Practical linear controllers for power electronics and drives

ảo-Huy Nguyễ
Bả ễna and João Pedro F. Trovãob, a CTI Lab. for EVs, School of Electrical and Electronic Engineering, Hanoi University
of Science and Technology, Hanoi, Vietnam; and b e-TESC Lab., Université de Sherbrooke, Sherbrooke, QC, Canada
© 2023 Elsevier Inc. All rights reserved.

Fundamentals of controller design 268

Basic concepts 268
Controller design principle 269
Small- and large-signal methods 270
Motivation: nonlinearity of bidirectional DC/DC converter 270
Small-signal composition method 271
Large-signal decomposition method 272
Methodology 272
Voltage controller design 273
Current controller design 274
Design example 274
Techniques for performance improvement 276
Motivation: speed control of electrical drives 276
Integral-proportional (IP) controller 277
Disturbance observer 279
Conclusion 281
References 282

Abstract

Power electronics and electrical drives are large application fields of automatic control. Even though all these systems in real
world are nonlinear, the linear methods are useful for control engineers due to their systematic approach and mature
development. This article presents several practical linear control techniques illustrated for regulating a DC/DC (direct
current) boost converter and a permanent magnet DC motor. Despite that the applied systems are simple, they are standards
for understanding, analysis, modeling, and controlling power electronics and drives. Small- and large-signal methods are
developed for the bidirectional converter to highlight the interest of the large-signal decomposition approach. Then integral-
proportional (IP) scheme and disturbance observer (DOB) are presented as techniques for performance improvement
examined in the motor speed control application. Examples with simulation results are given to demonstrate the advantages
of the presented methods. The article could be of interest for students and practical engineers who need to regulate power
converters and motor drives.

Fundamentals of controller design

Basic concepts
This section presents the basic principles of controller design for dynamical linear systems. For that, some brief definitions should be
given, not in strictly theoretical expressions but by an intuitive approach.
A control system, as illustrated in Fig. 1, is a system that aims to regulate the output y of a plant P to follow a reference r. The
controller computes the control input u of the plant so that the error e between r and y is minimized (ideally becomes zero). The
input d is the disturbance from the external environment imposed on the system.
In order to design the controller, the mathematical relationship between u and y should be conducted. That is the modeling step
to deduce the mathematical model of the plant. In physical systems, there is always a delay between the input and the output of the

Fig. 1 A typical control system.

268 Encyclopedia of Electrical and Electronic Power Engineering, Volume 3 https://doi.org/10.1016/B978-0-12-821204-2.00028-3

 Practical linear controllers for power electronics and drives 269

plant. If the delay is not negligible, the mathematical model is a set of differential equations of time called a dynamical model. If the
differential equations are linear, the model is a linear system. Although most of the physical behaviors in reality are nonlinear, many
plants can be modeled as linear systems for convenience in analyses and control.
In a dynamical model, the variable x that is currently dependent not only on the input u but also on its own previous value is the
state variable. In a system, some states can be measured to give the output y and the other may be unmeasurable. If unmeasured
states need to be controlled, they can be calculated from a state observer. In this article, it is assumed that all the states are measur-
able and the measurement gains are unit, i.e., y ¼ x.
Controlling a plant is, in fact, to form a closed-loop system which has a new dynamical model that behaves in the way that the
control engineer expects. Controller design is, therefore, to develop a set of control equations that combines with the plant model
equations to produce the desired new mathematical model of the closed-loop system. Unfortunately, it is often difficult to directly
deal with the differential equations due to their complexity, especially in practical applications. To tackle this complexity, the differ-
ential equations are often transformed into the other representations of transfer function (in frequency domain) or state space (in
time domain) which can be dealt with by using algebraic tools.

Controller design principle

For illustrating the controller design principle, we consider a single-input single-output (SISO) plant of which the transfer function
is P(s). The plant is regulated by a controller C(s) in a feedback loop as portrayed in Fig. 1. The closed-loop transfer function of the
control system is given by:
yðsÞ
GðsÞ ¼
rðsÞ

CðsÞPðsÞ
¼ : (1)
1 þ CðsÞPðsÞ
The objective of the control system is that the output y(s) follows the reference r(s), i.e., GðsÞ/1. Fig. 2 illustrates the represen-
tative responses of a control system in time domain. Without a controller, the uncontrolled open-loop system (plant itself) cannot
follow the reference as shown in Fig. 2A. By contrast, Fig. 2B shows an ideal case when the system perfectly responds to the step
reference. It can be seen from (1) that the gain of the controller C(s) must be infinite in order to obtain this ideal response which
is unrealistic in real world.
A typical response is given in Fig. 2C, in which the output oscillates with a damping to converge to the reference. Some
commonly used control performance indices are depicted. The overshoot yp1, in per-unit (p.u.) system, is determined based

A B

C D

Fig. 2 Illustration of control system response.

