Energy 89 (2015) 55e64

Contents lists available at ScienceDirect

Energy
journal homepage: www.elsevier.com/locate/energy

Analytical solution of optimized energy consumption of Double Star

Induction Motor operating in transient regime using a
HamiltoneJacobieBellman equation
Imen Kortas*, Anis Sakly, Mohamed Faouzi Mimouni
Research Unit of Industrial Systems Study and Renewable Energy (ESIER), National Engineering School of Monastir (ENIM), Ibn El Jazzar, Skaness, 5019,
Monastir, University of Monastir, Tunisia

a r t i c l e i n f o a b s t r a c t

Article history: The problem of energy optimization of a DSIM (Double Stator Induction Motor) using the concept of a
Received 4 February 2015 RFOC (Rotor Field Oriented Control) can be treated by an OCS (Optimal Control Strategy). Using OCS, a
Received in revised form cost-to-go function can be minimized and subjected to the motor dynamic equations and boundary
22 June 2015
constraints in order to find rotor flux optimal trajectories. This cost-to go function consists of a linear
Accepted 10 July 2015
Available online 6 August 2015
combination of magnetic power, copper loss, and mechanical power. The dynamic equations are rep-
resented by using a reduced Blondel Park model of the DSIM. From the HJB (HamiltoneJacobieBellman)
equation, a system of nonlinear differential equations is obtained, and analytical solutions of these
Keywords:
Double Star Induction Machine
equations are achieved so as to obtain a time-varying expression of a minimum-energy rotor flux. This
Rotor Field Oriented Control analytical solution of rotor flux achieved maximum DSIM's efficiency and was implemented in the
HamiltoneJacobieBellman ORFOC (optimal rotor flux oriented control) and compared to the conventional RFOC at different dynamic
Optimal control regime of the DSIM. Simulation results are given and improved the effectiveness of the proposed
Energy minimization strategy.
© 2015 Elsevier Ltd. All rights reserved.

1. Introduction require robust algorithms of energy minimization strategies. To

achieve performance, a strategy of optimizing energy of the DSIM
Some of the largest opportunities to save energy and reduce in transitory regime using calculus of variation theory is developed.
operating costs in buildings and industrial facilities come from An integral function is considered and decomposed into a weighted
optimizing electric motor systems. In general, the most part of sum of power-energy of the DSIM for a given time interval. This
electricity assumed flows through motors mainly induction motors function called the cost function will be constrained to boundaries
[13]. The DSIM (Double Stator Induction Motor) is the dominant conditions and to rotor flux and motor speed dynamical equations
technology used today due to its high performance, its high reli- which are developed from the DSIM transient model in a turning
ability, and its speed and torque capabilities. In many electrical (d,q) reference frame [1]. However, the variational calculation leads
drive applications, the Double Star Induction Motor does not to solving a nonlinear differential equation which depends explic-
operate at the nominal point since the desired torque may change itly on the boundary conditions.
according to position or velocity. It is then interesting to consider In this context; Lorenz et al. [2,3] have opted for an operating
other flux operation modes in order to optimize the system per- loss modeled by copper and core losses. In order to compute the
formance, particularly when the task is to reduce the DSIM energy minimal loss flux trajectory, they developed an optimal control
consumption under transient modes. In the industrial applications, strategy using a dynamic programming technique. C. Canudas de
many electrical drives operate with a rapidly-varying load torque Wit et al. [4,5] considered a convex energy cost function including
[15]. The transient regime is explained by nonlinear models that the stored magnetic energy and coil losses. They developed a
nonlinear EulereLagrange equation from which an optimal flux
norm trajectory can be derived. The obtained equation was un-
* Corresponding author.
solvable for an arbitrary torque. They proposed a suboptimal
E-mail addresses: [email protected] (I. Kortas), [email protected] analytical solution aligned with a constant torque operation. In Ref.
(A. Sakly), [email protected] (M.F. Mimouni). [6], the authors present the solution of the nonlinear optimal

http://dx.doi.org/10.1016/j.energy.2015.07.035
0360-5442/© 2015 Elsevier Ltd. All rights reserved.
56 I. Kortas et al. / Energy 89 (2015) 55e64

