Professional Documents
Culture Documents
AbstrPct- In this paper the theoretical analysis, simulations to the transitions as possible in order to cancel the
and practical results of a new speed and disturbance torque quantization, but as it cannot be completely suppressed due
observer are proposed. The analysis of the nonlinearity to hardware limitations, the residual measurement noise is
introduced by the encoder quantization allows improving the incorporated to the optimal observer gain calculation.
behaviour of the fdter in all the speed range. A Kalman fdter This filter can be viewed as an asynchronous multirate
type observer has been developed. The main novelty is that the
state estimate observational update takes place at the observer executed periodically, without the need to use
quantized position signal discontinuities. Unlike the usually interrupts that makes the implementation difficult. The
applied Kalman fdter, the measurement noise covariance is discrete time Kalman Filter equations are updated in virtual
variable. In this observer, it is a function of the shaft speed sampling instants characterized by their best noise
because the selection of the sampling instant results in a measurement properties. This is a simple method to take
reduction of the output noise covariance. The observer adapts advantage of all the information provided by the position
its gain depending on the state variables. The optimal gain transducer. Its implementation only requires an additional
calculation results in an observer bandwidth increase. counter and two-plant discretization every period. The
Another advantage of this observer is that the self-sustained
observer bandwidth increase can be up to ten times higher
oscillations are suppressed, obtaining asymptotic stability.
in our case.
I. INTRODUCTION Robust performance with inertia variations has been
extensively studied [l 11 [ 121. In this study, we suppose that
The encoder is the position transducer most widely used the inertia remains constant or that its variations can be
in industrial applications. The speed regulation on a wide identified fast enough.
range is fundamental in the drives control for industrial
applications as tool machines, robots or electrical vehicles. 11. THEMECHANICAL
MODEL
The ordinary control uses the information of the average
speed [1] for feedback. However, when a low-resolution The model of the mechanical part in Fig. 1 is the
encoder is used, a delay that causes instability in the classical third order state space model. The disturbance
feedback loop compels to decrease the controller gain. torque includes all type of external torques, including
Speed and disturbance torque observers have become an viscous fiiction. If friction torque model exists, it can be
included. The motor torque of the field oriented controlled
effective technique of calculating the instantaneous speed
using Luenberger-type observers [lo] [2] [3] or Kalman induction motor is
Filters [8] [9].
When the elapsed time between encoder pulses is very
large, the filter applied in its standard form can cause
oscillations in steady state. A projection observer that is run
at the instants in which the transitions of the quantized
output occurs was developed in [5]; it is in those instants I Tdis
when the measurement is fkee of quantization noise. It is
possible to reduce the quantization effect by reducing the
filter bandwidth at low speed, however it is difficult to tune
it correctly for a wide speed range and assure the observer
stability. In [6] Sakai attempted to improve the behaviour at
low speed adding static fiiction compensation to the Tm
observer. Yang [7] used an interpolation method of the
position using the average speed with the drawback that it
introduces an additional delay producing oscillations in the
transient.
In this paper, a new observer whose main property is the
encoder nonlinearity cancellation is proposed. The aim is to
achieve asymptotic stability at very low speed and to
increase the global dynamic performance in all the speed Fig. 1. Mechanical system and observer
range. The observer output equations are updated as close
0 2002 EEE.
0-7803-7262-x/Ml$10.00 540
Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on December 14, 2008 at 00:31 from IEEE Xplore. Restrictions apply.
State equation of the continuous system:
f = Ax+ Bu + w
y = cx+v
c=(o 1 0)
1) f 1)
-
0 0 -
J" J"
(2)
0 0 0 0
ImC)
Fig. 2. Discrete update of the projection observer
State equation of the discretized system:
x,,, = Akxk+ B,u + wk State estimate extrapolation
y=Cx+v, i=Ai+Bu Vtgt, (5)
A =eATS
State estimate observational update
inw
ti-K(y-y,) V t e tk (6)
where K = M-'CT ( CM-'CT)-I
In Fig. 2 the state space is the direct sum of Im(CT) and Error covariance extrapolation
N(C). Given a certain positive matrix M the vectors
belonging to the subspace Im(M'CT) are orthogonal to
e = Ak-,pk-,Ak-,T+ GQk-,GT
those of N(C). The orthogonal projection matrix on the
vector subspace N(C) is State estimate observational update
Zk t ik+ K k ( Z k -ak) (9)
(
P = I - M--'CT(CM-'CT ) I c) (4)
Error covariance update
541
Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on December 14, 2008 at 00:31 from IEEE Xplore. Restrictions apply.