270 Practical linear controllers for power electronics and drives

on the maximum value yp that the output goes beyond the reference. The output then converges within a band of 2% of the refer-
ence. The time it takes for this convergence is called settling time tsett0. In steady state (theoretically t/N, i.e., s/0), the output
should be equal to the reference. The difference between them, if any, is called steady state error. A high-performance response can
be illustrated in Fig. 2D where the settling time is small, there is no (or very small) steady state error, and there is no (or small)
overshoot.
Designing a controller is to consider a plant that responses as Fig. 2A, then to develop a control scheme so that the system
responses as Fig. 2C, then synthesis the controller parameters targeting high-performance response as Fig. 2D. Let us consider
the controlled system (1). The controller can be designed by forming the closed-loop transfer function G(s) as a standard transfer
function Gstd(s), i.e.,
GðsÞ ¼ Gstd ðsÞ: (2)
The performance indices can be determined based on the characteristics of Gstd(s). The controller parameters are then calculated
as a function of the plant P(s) (system property) and the standard transfer function Gstd(s) (desired performance). In this article, the
first-order and second-order standard transfer functions are used to develop simple-but-effective linear controllers for several power
electronic and motor drive applications.

Small- and large-signal methods

Motivation: nonlinearity of bidirectional DC/DC converter
A bidirectional boost DC/DC converter is considered with the topology given in Fig. 3. The source supplies the voltage us to the
inductor which is connected to the low side of the chopper. The high side is associated with the DC link capacitor which is in parallel
with the load considered as a current source.
Considering that the inductor has an inductance L and a parasitic resistance r, the induced voltage on the inductor is given by:
dis ðtÞ
us ðtÞ  uch ðtÞ ¼ L þ ris ðtÞ: (3)
dt
The chopper voltage uch is modulated by switching the chopper with the modulation function m(t) as follows:
uch ðtÞ ¼ mðtÞudc ðtÞ (4)
where udc is the DC link voltage associated with the capacitor as:
dudc ðtÞ
C ¼ ich ðtÞ  il ðtÞ (5)
dt
in which, ich is the chopper current and il the load current. The chopper current can be deduced by balancing the power between the
low side and the high side of the chopper:
ich ðtÞ ¼ mðtÞis ðtÞ: (6)
Here, we assume that the chopper losses are neglected and the circuit parameters r, L, and C are constants.
Combining (3)–(6), we have the system model in state space as follows:
dudc ðtÞ 1 1
¼ mðtÞis ðtÞ  il ðtÞ
dt C C
(7)
dis ðtÞ 1 r 1
¼  mðtÞudc ðtÞ  is ðtÞþ us ðtÞ
dt L L L
in which, m(t) is the control variable, udc(t) and is(t) the two state variables, and il(t) and us(t) the two disturbance inputs of the
system. Looking at the products of the control and state variables, it can be seen that the system model is nonlinear because of the
violation of the superposition principle Kuo (1991).

Fig. 3 A bidirectional DC/DC boost converter configuration.

 Practical linear controllers for power electronics and drives 271

Small-signal composition method

Since the model (7) is nonlinear, in order to apply linear control techniques, the model should be linearized. Suppose that at an
equilibrium, the converter works at the steady-state DC bus voltage U dc , source current Is , and modulation ratio M. When the system
works around this equilibrium, i.e., the variations of the variables are small, its dynamics can be considered as linear relations of the
small-signal variables. This approximation step is the small-signal linearization, or can also be called linearization at an operating
point.
To perform the approximated linearization, the variables are considered as the combinations of large DC components and small
AC variations as follows:
mðtÞ ¼ M þ mðtÞ e
udc ðtÞ ¼ U dc þ e udc ðtÞ (8)
is ðtÞ ¼ Is þ eis ðtÞ
where the over line denotes the DC component and the tilde means AC variation. Replacing (8) to (7) and neglecting the
disturbances il and us, which are the uncontrolled inputs, we have:
  !
deudc ðtÞ 1
¼ e
M þ mðtÞ Is þ eis ðtÞ
dt C
   ! (9)
deis ðtÞ 1 r
e
¼  M þ mðtÞ U dc þ e udc ðtÞ  Is þ eis ðtÞ
dt L L

that can be rearranged as:

udc ðtÞ
de MIs Me Is 1
¼ þ is ðtÞþ mðtÞ e þ eis ðtÞmðtÞ
e
dt C
|{z} C C
|fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl} C
|fflfflfflfflfflffl{zfflfflfflfflfflffl}
DC small linear AC variation very small
component nonlinear AC
variation
(10)
deis ðtÞ MU dc  rIs M r U 1
¼  e udc ðtÞ  eis ðtÞ  dc mðtÞ e  e e :
u ðtÞmðtÞ
dt L L L L L dcffl{zfflfflfflfflfflfflfflffl}
|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflffl
DC component small linear AC variation very small
nonlinear AC
variation