control problem of threeephase IM (induction motor). A third or- objective optimization has been utilized distinctly with each
der nonlinear model described in arbitrary rotating frame of in- objective function, nonetheless of other objectives. Throughout the
duction motor is used in this paper along with a quadratic last setting, input power, coefficient of performance and ecological
performance index. The problem is solved using the quasilinera- function objectives are optimized concurrently employing a non-
ization approach which converts the nonlinear optimal control dominated sorting GA (genetic algorithm) named the non-
problem into sequence linear quadratic optimal control problems. dominated sorting genetic algorithm (NSGA-II). As in multi-
In Ref. [7], a new minimum-time minimum-loss speed control al- objective optimization, an assortment of optimum results named
gorithm for induction motors is suggested to obtain high perfor- the Pareto optimum frontiers are gained rather than a single ulti-
mance, as well as high efficiency, under field-oriented control with mate optimum result gained via conventional single-objective
practical constraints on voltage and current. This algorithm utilizes optimization. Thus, a process of decision making has been uti-
a two-stage control. In the transient stage, a maximum torque lized for choosing an ultimate optimum result.
control algorithm is utilized to get the minimum-time response. In In Ref. [18], an optimization investigations of an irreversible
the steady state, a minimum-loss control algorithm is applied to absorption heat pump system on the basis of a new thermo-
improve the efficiency. Simulation studies show the performance of ecological criterion is developed. The objective functions which
the proposed minimum-time minimum-loss control algorithm considered are the specific heating load, COP (coefficient of per-
under field-oriented control. formance) and the ECOP (ecological coefficient of performance).
S. M. Yang and F.C. Lin [8] propose a scheme that uses power Three objective functions of the ECOP, COP and the specific heating
factor control with automatic measurement of the minimum-loss load are optimized simultaneously using the multi-objective opti-
power factor commands. A fuzzy logic compensator is included in mization algorithm NSGAII. COP and ECOP are maximized and
the controller to improve the accuracy of the generated commands. specific heating load is minimized in order to get the best perfor-
The scheme is simple for implementation and does not require an a mance. Decision making is done by means of three methods of
priori knowledge of motor parameters. Experimental results have LINAMP and TOPSIS and FUZZY. Finally, sensitivity analysis and
validated the effectiveness of this scheme to minimize the motor error analysis was performed for the system.
operating losses. A design of an adaptive nonlinear control system S. Hoseyn et al. [19] present an optimal design of a solar-driven
for high performance induction motors is developed in Ref. [9]. The heat engine Based on thermal and ecological criteria. In the present
proposed control system is of the explicit model reference type. It investigation, thermodynamic analysis and an EA (evolutionary
consists of a nonlinear controller (inner loop) that controls the rotor algorithm) were employed to optimize the dimensionless ecolog-
speed, an adaptation mechanism (outer loop) that involves a ical function, thermal efficiency, and dimensionless power of a
maximum likelihood estimator, communicating with a feedback solar-driven engine system. Four scenarios were conducted for
control law that uses the results of the adaptation mechanism to optimization of the solar heat engine. In the first three, a traditional
redesign the inner loop controller online. The advantage of syn- single objective optimization was employed separately with each
thesizing this type of controller lies in the fact that the desired objective function, regardless of other objectives. In the fourth
trajectory of the rotor speed is determined from the output of the scenario, efficiency and power objectives were optimized simul-
reference model, while the control trajectories that lead to that taneously using a nondominated sorting GA (genetic algorithm)
behavior are computed through the developed state feedback called the nondominated sorting genetic algorithm (NSGA-II). As in
control law. The control system is simulated under a situation multiobjective optimization, a set of optimal solutions called the
where some of the parameters vary in the presence of noise. It is Pareto optimal frontier was obtained instead of a single final
shown that the adaptive controller keeps the performance of the optimal solution obtained in traditional single-objective optimi-
drive system close to the desired performance even in the presence zation. Therefore, a process of decision making was employed for
of uncertainty. The effect of measurement noise is also taken into selecting a final optimal solution. Three decision-making pro-
consideration to show that the controller is feasible for practical cedures were applied to find optimized solutions from the Pareto
situations. optimal solutions in the objectives' space. The results obtained
M. H. Ahmadi et al. [16] present a developed ecological function from four optimization scenarios were compared and discussed
for absorption refrigerators with four-temperature-level. Moreover, using a deviation index introduced in this paper. It was shown that
aforementioned absorption refrigerator is optimized by imple- the optimal results obtained in single-objective optimization with
menting ecological function. With the aim of the first and second an ecological objective are very close to the corresponding results
laws of thermodynamics, an equivalent system is initially deter- obtained in the multiobjective optimization, in that the power and
mined. To reach the addressed goal of this research, three objective thermal efficiency are optimized simultaneously.
functions that the COP (coefficient of performance), the ecological In this paper, minimum energy control will be solved by an
function and thermoeconomic criterion have been involved in alternative approach called the HJB (Hamilton-Jacobi Bellman), in
optimization process simultaneously. Three objective functions are which the optimal control is not interested in the specific initial
maximized at the same time. Developed MOEAs (multi objective states, but in any unspecified initial conditions. In fact, the cost
evolutionary approaches) on the basis of Non-dominated Sorting function subject to the criterion index is transformed into a cost-to-
Genetic Algorithm (NSGA-II) method are implemented throughout go function. By imposing a transient mode to the DSIM drive, the
this work. nonlinear HJB equation was successfully solved in an analytical
In Ref. [17], authors propose an ecological and thermal ap- form and offered a time-varying expression of minimum energy
proaches for the Ericsson cryogenic refrigerator. Three objective rotor flux. This analytical solution was implemented in an optimal
functions (input power, coefficient of performance and ecological RFOC (Rotor Field Oriented Control) in which both a deadbeat rotor
objective function) are gained for the suggested system. flux controller and a saturation model are introduced.
Throughout the current research, an EA (evolutionary algorithm) This paper is organized as follows: Section 2 is mainly intended
and thermodynamic analysis are employed to specify optimum to describe the full DSIM model. Section 3 is devoted to present the
values of the input power, coefficient of performance and ecological reduced model. In the fourth section the energy model of the DSIM
objective function of an Ericsson cryogenic refrigerator system. is explained. In Section 5, the optimal control strategy is presented
Four setups are assessed for optimization of the Ericsson cryogenic in details by the energy-power cost function, the presentation of
refrigerator. Throughout the three scenarios, a conventional single- the optimal control problem, the mechanical system operation (the
 I. Kortas et al. / Energy 89 (2015) 55e64 57