IV.THEPROPOSED
OBSERVER STEP2: MeasurementUpdate
IF (New pulse OR Position error > %lax),
At low speed, the behaviour related to the position in THEN {
stationary state is shown in Fig. 3. The speed and Error covariance update (11).
disturbance torque estimates become oscillatory due to the Kalman Gain Matrix (10)
encoder quantization. The output of the observer grows State estimate observational update (9) 1
quickly after a new pulse arrives, and decays immediately
afterwards to maintain the fixed quantized position, while STEP3: Tz - Extrapolation
the actual position follows a different path. {Discrete Matrix Calculation using T2 (3)
State Estimate Extrapolation (7)
It can be observed the similarity between equations (6) Error covariance Extrapolation (8)}
and (11). The only difference is the presence of
measurement noise. The fundamental idea that underlies A. High-Frequency Counter
this study is that if the delay is sufficiently reduced, the
observer will approach to the projection observer ( R & I ) In the new observer, the time that elapses from the last
eliminating in practice the nonlinearity. This is encoder pulse until the present sampling instant is
accomplished projecting (updating the measurement) near employed to improve the estimate of the states. In order to
the encoder pulses. If there is no output feedback for many measure this time an additional counter is necessary. This
sampling periods, the estimated position can be greater than counter is reset with encoder signal edges and is latched
one encoder line. Then the state estimate is updated with together with the position up/down counter at a constant
the constant position, adjusting the Kalman gain to the period T,. The timer width W must comply with
standard filter one. Each sampling period T, the following is 2w 2 f,T, ; e.g. in simulations a high frequency clock f m
run:
=1.25 MHz is used under a Ts=200 ps sampling period.
STEP1: TI - Extrapolation Therefore, the timer should have at least eight bits. Calling
{DiscreteMatrix Calculation using TI (3) n to the output of the counter, I; = ( N - n )THFit is the time
State Estimate Extrapolation (7) since the previous sampling until the arrival of the last
Error covariance Extrapolation (8) } pulse before the current sampling, and T, = nT, is the time
since the last pulse.
0 . 1
0.2
4 r .
0.22
,
0.24
,
0.26
.
0.28 0.30
,
0.32
1 is reduced to e(@THF)=dHF,preserving the limit
corresponding to the angle of a line e - . Concerning the
distribution of the quantization noise, it has a uniform
time($ distribution with covariance R=ez/12. In Fig. 5 the
dependency of R with the speed, which is recalculated each
Fig. 3. Quantized position measurement Q[O] and observed position, Om time a new pulse amves, is represented.
using the standard Kalman filter
6
IO
Standard Kahnsn Filter f
IO
.I I
I I
0 1 4
10 IO 10
S& Speed (rpm)
Fig. 5. Measurement Covariance (R)vs. shaft speed for several high
Fig. 4. Time magnitudes employed in the proposed observer kquency clock values V;IF)
542
Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on December 14, 2008 at 00:31 from IEEE Xplore. Restrictions apply.
101 ..................... Fig. 6 and Fig. 7 show how the optimal gain calculation
results in an observer bandwidth increase, which can be up
to ten times higher, depending on the rotational speed.
v. SJMULATIONRESULTS
Simulations using Matlab for different ranges of shaft
speed, contrast the performance of the proposed observer
with the standard one. Speed and disturbance torque
estimates are compared with real variables.
....................