The DC components do not affect the dynamics of the model hence they can be neglected. On the other hand, the nonlinear AC
variation can be eliminated due to their very small values. Consequently, the control-oriented approximated small-signal linear
model of the boost DC/DC converter can be deduced as:
deudc ðtÞ Me Is
e
¼ is ðtÞþ mðtÞ
dt C C
(11)
deis ðtÞ M r U
¼ eu ðtÞ  eis ðtÞ  dc mðtÞ:
e
dt L dc L L
Due to its linearity, the model (11) can be regulated by applying well-known linear control techniques. For example, we may
write the model in the state space as follows:
2 3 2 3
2 3 M 2 3 Is
udc ðtÞ 7 6 7 u ðtÞ 6 7
0
d 6e 6 C 76 e dc 7 6 C 7e
4e 5¼6 74 e 5þ6 7mðtÞ: (12)
dt is ðtÞ 4 M r 5 is ðtÞ 4 U 5
   dc
L L L
Then, various state-space control methods can be utilized, e.g., pole placement or linear quadratic regulation (LQR) that can be
found in Ogata (2010) as a masterpiece in automatic control introduction.
The controller(s) can also be designed in the frequency domain by applying Laplace transformation on (11). Practically, one just
needs to replace the operator d/dt by s and transform the variables to the functions of the Laplace operator s. By that, the small-signal
model is written by:
M Is
udc ðsÞ ¼ eis ðsÞþ mðsÞ
se e (13a)
C C

M r U
u ðsÞ  eis ðsÞ  dc mðsÞ:
seis ðsÞ ¼  e e (13b)
L dc L L
272 Practical linear controllers for power electronics and drives

From (13b), we have:

 
eis ðsÞ Ls þ r ¼ Me U
u ðsÞ  dc mðsÞ
e
L L dc L
(14)
Me e
udc ðsÞ  U dc mðsÞ
5eis ðsÞ ¼ :
Ls þ r
Applying the above eis ðsÞ to (13a) leads to:
Me e
udc ðsÞ  U dc mðsÞ M Is
udc ðsÞ ¼
se e
þ mðsÞ
Ls þ r C C

2
M MU dc Is
¼ e
u ðsÞ  e
mðsÞþ e
mðsÞ: (15)
LCs þ rC dc LCs þ rC C
From that, we have
2  
LCs2 þ rCs þ M Is MU dc
e
udc ðsÞ e
¼ mðsÞ  : (16)
LCs þ rC C LCs þ rC
Eventually, we deduce the transfer function from the small variation of the modulation ratio to the small variation of the DC bus
voltage given by:
e
udc ðsÞ Is Ls þ rIs  MU dc
¼ 2: (17)
e
mðsÞ LCs2 þ rCs þ M
Normally, rIs  MU dc , hence the model is a non-minimum-phase second-order system. A common method to develop
a controller for this system is to use the Bode plot characteristics of the open-loop model and the desired closed-loop system. Inter-
ested readers may refer to Erickson and Maksimovic (2020) as a rigorous textbook of power converters analysis and control, espe-
cially the Bode plot method.
Small-signal linearization is a systematic approach to apply linear control method for regulating power electronics converters.
However, its essential drawback is that the system is approximated around a certain operating point. Hence, the desired perfor-
mance may therefore not be achieved for the system working in a wide-range condition. In the next part of this section, we introduce
a large-signal linearization method that is valid for the whole working space without any approximation.

Large-signal decomposition method

Methodology
An alternative approach is to decompose the plant model into subsystems that can be controlled without small-signal approximate
linearization as portrayed in Fig. 4. We can see that even though the whole system is nonlinear, the subsystem models, including the