case of a linear time-varying motor speed), the Hamil- 8

> 4_ _ rðd;qÞ þ bU
¼ ðaI þ rJÞ4
toneJacobieBellman equation and its resolution and finally the < rðd;qÞ
cUt J4rðd;qÞ (3)
RFOC scheme. Section six deals with the simulation results by >
:U _ ¼  Kl þ
comparing the optimized RFOC with the conventional one. Jm Jm

2. Modeling of the DSIM

2.3. Energy model of DSIM
2.1. Full-order dynamic model of DSIM
The instantaneous active power in the (deq) rotating frame is
By selecting the stator current, the rotor flux vectors and the given by:
motor speed as state variables, the complete DSIM dynamic model  t
in the synchronous Park reference frame is given as [10]: Pa ¼ 3 2 Vsðd;qÞ Isðd;qÞ
=
(4)
8 From the system (1), the input power is given by:
>
>
> I_sðd;qÞ ¼ ðgI þ ðr_ þ pUÞJÞIsðd;qÞ 
>
>   t  
<
4_ rðd;qÞ ¼ ðaI þ rJÞ4
_ rðd;qÞ þ bI Pa ¼ 3 2 s1 Ls þ s2 Lsp : I_sðd;qÞ
=
Isðd;qÞ
(1)
> 
>
> t    t  
>
>
:U _ ¼  Kl þ Y
Jm Jm þ g: I_sðd;qÞ Isðd;qÞ  2ha 4rðd;qÞ Isðd;qÞ
 t
where: þ hpU 4rðd;qÞ JIsðd;qÞ
      (5)
Is1d þ Is2d Isd Vsd1 þ Vsd2
Isðd;qÞ ¼ ¼ ; Vsðd;qÞ ¼
Is1q þ Is2q Isq Vsq1 þ Vsq2 The relation between the stator and rotor current can be given
    as follows:
Vsd 4rd
¼ ; 4rðd;qÞ ¼
1 
;
Vsq 4rq
Irðd;qÞ ¼ 4rðd;qÞ Ls Isðd;qÞ (6)
Lr

