Variable sampling period IM vector control for high performance medium

voltage drive

F. Terrien, S. Siala and G. Flury

CONVERTEAM
3, avenue des trois chênes
90018 BELFORT Cedex, FRANCE
tel: +33/(0) 3 84 55 64 50
fax: +33/(0) 3 84 55 19 49
[email protected]/[email protected]
[email protected]

Keywords
"Vector control", "Induction motor", " Variable speed drive", "Industrial applications"

Abstract
This paper deals with the control of high performance Induction Machine vector control working with
variable sampling period. In order to optimise the efficiency of a 6/8MW 3 level medium voltage
inverter, used in Zendzimir mill steel (Zmill) applications, the PWM frequency is varied to minimise
the switching losses. The experimental results on 6MW tension reel drive will be presented..

Introduction
Medium voltage drive are usually used at the maximum of their performances. For large adjustable
drive, the switching frequency and the PWM type (Space Vector, H3, GDPWM,…) can be chosen
according to the speed and the current level in order to [3],[4],[7]:
• Optimise the switching losses between the inverter components at low speed, and avoid the
low speed de-rating
• Optimise the waveform and minimise the THD
• Minimise the inverter losses for high supply frequencies

The Asynchronous PWM is used. The control sampling period is equal to the half of the PWM period.
This period is variable according to the PWM period. The vector control is based on the polynomial
regulators (RST) for the current control, and a Kalman Filter for the flux estimation. All the vector
control functions parameters are automatically re-computed when the sampling period changes. The
soft switching of the torque is very important for the Zmill application. An analytical digital state
space model for the induction machine is developed. The analytical equations of the RST parameters
are also developed. The speed varies in a large band from -3pu to 3pu, with a flux weakening of a ratio
2,5. The experimental result of the first application of this control on a Zmill application will be
presented.

ZMill overview
A Zendzimir mill steel is constituted by four converters, one active front end and three machine
bridges, each one is driving an induction motor. The converters are linked with their DC bus. Figure 1
shows an overview of the electrical installation.
The tension reel motors ensure the passing of the a steel coil and the stand motor has to pressure the
steel. The maximum speed for a coil can reach 800m per minute.
The stand motor is driven in speed and the tension reel motors in torque. As the tension reel motors are
linked mechanically together by the steel, When one machine, driving one coil is working in the motor
area the other one is working in the regenerating area.
The vector control structure used by each motor is strictly the same. As the tension reel motor torque
precision gives the global performance. It's control will be detailed in this paper.

Tension
6MW
reel
Machine
Bridge

Factory
network

8MW Stand
8MW Machine
AFE Bridge

6MW Tension
Machine reel
Bridge

Figure 1: Global electrical scheme overview

In order to achieve the required torque power on all the working points, without any over sizing of the
converter, the PWM frequency and the type of PWM is modified according to speed range. That
ensure an optimized using of PWM converter on all the working points.

Control synoptic
The vector control of figure 2 presents the overview induction vector control, that could be used with
or without encoder. This application uses an encoder [5],[6],[10].
The sampling period is an input for the control. The vector control parameter and the Kalman filter
matrix are computed on the fly. The sampling period could vary from 360µs to 1100µs. The PWM
pulses are computed by the same processor according to the sampling period. A Power Interface Board
(PIB) generates the inverter pulses.
The PIB is the electronic boards developed by Converteam to drive the converter (Analog and logical
input/output, pulses generation, over-voltage and over-current management, sampling period
management).

Sampling period PIB

generation VDC interface

EVC
I1
Iα Vα
PWM
3 to 2 To PIB
I2 currents reference
Vector
transform Iβ Vβ
ψR control computation
Flux
I3 estimator θR
using a ω
Kalman
filter

Speed
estimator

Encoder ωmeas

Figure 2: General control synoptic

The vector control design is presented in figure 3. It is developed in detail in [10]. The structure allow
an efficient management in limitation with an efficient anti-windup. The type of the three regulators is
RST.