Fig. 4 Block diagram of the DC/DC boost converter control system with chopper switching model (white: model; gray: control).
 Practical linear controllers for power electronics and drives 273

inductor, the chopper, and the capacitor are linear. Hence, this approach is a structural linearization method. The subsystems can
then be controlled by linear controllers considering the decomposed large-signal linear models instead of a composed approxi-
mated small-signal one. In the converter model, one thing should be noted that the initial value of the integrator, i.e., the initial
voltage of the capacitor, must be given. It can be set equal to the source voltage us.
Moreover, controllers can be considered as inversions of plant models. More details of this philosophy can be referred to Bous-
cayrol et al. (2013). Adopting this approach to the examined system, the closed-loop voltage controller is the inversion of the capac-
itor model which is here an integrator, in which, the control objective is the capacitor voltage udc. The controller computes the
current reference that should be imposed on the capacitor to obtain the DC bus voltage reference udc . The load current il is
then compensated to deduce the chopper current reference ich . After that, the modulation function m is taken into account to
compute the source current is . This is the inversion of the current relation in the chopper model (6). By using the information
of m, the nonlinear structure of the boost converter is linearized so that we do not need to conduct any approximated small-
signal linearization. One may notice here that in some simulation environments, e.g., MATLABÒ/SimulinkÒ, the feedback of m
may cause an algebraic loop. The solution is to put a dynamical delay, e.g., a z1 block, in the feedback line. In practice, when
programming microcontroller, a register may be used to store the previous value of the computed modulation function. This stage
is considered as an open-loop control (without feedback) of the chopper current since the chopper model contains no dynamical
delay. By contrast, a closed-loop current controller is required as an inversion of the inductor model, which is a first-order plant. The
control objective of the current loop is the inductor current is. The output of the current controller is a voltage reference that should
be placed on the inductor to produce the desired current is. This voltage should be compensated by the source voltage us to conduct
the chopper voltage reference uch . Finally, by dividing uch to udc, we have the modulation function m which is imposed on the PWM
block to generate the pulses S that switch the chopper.

Voltage controller design

After the structural decomposition, we just need to design a simple linear proportional-integral (PI) controller for the capacitor
model as an integrator. As its name suggests, a PI controller is composed by a proportional (P) part and an integral (I) one;
both components receive the control error as their inputs. When the error exists, the controller acts to reduce it by imposing the
control signal on the plant. When the error is zero at the steady state, the I component maintain the control signal to keep the system
at this desired state. That is why the integral action is often necessary to cancel the steady-state error. Here, one may see that since the
capacitor model is an integrator, only P controller is required to attenuate the error. However, this is true only if the disturbance is
absolutely rejected. In practice, it is difficult to ensure this condition, hence, a PI controller should be employed. On the other hand,
even though the PI scheme is used, disturbance compensation is still of interest because it can improve the control performance by
reducing the impact of the load change. In case of this application on controlling the DC/DC converter, it is effortless to measure the
load current il. Hence, this measured disturbance is compensated by adding to the control signal of the PI voltage controller as seen
in Fig. 4.
Applying (1), the closed-loop transfer function of the voltage control loop is deduced as follows:
  
kIu 1
kPu þ
s Cs
Gu ðsÞ ¼   
kIu 1
1 þ kPu þ
s Cs

kPu kIu
sþ
¼ C C : (18)
2 kPu kIu
s þ sþ
C C
To conduct the above closed-loop transfer function, we assume that the dynamics of the inner current loop can be neglected, i.e.,
the chopper current ich responds immediately to the reference ich . This assumption is valid if the response time of the inner loop is
much smaller than the outer loop as we will discuss when developing the current loop below. The control engineer must consider
this condition when designing the controllers.
The voltage loop transfer function Gu(s) has a form of a second-order system with a left half-plane (LHP) zero. This sort of zero
makes the system respond faster and causes an overshoot that is not a big deal for voltage control. Hence, we can design the closed-
loop system as a standard second-order system and let the LHP zero add its derivative effect.
The denominator (characteristic equation) of a second-order system has a form of:
DðsÞ ¼ s2 þ 2xun s þ u2n (19)
where un is the natural frequency and x the damping factor. The frequency un reflects how the response oscillates and the factor x
how the oscillation damps to the steady state. Fig. 5 illustrates the unit step response of a standard second-order system regarding
the different values of x with a common natural frequency un ¼ 2p, i.e., the oscillation period, if applicable, is 1 s.
274 Practical linear controllers for power electronics and drives

Fig. 5 Step response of a standard second-order system with un ¼ 2p.

Hence, one may choose the damping factor and the natural frequency to specify the desired performance of the closed-loop
system. It can be seen that the case of x ¼ 1 is the fastest response without overshoot and the system responds in approximately
one period of the oscillation, which is already damped. Hence, as a design recommendation, the natural frequency can be chosen
as:
2p
un ¼ (20)
Tr
where Tr is the desired response time of the closed-loop system.
Unifying (19) with the denominator of (18), the PI controller parameters are as follows:
kPu ¼ 2xu unu C
(21)
kIu ¼ u2nu C

in which xu and unu are calculated as above regarding the desired response time Tru of the voltage control loop.