  2 
M2 M The instantaneous active power becomes then:
s1 ¼ 1  ; s2 ¼ 1  ;g
L r Ls Lr Lsp  
!   t 
Pa ¼ 3 2 s1 Ls þ s2 Lsp : I_sðd;qÞ
=
Isðd;qÞ
1 M2
¼ Rs þ 2 Rr ; 
s1 Ls þ s2 Lsp Lr 1  t     t  
þ : 4_ rðd;qÞ 4rðd;qÞ  3 2 Rs I_sðd;qÞ
=
Isðd;qÞ
Lr
      t  
1 0 0 1 Rr
I¼
0 1
;J ¼
1 0
; a ¼ ; b ¼ aM; þ Rr I_rðd;qÞ Irðd;qÞ þ UY (7)
Lr
By noting us ¼ r_ is the sleep frequency, Rs and Rr are stator and Finally, the instantaneous active power is given as:
rotor resistances, Ls and Lr are stator and rotor inductances; M is
v
the magnetizing inductance, Lsp is the principal cyclic inductance. Pa ¼ W þ PJ þ Pm (8)
Jm is the total moment of inertia of the rotor, Kl is the load torque vt
constant. Isd1 , Isd2 , Isq1 and Isq2 are respectively the direct and By means of a field-oriented control drive, we define from
quadrature current of stator 1 and stator 2. Vsd1 ; Vsd2 ; Vsq1 ; Vsq2 Equation (7):
are respectively, the stator voltages in d-q axis of each stator, p is  the derivate of stored magnetic given as follows:
the poles number, Y is the electromagnetic torque.

  
2.2. Reduced model of the DSIM v s1 Ls þ s2 Lsp  2  1 2
W ¼ 32 =
: u1 þ u22 þ 4r (9)
vt 2 2Lr
In order to eliminate the nonlinear terms in the DSIM dynamic
model and allow an easy resolution of a possible optimal control
problem, the system can be forced into a current-command mode
 the Joule losses:
using high-gain feedback. Furthermore, the reduced order current-
fed DSIM model can be simplified by restraining the Proportional -
Integral (PI) to a sample proportional gain [4], as follows:
  t     t  
s1 Ls þ s2 Lsp PJ ¼ 3 2=
Rs I_sðd;qÞ Isðd;qÞ þ Rr I_rðd;qÞ Irðd;qÞ
Vsðd;qÞ ¼ (2)
ε
(10)
with By using the first equation of system (1) and Equation (10), the
    Joule losses can be expressed with respect to U and 4r as follows:
u1 Is1d þ Is2d
U¼ ¼
u2 Is1q þ Is2q  2 !  !
M  3R 3 d42r
2 2 r 2
PJ ¼ 32
=
Rs þ Rr : u1 þ u2  4 
which 0 < ε < 1 and U is a new command of the system. Lr 2 L2r r 2Lr dt
The standard reduced model form is obtained through singular
(11)
perturbation. It is given by the following system:
58 I. Kortas et al. / Energy 89 (2015) 55e64

 The mechanical power of the DSIM:  2 !

 M
r1 ¼ 34
=
s1 Ls þ s2 Lsp 41 þ 2 Rs þ Rr
3 =
42 ;
Lr
Pm ¼ UY (12) r2 ¼ r1 þ r0 a3
In term of rotor variables and torque current ðIsq Þ, we get:
M a1 3 Rr 3M
Pm ¼ 3 2 =
4 Isq U (13) q1 ¼ 3 4 =
 2 2 a2 þ a3 q0 ; q2 ¼
=
a
Lr r Lr Lr 2 Lr 4
In this paper, the mechanical process operations are restricted to
3. The optimal problem control the accelerate modes in order to limit the study to the transient
operation. Then the motor speed can be chosen as the following
3.1. Determination of the cost function expression:

U* ¼ c0 t þ c1 (19)
Cost function can be defined as an integral of the index,
f ðIsd ; Isq ; 4r ; UÞ given as follows:
with c0 > 0.
ZT

J¼ f Isd ; Isq ; 4r ; U dt (14) 3.2. Determination of HamiltoneJacobieBellman equation
0
In order to increase the subsequent equations' readability, we
The index corresponds to the weighted sum:
denote x1 ¼ 4r and x2 ¼ U.

f Isd ; Isq ; 4r ; U ¼ b1 WL þ b2 PJ þ b3 Pm (15) The dynamic system and the cost function described in (17) and
(18) respectively become:
The weighting factors b1 , b2 and b3 are used to scale power-
energy combined convex criteria terms defined above. Minimize
the cost function provides two important advantages: first, mini- ZT 
b2  2  
mizing the corresponding magnetic energy stored, allowing thus Jr ¼ 4r ð0Þ  42r ðTÞ þ r1 u21 þ r2 u22 þ q1 x21 þ q2 4r u2 x2 dt
maximizing the power factor, the second being the minimization of Lr
0
losses in the winding thus increasing the machine efficiency.
(20)
Using equations (9), (11), (13) and (14), the cost function is given
as follows:

2    2 ! !3
   1 M  R
r
ZT 6 b1 s1 Ls þ s2 Lsp : u21 þ u22 þ : 42 þ b2 Rs þ Rr u21 þ u22  2 42r 7
6 2Lr r Lr Lr 7
J ¼ 32 = 6 7dt (16)
6   7
4 b2 d 2 M 5
0  4r þ b3 p 4r u2 U
Lr dt Lr

and
Z T 
b d 2 b 8
with  2 ð4r Þ dt ¼ 2 ð42r ð0Þ  42r ðTÞÞ representing the
0 Lr dt Lr < x_1 ¼ ax1 þ bu1
>
boundary conditions of the rotor flux in [0, T].   Kl cu2 x1 (21)
Considering the new control vector U ¼
u1
, the system
>
: x_2 ¼  x2 þ
u2 Jm Jm
described by (4) can be defined as follows:
8 Therefore, the task is to find an optimal cost-to-go function
< 4_ r ¼ a4r þ bu1
> VðT; x1 Þ that must necessarily satisfy the boundary condition
(17)
>
:U _ ¼  Kl U þ cu2 4r VðT; x1 Þ ¼ Kðx1 ðtÞÞ, with Kðx1 ðtÞÞ is defined as the final state pen-
Jm Jm alty term. In our case:
Using the new constraint given by (17) and the cost function
b2  2  b  
given by (16), an optimal control problem can be presented as Kðx1 ðtÞÞ ¼ 4r ð0Þ  42r ðTÞ ¼ 2 x21 ð0Þ  x21 ðTÞ (22)
follows: Find the optimal control variables ðu*1 Þ and ðu2 Þ that Lr Lr
transfers the dynamic system described in (17) from the initial The Optimality principle has shown that the optimal co-state
states: 4r ð0Þ ¼ 4r0 and U(0) to admissible final states 4r ðTÞ and l* ðtÞ as defined in the Hamiltonian-Jacobi [11] equation corre-
U(T) so that the cost function is minimized: sponds to the gradient of the optimal cost-to-go function given as
follows:
ZT 
b2  2   
Jr ¼ 4r ð0Þ  42r ðTÞ þ r1 u21 þ r2 u22 þ q1 42r þ q2 4r u2 U dt * vV T; x*1 ðtÞ
Lr l ðtÞ ¼ (23)
0 vx*1
(18) vV ðT;x* ðtÞÞ
where vx*1
1
is continuously differentiable with respect to x1 et
where the weighting factors r1 , r2 , q1 and q2 must be positives. x1 ¼ x*1 ðtÞ.
 I. Kortas et al. / Energy 89 (2015) 55e64 59

All along an arbitrary admissible trajectory u1 ð:Þ and x1 ð:Þ, the

l1 ¼ r1 ; l2 ¼ r 1 A2 ; l3 ¼ q2 c0 A; l4 ¼ 2r1 AB; l5
corresponding suboptimal cost-to-go function
¼ q2 ðc0 B þ c1 AÞ; l6 ¼ q1 ; l7 ¼ r1 B2 ; l8 ¼ q2 c2 B
ZT