IO1 '01 Frcquecy (radts) lo' 10' TABLE I
SMULATION PARAMETERS
Fig. 6. Bode graph of estimateddisturbance toque (Td CSJTdieol). Total Inertia (Jn): 0.012 Kgm'
Encoder Resolution (PPR): 4096
?{ -
7 Sampling Time (Ts): 200 p
High frequency clock (THF): 833 ns
Disturbance Torque Peak (Tdis) 0.1 N.m
0.8
0.6
0.4
0.2
0.15
0.1
P
$ 0.05
P
c - 0 0
8
B
-
f
%
-0.05
5 -0.1
I
0
I
0.1
l
0.2
l
0.3 0.4 0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4
1
.
0
..
1
0.1 0.2 0.3 0.4
time (s) time (s) time (s) time (s)
Proposed Filter Standard Filter Proposed Filter Standard Filter
a) Medium speed range , b) Low speed range
543
Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on December 14, 2008 at 00:31 from IEEE Xplore. Restrictions apply.
encoder pulses to "up/down" and "clock" inputs to the first
counter and the reset signal for the second. Synchronization
of the two counters is essential in order to obtain correct
measurements. This configuration allows us to measure the
average speed with the M/T method [13 without the need of
extra logic.
TABLE XI
PRACTICAL hfPLEMENTATION PARAh4ETERS
544
Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on December 14, 2008 at 00:31 from IEEE Xplore. Restrictions apply.
[4] M. Grewal, A. Andrews “Kalman Filtering, Theory and practice”.
Prentice Hall, 1993.
[5] J. Sur, B. Paden, “State observer for linear time-invariant systems
with quantized output”. Trans. ASME. Jour. of dynamic systems,
20 measurement and control. Vol120, pp. 423-426
[6] S. Sakai, Y. Hori, “Ultra-low speed control of servomotor using low
resolution rotary encoder”, IECONproc. 1995, p p . 615-620.
15
h [7] S. Yang, S. Ke, “Performance evaluation of a velocity observer for
s accurate velocity estimation of servo motor drives”. U S conj 1998,
110 pp. 1697-1702.
[8] T. J. Kweon, D. S. Hyun, “High-Performance speed control of
z5 electric machine using low precision shaft encoder”. IEEE Trans.
Power Electronics, vol. 15, no 5, Sep. 1999, pp. 838-849
[9] H. W. Kim, S. K. SUI, “A new motor speed estimator using Kalman
filter in low speed range”, IEEE Trans. On Industrial Electronics,
0 Vol. 43, No 4, Aug. 1996 pp. 499-504
[lo] D. G. Luenberger, “An introduction to observers”, IEEE Trans. On
Automatic Control, vol. 16, N”6, 1971, pp. 596-6
-51
0
’
0.1
’
0.2
’
0.3
’
0.4
‘
0.5
’
0.6
’
0.7
‘
0.8
I
0.9
I
1 [ l l ] N. Kim, H. Moon, D. Hyun, “Inertia Identification for the speed
observer of the low speed control of induction Machines”, ZEEE
(a) Estimated Speed and MIT average speed. Trans. On Industry Applications, vol. 32, N”6, 1996, pp. 1371-9
[12] M. Iwasaki, N. Matsui, “Robust speed control of IM with torque
feedfomard control“,IEEE Trans. On Industrial Electronics, vol. 40,
2.5
N”6, 1993, pp. 553-60
2
1.5
2 1
20
0.5
c
i o 18
a
p
0
-05
-1 5
7
-I
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
time@) 6
(b) Disturbance torque (Measured and Estimated) 4
VII. CONCLUSION
REFERENCES
0.49
I
0.5
I
0.51 0.52
I
512 PPR
0.53 0.54
1
Y. Hori, “An instantaneous speed observer for high performance
control of DC servomotor using DSP and low precision shaft (b) Zoom at low speed
encoder”, Proc. EPE, 1991, vol. 3, pp. 647-652
Y. Hori, “Robust and adaptive control of a servomotor using low
precision shaft encoder”, Proc. IECON’93 vol. 1, pp. 73-78 Fig. 12. Estimated and M/T average speed.
545
Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on December 14, 2008 at 00:31 from IEEE Xplore. Restrictions apply.