Current controller design

The same approach can be applied to design the current PI controller as an inversion of the inductor model, which is a first-order
plant. The closed-loop transfer function of the current loop control system is:
  
kIi 1
kPi þ
s Ls þ r
Gi ðsÞ ¼   
kIi 1
1 þ kPi þ
s Ls þ r

kPi kIi
sþ
¼ L L : (22)
kPi þ r kIi
s þ
2
sþ
L L
The closed-loop current control transfer function has also a form of second-order oscillator with a LHP zero. Hence, similar to
the voltage controller, the PI current controller gains are given as follows:
kPi ¼ 2xi uni L  r;
(23)
kIi ¼ u2ni L;

where xi ¼ 1 and uni ¼ 2p=Tri are also recommended. One may see that the only difference of both model and control of the
capacitor voltage control loop and inductor current control loop is the parameter r. If r ¼ 0, i.e., no parasitic resistance of the
inductor considered, the inductor model is also an integrator like the capacitor model and the PI controller gains of (23) has
the identical form as the ones of (21).
Here, it should be noted that the current controller is at the inner loop whereas the voltage control is the outer meaning the
output of the voltage control loop ich is the reference to the current control loop. Hence, the current loop must respond much faster
than the voltage loop, i.e., Tri  Tru. It is often referred that the inner loop should be at least 10 times faster than the inner loop
Mathworks (2021).

Design example
Let us consider a sample converter (Fig. 3) with L ¼ 0.5 mH, r ¼ 0.1 U, C ¼ 2200 mF, modulated at 20 kHz, supplied by a 24-VDC
source, and supplying to a resistive load rl ¼ 20 U. The model and control of the converter are realized in MATLABÒ/SimulinkÒ
 Practical linear controllers for power electronics and drives 275

Fig. 6 Simulation results of the boost converter with the large-signal control method.

using only the basic Simulink blocks without any need of special toolboxes, e.g., SimscapeTM. The voltage loop and current loop
response time are designed as Tru ¼ 100 ms and Tri ¼ 5 ms. The simulation results are reported in Fig. 6.
To demonstrate the reference following and the disturbance cancellation performance of the controlled system, step changes of
udc and the current load il are applied. Moreover, the robustness of the control to the system uncertainty is examined by assuming
that there are up to 50% uncertainty of the parameters L, r, and C, and 5% of current measurement error.
276 Practical linear controllers for power electronics and drives

Considering that the converter is working at a no-load steady-state point with the output voltage of 36 V, at the time 0.1 s, the
reference udc steps to 48 V (Fig. 6A). The voltage udc responses to the new set point after about 0.1 s as the controller was designed.
The overshoot is caused by the LHP zero of the voltage loop transfer function. At the time 0.3 s, the resistive load is connected to the
DC bus that causes a step disturbance to the converter. The capacitor voltage therefore slightly drops and then quickly recovers to the
set point of 48 V thanks to the disturbance rejection ability of the voltage controller.
Fig. 6B shows the inductor (also the source) current response is to the reference is , which is generated from the outer voltage
control loop. Since is varies as a function of the voltage control behavior, the response time and overshoot of the current loop
are not directly shown. Instead, we can see that the current controller reacts well to the quick changes of the reference so that
the outer voltage loop performance can be guaranteed.

Techniques for performance improvement

Motivation: speed control of electrical drives
Beside power converters regulation, controlling of electrical machine drive is a fundamental application of automatic control. A
historical note: at its early stage, feedback control theory was mostly addressed as “servomechanism”. Despite various types of elec-
trical machines, DC motor control can be considered as the basic scheme of electrical drives. For example, the principle of field-
oriented vector control of AC machines, including induction and synchronous motors, is to transform the AC machines models
from the static frame to the synchronous rotating frame so that the models can be controlled in a similar manner as the DC motor.
Hence, DC motor control is essential for studying electrical drives. As a bibliography note, Nam (2017) is an excellent text that also
considers DC motor as “a standard model for electro-mechanical systems”.
The examined configuration of a permanent magnet (PM) DC motor drive is portrayed in Fig. 7. Since the magnetic field is
produced by the PM, only the armature equivalent circuit of the motor is addressed as a coil with the inductance La and the parasitic
resistance ra linked in series with the back electromotive force (EMF) em. The armature terminal is connected to a chopper, which is
supplied by a battery as a DC voltage source.
The chopper modulates the DC bus voltage udc to deduce the armature voltage ua given by:
ua ðtÞ ¼ udc ðtÞmðtÞ; (24)
where m(t) is the modulation function. This terminal voltage is imposed on the armature coil, interacting with the back EMF em, to
produce the armature current ia as follows:
dia ðtÞ
ua ðtÞ  em ðtÞ ¼ La þ ra ia ðtÞ: (25)
dt
The armature coil model is transformed into frequency domain as:
1
ia ¼ ðua  em Þ: (26)
La s þ ra
The electrical power, reflected by ia and em, is converted to the mechanical power of the motor and vice versa:
Tm ðtÞ ¼ kT ia ðtÞ;
(27)
em ðtÞ ¼ ke um ðtÞ;
where Tm and um are the electromagnetic torque and rotational speed, kT and ke the torque and back EMF constants of the motor,
respectively.
The motor torque Tm interacts with the resistant torque including the load torque TL and the viscous friction torque Tf to produce
the motor speed regarding the Newton’s second law:
dum ðtÞ
J ¼ Tm ðtÞ  TL ðtÞ  Tf ðtÞ; (28)
dt
in which,
Tf ðtÞ ¼ Bum ðtÞ: (29)