V t; x*1 ¼ Kðx1 ðtÞÞ þ LðxðtÞ; uðtÞ; t Þdt (24) The optimal control variable u*1 ðt; x1 Þ is obtained from
vHðt;x1 ; u1 Þ
t vu1 ¼ 0, and then we have:

is given as: b vVðt; x1 Þ

u*1 ðt; x1 Þ ¼  (32)
2l1 vx1
ZT 
b2  2 
By replacing the Hamiltonian expression in (31), the equivalent
Vðt; x1 Þ ¼ x1 ð0Þ  x21 ðTÞ þ r1 u21 þ r2 u22 þ q1 x21
Lr HJB equation becomes:
0
  ! !
þ q2 x1 u2 x2 dt (25) vV t; x*1 l2 l4
¼  min l1 u*1 þ þ l 2
3 t þ þ l 2
5 t þ l6 x1
vt u*1 x21 x21
according to the following differential equation: !
l7 vVðt; x1 Þ
þ þ l8 þ ðax1 þ u1 Þ
Vðt; x1 Þ Vðt; x1 Þ Vðt; x1 Þ vVðt; x1 Þ x21 vx1
¼ x1 þ ¼ lð  ax1 þ bu1 Þ þ
dt dx1 dx1 vt
(33)
¼ Lðx1 ; u1 ; tÞ Substituting u*1 ðt; x1 Þ expression in (33), we deduce:
(26)   !2  * !
hence, vV t; x*1 b2 vV t; x*1 * vV t; x1 l2 2
¼ 2 þ ax  þ l3 t
vt 4l1 vx*1 1
vx*1 x21
vVðt; x1 Þ !
¼ lð  ax1 þ bu1 Þ  Lðx1 ; u1 ; tÞ ¼ Hðt; x1 ; u1 ; lÞ
vt l4 l
 þ l5 t  l6 x21  72  l8
This corresponds to the partial differential equation for the x21 x1
optimal cost-to-go function Vðt; x1 ðtÞÞ which we call the HJB (34)
equation [11,12].
The control problem is, then, expressed as follows: Find an op-
timum control variable u* ðt; x1 Þ which generates the corresponding
state variable x*1 ðtÞ, with the consideration of the following 3.3. Resolution of the HJB equation
condition:

 c 2  ZT 
*
V t; Fr ¼ min 2
Fr ð0Þ  Fr ðTÞ þ r1 u21 þ r2 u22 þ q1 F2r   2 1
u*1 Lr V t; x*1 ¼ a1 ðtÞ x*1 þ a2 ðtÞ  2 þ a3 ðtÞ (35)
t x*1

þ q2 Fr u2 U vt These yields:

(27) vV t; x*1 2
¼ 2a1 ðtÞx*1  a2 ðtÞ  3 (36)
which transfer the initial state Fr ð0Þ ¼ Fr0 to an admissible final vx*1 x* 1
state x1 ðTÞ ¼ Fr ðTÞ under the following constraint:
and
x_1 ¼ ax1 þ bu1 (28)

vV t; x*1  2 1
This determines directly the following HJB equation: ¼ a_ 1 ðtÞ x*1 þ a_ 2 ðtÞ  2 þ a_ 3 ðtÞ (37)
   vt x*
vV t; F*r vVðt; Fr Þ 1
¼  min H t; F*r ; u*1 ; (29) Using the expressions of the partial derivatives given by (36)
vt u*1 vFr
and (37) into (34), we obtain the following equation:
then 0 12
 * 2 1 b2 @ * 2 A
1 a_ 1 ðtÞ x1 þ a_ 2 ðtÞ  2 þ a_ 3 ðtÞ ¼ 2a1 ðtÞx1  a2 ðtÞ  3
Hðt; x1 ; u1 Þ ¼ r1 u21 þ r1 ðAt þ BÞ2 þ q1 x21 x*1 4l1 x*1
x2 0 1
vVðt; x1 Þ 2
þ q2 ðAt þ BÞðc0 t þ c1 Þ þ  ðax1 þ u1 Þ þ ax*1 @2a1 ðtÞx*1  a2 ðtÞ  3 A
vx1
x*1
(30) ! !
l2 2 l4
 þ l t  þ l
with A ¼ kl :cc0 , and B ¼ Jcm ðc0 þ kl c1 Þ x21
3
x21
5