PI

Φ Id current
Rotor Flux
×
ref V
d

Regulation Regulation
Φ
I d
V deq

Γref Vqeq
V ÷
Γref Iq current
÷ Regulation × V q

I
q

Figure 3: Vector Control synoptic in the dq frame

In order to achieve a good torque control when the sampling period switches on fly, some
improvement are added on this control. The main improvement deals with the exact discrete-model of
the induction machine, that makes possible the re-computing in the fly of IM model when the
sampling period switches. The second point concerns the controller design and polynoms parameters
computing on the fly.

On-line parameters computing

Efficient IM model in the α,β digital frame

An analytical discrete time model of induction machine is developed and implemented in the digital
controller. The goal is to compute, on every fast task, the state space matrix according to the shaft
speed and the sampling period T.
The discrete time state-space is given by:

 X [ k + 1] = Ad (ω m , T ). X [k ] + B d (ω m , T ).U [ k ]
 (1)
Y [k ] = C . X [k ]

The state vector is defined as:

[
X [ k ] = Ψ Sα Ψ Sβ Ψ Rα Ψ Rβ ] T
(2)

The input vector is defined as:

[
U [ k ] = V Sα V Sβ ] T
(3)

The output vector is defined as:

[
Y [ k ] = I Sα I Sβ ] T
(4)

Using a specific writing of the continuous analytical electrical induction motor model, allowing to
show particular bloc triangular matrix and to replace the blocs by their digital expressions, an efficient
digital model is calculate versus the speed and the sampling period.
By this accurate digital model, the induction motor state model is computed on-line and used by the
Kalman Filter [1],[2],[8],[9]. By this way, the sampling period variation is taken into account without
loss of precision or have transient effect on the flux estimation.

Figure 4 shows the simulation curves of a Zero order hold rotor flux obtained, by the continuous
model, and the rotor flux obtained by the discrete-time state space model defined above, at 90Hz with
a sampling period of 1ms.
The α and β rotor flux component are very close between the continuous model and the digital model,
validating the accuracy of the analytical digital model.

Figure 4: comparison of α,β rotor flux component

On-line parameters RST controller calculation

Based on the discrete characteristic polynomial equation (pole placement), expressing the dominant
poles polynomial in the general continuous form, two polynomials are defined for set the closed loop.
The filtering F ( s ) = s 2 + 2.ξ .ω .s + ω 2 and the auxiliary poles polynomial equation
as C ( s ) = ( s + ω 0 ) where ω 0 , ω are the signal tracking and disturbance rejection bandwidths set by
Tc and T f respectively.
In discrete-time frame, C(z) can be expressed as: C ( z ) = z − z 0 where, z 0 = e −ω 0T and T is the
sample period.
The two roots of the polynomial F(s) are − ξω ± jω 1 − ξ 2 .
F(z) can be written as:

F ( z ) = ( z − z1 )( z − z 2 ) (5)
2
− ξω T ± j ω T 1−ξ
where z 1 , z 2 = e .
Expanding and simplifying F(z) gives:
F ( z ) = z 2 − 2αβ z + α 2 (6)

where α = e −ξωT , β = cos(ωT 1 − ξ 2 ) .

So, the discrete characteristic polynomial of the closed loop system is given by:
P ( z ) = F ( z ).C ( z ) = ( z 2 − 2.α .β .z + α 2 ).( z − z 0 ) (7)

After writing the exponential expression in the complex plan according to the previous pole
placement, it is possible to develop an analytical expression for each parameter of the RST controller.
In this case, the parameters could computed on-line with the actual sampling period.

Vector Control improvement with sampling period switching

A new RST controller design

The classical RST controller design, presented in figure 5, is not efficient for switching of the
sampling period, that induces a prohibitive transient following a sampling period switching. Indeed, as
the Kalman Filter, the RST digital controller parameters are computed on-line according to the
required closed loop dynamics and the sampling period, based on the equations (5) to (7).

As example the equation of the reference parameter ‘h’ changes according to the sampling period. In
the discrete-time frame (where z-1 is the delay operator), the polynomials shown in figure 5 could be
expressed as :

S ( z ) − F ( z ) ( s1 − f 1 ) − ( f 0 + s1 + s 2 ) z −1 −1
= .z (8)
F ( z) 1 + f 1 z −1 + f 0 z − 2

R(z) ( r1 + r0 z − 1 )
= . z −1 (9)
F ( z ) 1 + f 1 z −1 + f 0 z − 2

r1 + r0
And the gain h = (10)
1 + f1 + f 0
As the parameters of the polynomials R(z) and F(z) change according to the sampling period, the "h"
parameter and the outputs of RST transfer functions also change. The output of the gain "h" changes as
a step, where the outputs of the two other transfer functions change with the F(z) transient time
response. The output of the comparator (Add) will temporary vary according to the F(z) transient time
response.

