor
pletely known. In particular, terms dependent on speed speed, by means of a 512 ppr incremental encoder and
estimation error (pω̂ − pω) have been neglected. By the only for monitoring purposes), implements the control al-
way, according to V̇ , this solution is effective only when gorithm and generates the PWM signals for the inverter
n̂r , dˆr are sufficiently close to their real values. Noting actuation. The discrete-time version of the control algo-
that n̂r and dˆr are approximatively proportional to i∗q and rithm (3)–(6), (14), obtained using the Euler method, and
ω0 respectively, the adaptation law (14) is enabled only the On/Off adaptation supervisor according to (16) have
if these values are large enough, i.e. when the following been implemented with 200 µs sampling time;
condition is satisfied: • the 1.1kW induction motor;
• a vector controlled permanent magnet synchronous mo-
(|i∗q | ≥ i∗qM ) and (|ω0 | ≥ ω0M ) (16)
tor used to provide load torque and load inertia.
where i∗qM , ω0Mare positive constant thresholds, otherwise Controller and estimation parameters have been set accord-
r̂s is not updated. ing to tuning rules given in Section III as follows: current
The saturation/dead-zone function fs has been added for gains are kid = 350, kiq = 500, speed controller gains are
the following reasons. The dead-zone is useful to increase kω = 120, kωi = 7200, observer gain is kio = 125000, tun-
robustness with respect to measurement noise, stator volt- ing gain is set to γ1 = 25.9. With the selected gains the
age distortion and other model uncertainties. It prevents speed control and speed estimation dynamics show a time
1266
Rated power 1.1 kW Rated speed 1410 rev/min a) Speed reference and load torque profile b) Est. stator resistance
Rated torque 7.0 Nm Rated frequency 50 Hz 20 12
11.5
15
Pole pairs 2 Rated voltage 220 V 11
rad/s, Nm
10
10.5
Rated current 2.8 A Excitation current 1.4 A
Ω
5
10
0 9.5
Rs = 10.3 Ω Rr = 5.2 Ω Lm = 0.434 H −5 9
rad/s
rad/s
15 30
10 20
5 10
0
0 0.5 1 0 0.5 1
constant of τω = 17 ms and τo = 4 ms respectively, with e) d−axis current reference f) q−axis current reference
imposed time scale separation expressed by = ττωo = 0.25. 4
3
4
3
2 2
Gains kid and γ1 are selected considering the trade-off be- 1 1
0 0
A
tween bandwidth increase of the d-current loop and sensi- −1 −1
−2 −2
tivity to noise superimposed on ĩd . Stator resistance adap- −3 −3
eters have been set as follows: fdz = 0.1 Ω/s, fM = 4 Ω/s, 0.3
g) d−axis current error
0.3
h) q−axis current error
A
mode, which represent the most critical operating condi- −0.1 −0.1
−0.2 −0.2
tions of sensorless controlled IMs, have been performed to −0.3 −0.3
0 0.5 1 0 0.5 1
study the performance of the proposed controller. time (s) time (s)
10
stant speed reference equal to ω ∗ = 15 rad/s. The stator 10.5
Ω
5
10
0
resistance adaptation law is removed and the estimated sta- −5
9.5
9
tor resistance is maintained constant at r̂s = 9.5 Ω (about −10
0 0.5 1
8.5
0 0.5 1
92% rated value). The estimated speed tracking error con- c) Speed ref., speed and est. speed d) Excitation freq.
30 60
verges to zero after the transient at t = 0.2 s, nevertheless 25 50
the speed tracking error diverges, even leading to instabil- 20 40
rad/s
rad/s
15 30
ity if this condition persists. This is due to the uncorrect 10 20
value of r̂s , as enlightened also by the non-null d-axis cur- 5 10
0
rent tracking error when load torque is applied, which is in 0 0.5 1 0 0.5 1
e) d−axis current reference f) q−axis current reference
accordance with (13). 4 4
3 3
In the second test, whose results are reported in Fig. 3, 2 2
1 1
the IM is controlled under the same operating conditions of 0 0
A
−1 −1
the first test, with the stator resistance adaptation starting −2 −2
−3 −3
with initial value r̂s (0) = 9.0 Ω. According to enabling −4
0 0.5 1
−4
0 0.5 1
condition (16), the stator resistance is adapted after load
torque is applied, i.e. for t ∈ [0.2, 1.2] s, when i∗q ∼
g) d−axis current error h) q−axis current error
= −2.8 A. 0.3
0.2
0.3
0.2
From Fig. 3 it follows that r̂s is adapted to the rated value 0.1 0.1
0 0
A
in about 0.4 s, the speed and d-axis current tracking errors −0.1 −0.1
TL
converge to zero and the q-axis current is close to µJψ ∗, −0.2 −0.2
−0.3 −0.3
enlightening correct flux orientation. 0 0.5 1 0 0.5 1
time (s) time (s)
In order to analyze the behavior of the stator resistance
estimation also during fast variable load torque, a square Fig. 3. Experimental results with ω ∗ = 15 rad/s and stepwise load
wave load torque with amplitude TL = ±7 Nm and period torque TL = −7 Nm (the same condition of Fig. 2) with rs adap-
tation law activated.
equal to 0.3 s is applied. In Figs. 4 and 5 it is shown that
1267
a) Speed ref., speed and load torque profile b) q−axis current (real and ref.) a) Speed ref., speed and load torque profile b) Est. stator resistance
30 4 30 12
25 3 20 11.5
20 2
rad/s, Nm
11
rad/s, Nm
15 1 10
10.5
10 0
Ω
0
5 −1 10
−10 9.5
0 −2
−5 −3 −20 9
−10 −4 −30 8.5
0 0.5 1 0 0.5 1 0 2 4 6 8 0 2 4 6 8
c) Est. stator resistance d) d−axis current error c) d− and q−axis current d) Excitation freq.
11.5 0.3 4 75
11 0.2 3
50
2
10.5 0.1 25
1
rad/s
Ω
10 0
A
0 0
A
9.5 −0.1 −1 −25
−2
9 −0.2 −50
−3
8.5 −0.3 −4 −75
0 0.5 1 0 0.5 1 0 2 4 6 8 0 2 4 6 8
time (s) time (s) time (s) time (s)
Fig. 4. Experimental results with ω ∗ = 15 rad/s and stepwise load Fig. 6. Experimental results during speed transient from 20 rad/s
torque TL = −7 Nm, with rs adaptation law. a) Speed reference to −20 rad/s with 7 Nm rated torque. a) Speed reference ω ∗
ω ∗ (marked), speed ω (solid), load torque TL (dashed); b) q-axis (marked), speed ω (solid), load torque TL (dashed); b) estimated
current iq (solid) and reference i∗q (marked); c) Estimated stator stator resistance r̂s ; c) d-axis current id (marked) and q-axis
resistance r̂s ; d) d-axis current tracking error ĩd . current iq (solid); d) excitation frequency ω0 .
a) Speed ref., speed and load torque profile b) q−axis current (real and ref.)
30
25
4
3
back-emf signal, represents a faster dynamics with respect
20 2
to the speed-flux tracking one, while the stator resistance
rad/s, Nm
15 1
10 0
A
10 0
A
1268