Fig. 7 A PM DC motor drive configuration.

 Practical linear controllers for power electronics and drives 277

Fig. 8 Block diagram of the PM DC motor control system with chopper average model (white: model; gray: control).

where B is the viscous friction coefficient. In some applications, Tf can be neglected because B is very small in comparison to the
moment of inertia J. By using Laplace transformation, the model can be written as an integrator:
1
um ¼ ðTm  TL  Bum Þ: (30)
Js
Even though the term Bum can be moved to the left-hand side of the above equation to form a first-order model of the shaft, in
this article, the viscous friction torque is considered as a disturbance to be compensated. Normally, the motor shaft starts from zero
speed; hence, the initial value of the integrator in this model can be set as zero.
The DC motor model is linear, thus, it can be controlled by linear control techniques as discussed in the previous session. By
using the decomposition large-signal approach, the model and control organization of a DC motor drive can be represented as
in Fig. 8. Similar to the large-signal control of the boost converter, the controllers of the DC motor are the inversions of their
subsystem models. The speed controller, as the inversion of the shaft model, imposes the motor torque reference Tm  on the inver-

sion of the field model. Since the current-torque relation is modeled by only a torque constant, the inversion is simply 1= kT that
deduces the current reference ia . The current controller then regulates the armature coil to make ia follow ia . The output of the current
controller is the desired armature voltage ua , which is modulated, as the inversion of the chopper model, by dividing to the DC bus
voltage udc.
Even though the system can work properly with traditional PI controllers as presented in the previous section, some techniques
can be applied to enhance control performance. In this article, we focus on the speed control loop due to the response performance
requirement and the presence of immeasurable disturbance.

Integral-proportional (IP) controller

As having a model in form of an integrator or a first-order function if considering the viscous friction, the shaft speed can be
controlled by a PI controller as addressed in the previous section. The closed-loop transfer function is therefore:
kPu kIu
J sþ J
Gu ðsÞ ¼ : (31)
s2 þ kPu
J sþ J
kIu

By that, the PI controller gains kPu and kIu can be calculated as:
kPu ¼ 2xu unu J;
(32)
kIu ¼ u2nu J;

where xu is the damping factor and unu the natural frequency of the speed control loop. Due to the LHP zero, the speed response
contains overshoot and requires more energy of the control variable input. To overcome this issue, the IP control scheme, proposed
by Harashima and Kondo (1982), depicted in Fig. 9, can be employed to cancel the zero.
278 Practical linear controllers for power electronics and drives

Fig. 9 Integral-Proportional (IP) scheme for motor speed control loop.

In fact, the IP scheme is a state-feedback inner loop combined with an integral output feedback outer loop to cancel the steady-
state error. For more details of state-feedback with integral control, the readers may refer to Kuo (1991). Following this scheme, the
inner loop transfer function is given by:
1
Js
Guin ðsÞ ¼
kPu
1þ
Js

1
¼ : (33)
Js þ kPu
The output of the state-feedback system (33) is regulated by an integral controller to form the following global speed control
loop transfer function:
kIu 1
s Js þ kPu
Gu ðsÞ ¼
kIu 1
1þ
s Js þ kPu
kIu
¼
Js2 þ kPu s þ kIu

kIu
J
¼ ; (34)
kPu kIu
s2 þ sþ
J J
which is a standard second-order system where the zero is canceled. One may find that the controller gains kPu and kIu of the IP
controller are exactly the same as those of the PI controller given in (32). That means with the same gains, we can just reconfigure the
proportional part to switch between PI and IP controllers.
Furthermore, it is interesting to see that the armature coil can also be regulated by a similar IP controller instead of using the PI
scheme as presented for the DC/DC converter inductor. Normally, the coil is considered as a first-order plant 1=ðLa s þ ra Þ, where
a PI controller is suitable as aforementioned. However, if we consider the coil as an integrator 1=La s with a negative feedback of
the gain ra, then it is analogous to the shaft with J and B. Thus, IP control can be applied for the coil with kPi ¼ 2xiuniLa and kIi ¼
u2ni La . By that, we do not need to know the parasitic resistance ra and can achieve smooth current control performance with lower
required armature voltage ua . The reduction of ua is especially vital when the motor drive works close to the voltage saturation
region.
 Practical linear controllers for power electronics and drives 279

Fig. 10 Speed control performance of the IP controller in comparison with the PI controller.