Rescaling this equation, the HJB equation is finally given by (30):

l7
! ! t  l6 x21   l8
x21
l2 l4 l
Hðt; x1 ; u1 Þ ¼ l1 u21 þ þ l3 t 2 þ þ l5 t þ l6 x21 þ 72 (38)
x21 x21 x1
To satisfy the relation (38) for any x1 > 0 we must solve the
vVðt; x1 Þ
þ l8 þ  ðax1 þ u1 Þ following sets of equations:
vx1
(31) b2 2
a_ 1 ðtÞ  a ðtÞ  2aa1 ðtÞ þ l6 ¼ 0 (39)
With: 4l1 1
60 I. Kortas et al. / Energy 89 (2015) 55e64

b2 0sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1
a_ 2 ðtÞ þ 2aa2 ðtÞ þ 2a1 ðtÞa2 ðtÞ þ l2 t 2 þ l4 t þ l7 ¼ 0 (40) 1 al1 @ l 6 b2
l1 a1 ðtÞ ¼ ! þ 2 1þ  1A
0 1" b l1 a2
t#
2b2 a1p
b2
 2aþ l1
2
a_ 3 ðtÞ þ l3 t þ l5 t þ l8 ¼ 0 (41) @ l1
A 1e
2b2 a 1p
2aþ l1

(52)
b2 1 2b2 a1 ðtÞa2 ðtÞ
 a22 ðtÞ 6 þ ¼0 (42)
l1 x1 l1 x21 Now consider the second Equation (40):

Start with solving Equation (41):    

b2
a_ 2 ðtÞ þ 2a þ 2a1 ðtÞ a2 ðtÞ þ l2 t 2 þ l4 t þ l7 ¼ 0 (53)
  l1
a_ 3 ðtÞ ¼  l3 t 2 þ l5 t þ l8 (43)
we pose
These yields:
   
b2
t3 t2 AðtÞ ¼ 2a þ 2a1 ðtÞ ; BðtÞ ¼ l2 t 2 þ l4 t þ l7 (54)
a3 ¼ l3 þ l5 þ l8 t (44) l1
3 2
Then, Equation (54) became:
From the first Equation (39) and for all x1 > 0, a first order dif-
ferential equation can be defined as follows:
a_ 2 ðtÞ þ AðtÞa2 ðtÞ þ BðtÞ ¼ 0 (55)
b2 2 This is a first order differential equation with second member
a_ 1 ðtÞ  a ðtÞ  2aa1 ðtÞ þ l6 ¼ 0 (45)
4l1 1 with non-constant coefficients.
The solution to Equation (56) gives the following expression of
This type of equation is called in the literature the Riccati
a2 ðtÞ:
equation. The integration of a Riccati differential equation requires
the knowledge of a particular solution a1p of this equation. It is
determined from the following equation: l4 t 2
a2 ðtÞ ¼ þ l7 t þ C1 (56)
2
b2 2 with C1 is the constant of integration defined from initial
a ðtÞ  2aa1p ðtÞ þ l6 ¼ 0 (46)
4l1 1p conditions.
vVðt;x*1 Þ
Then we find: Using (52) and (56), the expression x*1
in Equation (37)
becomes:
0sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1
al1 @ l6 b2  0 1
a1p ¼ 2 1þ  1A  
(47) vV t; x*1 1 al 1
b l1 a2 ¼ 2@ 2  t
þ 2 1
 1 Ax*1 ðtÞ
x*1 bl1b 1  e =b b 2ab
Using the following substitution: zðtÞ ¼ a1 ðtÞ  a1p ðtÞ, we get  
the following differential equation: 2l2 t 3 1
þ þ l4 t 2 þ 2l7 t þ C1 (57)
3 x*3
1 ðtÞ
 0 b2  2 
zðtÞ þ a1p ðtÞ  zðtÞ þ a1p  2a zðtÞ þ a1p þ l6 ¼ 0 Finally, from the expression of the optimal control given in (32),
4l1 is expressed by Equation (58):
(48)
0 1
 