Figure 5: classical RST controller structure

In order to ovoid this transient behaviour , an alternative design of the RST controller Figure 6 is
proposed and implemented. The polynomial filtering function F(z) is moved to the RST output, that
reduces the impact of the gain variation and compensates any transient effect. Globally, the transfer
function of the RST controller is the same between for both versions, as shown in the figure7.

Figure 6: new RST controller structure to allowing sampling period switching

Figure 7: comparison of the two RST response when the sampling period switching

Computing delay compensation

In addition to RST structure modification, the Park's transformation angle has to involve some delay
compensation, according to the shaft speed, the actual sampling period and the next one.

Experimental results

Start up time response

Figure 8 shows the fluxing sequence of the induction machine. The Id response time is equal to 3
sampling period, equals to the set value. During this sequence the Id current is saturated at its
maximum values, and the flux increases linearly till the reference value. This high dynamic allows a
fast flux weakening in the over speed range.
 Flux regulation
Estimated flux enabled

Id current time
response (#1ms)

Figure 8: Rotor flux and Id time response

Experimental results with the Zmill in production

The first application of this control is commissioned on August 2006 for high performance Zmill.
Three sampling periods are used (833µs
1100µs
550µs). The time response of the current is 3ms
for the three sampling periods. Figure 9 shows the plots of the switching behaviour when the period
changes from 1100µs to 833µs during the decelerating stage.

Period
switching

Figure 9: Experimental result in production

The average values of d-q components are constant. Only the current ripples changes, due to the
modification of the period and the type of the PWM. The red curves (Mes_Traction) shows the
stability of the traction when the sampling period switches.

Conclusions
The proposed control gives to power electronic designer the flexibility to change the PWM frequency
and type, in order to optimise the MV inverter performances without any de-rating. The high dynamic
and precision of the control give a high accuracy that fulfil the contractual requirements (thickness
error <3µm) with an induction machine, where the synchronous motor was usually used for this type
of applications.

References
[1]. "Two efficient Kalman filters for flux and velocity estimation of induction motors" M. Hilairet, C.
Darengosse and F. Auger, PESC2000
[2]. "Adaptative flux observer for sensorless induction motor drives with enhanced dynamic
performance" J. Maes and J. Melkebeek, EPE 99
[3]. "Electrical propulsion - Active torque smoothing.", J. Courault and C. Nicolas, First International
symposium and Exhibition Civil or Military all Electrical Ship,AES97
[4]. "Analysis Of Double Star Motor Drives For Electrical Propulsion", F. Terrien, M.F. Benkhoris,
IEE: 9th Internation Conference Elecrical Machines and Drives EMD 99, Canterbury, UK; 1-3
septembre 1999; conference publication 468
[5]. "La propulsion électrique des navires", L.Mazodier, Revue de l'électricité et de l'électronique,
No.3, pp. 30-36, mars 1997
[6]. "Control of Electrical Drives" W. Leonhard, Editions Springer 1996
[7]. "Critères de choix pour les ensembles machine-convertisseurs de forte puissance utilisés dans le
domaine de la propulsion électrique des navires", M. Fracchia, L. Pierrat, EPF 96
[8]. "A speed adaptative Kalman filter observer for induction motors", J.L. Mora, A. Torralba, L.G.
Franquelo, EPE 2001
[9] “Comparison of three Kalman filters for speed estimation of induction machines“, Jie Li; Yanru
Zhong; Industry Applications Conference, 2005. Fourtieth IAS Annual Meeting. Conference Record of
the 2005 Volume 3, 2-6 Oct. 2005 Page(s):1792 - 1797 Vol. 3
[10] “ Robust sensorless induction motor control for electric propulsion ship ” F. Terrien and S. Siala,
EPE 2003 proceeding, 2-4 september 2003, Toulouse, FRANCE