To demonstrate the performance improvement of the IP controller to motor speed regulation, let us examine a PM DC motor
supplied by a 40-V DC voltage source, with the armature coil having La ¼ 2 mH, ra ¼ 0.1 U, the field constants ke ¼ kT ¼ 0.25, and
the shaft having J ¼ 0.0015 kg m2 and B ¼ 0.001 Nms/rad. The motor rated speed is 1500 rpm, the rated torque is 2 Nm and the
peak torque is 3.5 times of the rated one. For both current and speed loops, the damping factor is selected as xu ¼ xi ¼ 1. The desired
response time of the speed loop is Tru ¼ 0.2 s and the one of the current loop is Tri ¼ 1 ms.
The simulation scenarios and results are plotted in Fig. 10, in which the upper subfigure is the motor rotational and the lower
one is the electromagnetic and load torques. Firstly, the motor is accelerated without load to its rated speed as a step response. After
0.2 s, as designed, both PI and IP controllers regulate the motor speed to the reference. Nevertheless, the transient trajectories are
different. In the case of PI control, the motor responds quickly to the reference step change, performs an overshoot, then reduces to
the set point. In contrast, with the IP controller, the response curve is smooth without any overshoot that requires less control effort
than the PI scheme. The control efforts are reflected via the motor torques. The PI controller needs a torque raising shapely to the
maximum limitation to accelerate the motor and then falling down rapidly due to the speed overshoot. Whereas the motor torque is
smooth with a much lower peak value when the IP controller is in use. That demonstrates the advantage of the IP scheme over the PI
counterpart.
When the speed is in steady state, the step rated load torque is suddenly enforced to the motor. The speed curve therefore drops
and recovers within 0.2 s. It can be seen that the two control schemes have the same behavior of reacting to the load change. One
may also see that in the steady state, the motor torque slightly differs from the load torque TL that is because of the viscous friction
torque Tf ¼ Bum.

Disturbance observer
Comparing the voltage response in Fig. 6 and the speed response in Fig. 10, it can be seen that the speed droop is notable
whereas the voltage is more robust to the step load change. It is because in the voltage control loop, the load current il is
measured and compensated to deduce the chopper current reference ich as shown in Fig. 4. It is, however, not the case of
the motor speed control loop. Even though there exist torque sensors, they are expensive and bulky so that they are often
used for motor characterization and testing, but not in application field. Hence, the motor load torque is often considered
as immeasurable. Thus, to get the information of the disturbance to compensate the speed control loop for performance
improvement, the load torque should be estimated. Estimation is to compute the unmeasured variables from the measured
ones. In this part of the article, we present a simple-but-effective method called disturbance observer-based control (DOBC)
originally proposed by Ohishi et al. (1983).
280 Practical linear controllers for power electronics and drives

Fig. 11 Principle of disturbance observer-based control scheme.

The principle of DOBC is illustrated in Fig. 11. The plant P(s) is regulated by the controller C(s), which produces the control
signal uc. The disturbance, in fact, may affect any part of the plant, including model uncertainty. It is assumed that we have no infor-
mation about the disturbance, thus, it can be considered as the unknown input d of the plant. The DOB is supposed to estimate the
disturbance bd to be directly compensated at the control input.
The idea of how to obtain b d is quite simple. Suppose that we have a nominal model Pn(s) of the plant. If the model is correct, i.e.,
Pn(s) ¼ P(s), and there is no external disturbance enforced to the system, then Pn(s) should produce the same output y as P(s) does
when we apply the same input u. Hence, if we have an inverted model Pn ðsÞ1 , which takes y as its input, the inverted model should
exactly produce the signal u. In case of external disturbance and model uncertainty, the output of Pn ðsÞ1 will be different from u. As
a consequence, their difference would be the information of the total disturbance.
Unfortunately, the ideal inverted model Pn ðsÞ1 cannot be realized in both theory and practice because all the physical plants
have causal models that perform dynamical delays between inputs and outputs. Thus, a pure inverted model of a physical plant is
noncausal that the numerator order of its transfer function is higher than the denominator order. In practice, because that sort of
model performs a derivative action, it would be very sensitive to the measurement noise of the output feedback that may cause the
system unstable. To overcome this issue, the inverted model can be associated with a low-pass filter (LPF) so that it makes the
inverted model causal, in term of theory, and it filters the measurement noise, in term of practice.
Adding a LPF to the inverted model causes a new issue: the signal is delayed. If the output of Pn ðsÞ1 is directly compared to the
control input u, then the difference does not truly reflect the disturbance. A twin LPF is, therefore, added to obtain an identical delay
of the signal u. As a result, the comparison produces the estimated disturbance b d that is delayed regarding the real disturbance d. It
should be noted that in implementation, the LPF should be combined with Pn ðsÞ1 in a unique block instead of a filter connected
with the inverted model as illustrated.
Designing the LPF is a trade-off between the dynamical delay and the noise filtering effect, as well as the system bandwidth. A
design reference could be that the response time of the LPF is at least ten times higher than the sampling time of the control system
and ten times faster, i.e., smaller, than the response time of the controller. Then, the control engineer may tune the filter to get the
desired performance.
For the application of motor speed control, the plant is an integrator 1=Js, thus, the inverted model is Js, which is a first-order
derivative. To achieve causality, a first-order LPF can be employed as:
1
GLPF ¼ ; (35)
ss þ 1
in which s is the time constant of the filter. For realization, the inverted model and the filter are combined as:
Js
GFI ¼ ; (36)
ss þ 1
where “FI” denotes “filtered inversion”.
To justify the advantage of the DOBC scheme, the same examined DC motor drive and simulation scenarios as the part of IP
controller are inherited. Since it takes a first-order system about five times of the time constant to reach more than 98% of the refer-
ence step change, the parameter s here is chosen as 5 ms to have the estimation delay of about 25 ms. This desired delay could be
suitable for a speed loop responding in 0.2 s of a motor control sampled at 0.1 ms, i.e., 10 kHz which is common for industrial drive
systems. In fact, the engineer may get much faster response with this system by reducing s, e.g., to 1 ms. However, this DOB design is
chosen for a comparative illustration purpose.
The speed and torque responses are given in Fig. 12 zooming at the load step change. Comparing to the conventional IP
control, the DOBC scheme dramatically lessens the speed droop by enforcing fast torque response with negligible higher
 Practical linear controllers for power electronics and drives 281

Fig. 12 Speed control performance of the IP controller with disturbance observer.

Fig. 13 Convergence of the observed disturbance.

peak torque. Furthermore, as aforementioned, the slight overshoot can be more decreased to be insignificant by reducing the LPF
time constant s. The convergence of disturbance estimation is given in Fig. 13. The total resistant torque includes the load and the
b res in Fig. 8, converges to the total disturbance within about 25 ms as the
viscous friction torques. The observed disturbance, T
DOB is designed.
In this simulation, to illustrate the estimation convergence as mentioned above, we consider that the moment of inertia J is
correctly known. The interested readers may want to test the design of the DOB with Jn s J, e.g., Jn ¼ 1.5 J, which means 50%
of parameter error, to find out that the DOBC scheme would properly work despite model uncertainty.

Conclusion

We have addressed several techniques to develop linear controllers for regulating power electronics converters and electrical
drives. The principles, development, and design guidance have been presented for typical examples of the bidirectional DC/
DC converter and the DC motor. The methods, however, are not limited to theses examined systems, but can be applied to
various applications, e.g., AC power converters and machine drives in rotating synchronous coordinates. The presented tech-
niques are simple but effective, easy to understand, design, and implement in real-world systems to achieve high control
performance.
282 Practical linear controllers for power electronics and drives

References

Bouscayrol, A., Hautier, J.P., Lemaire-Semail, B., 2013. Graphic formalisms for the control of multi-physical energetic systems: COG and EMR. In: Roboam, X. (Ed.), Systemic
Design Methodologies for Electrical Energy Systems: Analysis, Synthesis and Management. ISTE Ltd., pp. 89–124. Chapter 3.
Erickson, R.W., Maksimovic, D., 2020. Fundamentals of Power Electronics. Springer Nature.
Harashima, F., Kondo, S., 1982. A design method for digital speed control system of motor drives. In: 1982 IEEE Power Electronics Specialists Conference, pp. 289–297.
Kuo, B.C., 1991. Automatic Control Systems, sixth ed. Prentice Hall.
Mathworks, 2021. Tuning Multiloop Control Systems. https://www.mathworks.com/help/slcontrol/ug/tuning-multi-loop-control-systems.html.
Nam, K.H., 2017. AC Motor Control and Electric Vehicle Applications. CRC Press.
Ogata, K., 2010. Modern Control Engineering, fifth ed. Prentice Hall.
Ohishi, K., Ohnishi, K., Miyachi, K., 1983. Torque-speed regulation of DC motor based on load torque estimation. In: IEEJ International Power Electronics Conference, IPEC-Tokyo,
pp. 1209–1216.