Moreover, we have: b @ 1 al1 1
u*1 ðtÞ ¼  þ 2  1 Ax*1 ðtÞ
  l1 b2 b 1  et=b b 2ab
l1
b2 b2  3 
_ 
zðtÞ a1p þ 2a zðtÞ ¼ ðzðtÞÞ2 (49)
2l1 4l1 b l2 t l t2 1
 þ 4 þ l7 t þ C1 (58)
l1 3 2 x*3
1 ðtÞ
This simplification leads to the Bernoulli equation (a ¼ 2).
The solution to Equation (50) gives the following expression of 2b2 a
with b ¼ 2a þ l1 1p
zðtÞ:
Using the dynamic equation of the flux, and replacing the
command u*1 in (58), we obtain a nonlinear first order differential
1
zðtÞ ¼ ! (50) equation of rotor flux:
0 1"
t#
2b2 a1p
b2
 2aþ l1 0 1
@ l1
A 1e  
b 1A * b2 l2 t 3 l4 t 2 1
x_*1 ðtÞ¼ @
2b2 a
2aþ l 1p t=
þ x 1 ðtÞ þ þl7 t þC 1 *3 ðtÞ
1
1e b 2b l 1 3 2 x 1
Finally, the solution a1 ðtÞ is described by the following equation:
(59)
a1 ðtÞ ¼ zðtÞ þ a10 ðtÞ (51) Substitute sðtÞ for x*1 ðtÞ
as follows: sðtÞ¼ðx*1 ðtÞÞ3 .
By introducing this new variable sðtÞ in the Equation (59), a first-
whether order differential equation is obtained:
 I. Kortas et al. / Energy 89 (2015) 55e64 61

0 1 150
 
b 1 A b2 l2 t 3 l4 t 2
_ ¼ sðtÞ@
sðtÞ t
 þ  þ þ l7 t þ C 1
1  e =b 2b l1 3 2 reference rotor speed

(60)
where: sðt0 Þ ¼ ðx*1 ðt0 ÞÞ3
et sðTÞ ¼ ðx*1 ðTÞÞ3 100
The solution pðtÞ of the equation is given as follows:

0 1
Zt
1 b
@ t
 þ Adt Z t  
1  e =b 2b b2 l2 t 3 l t2 50
pðtÞ ¼ e 0  þ 4
l1 3 2
0
0 1
Zt
1b
 @ t
 þ Adt
1  e =b 2b
þ l7 t þ C1 e0 dt 0
0 1 2 3 4 5 6 7 8 9

(61) Fig. 1. Rotor speed reference.

The final solution of the Equation (61) is expressed as follows:

2   3
 2b2 þ 1 l t 3 l t 2 
4b2  
 6
6
2
þ 4 þ l7 t þ C1   2
2 l2 t þ l4 t þ l7 7
7
2b 3 2
b2 b
2
log 1t 1 6
6
2
2b þ 1 7
7
pðtÞ ¼ e
=
eb 6 7 (62)
l1 6 7
6 8b3
32b 4
l 7
6 2 7
4  3 ð2l2 t þ l4 Þ   4 5
2 2
2b þ 1 2b þ 1

4. Simulation results
Then we determine an analytical solution of the rotor flux which
minimizes energy consumption of the DSIM as follows: The Double Star Induction Machine models with conventional
and optimal RFOC were numerically simulated. The conventional

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2   33ffi
u2
u  2b 2
þ 1  3   
u6 l2 t l t 2
4b 2
u6  6 6 þ 4 þ l7 t þ C1   2
2 l2 t þ l4 t þ l7 77
77
u6 6 2b 3 2 77
u 6 b 2
log 1
1 6 2b 2
þ 1 77
u b
4 2 t
F*r ðtÞ ¼ u6 6 77
=
e eb (63)
u6 l1 6 77
u66
6
6 8b 3
32b 4
l 77
77
u4 4   3 ð2l t þ l Þ  
2
4 55
t 2 4
2 2
2b þ 1 2b þ 1

RFOC is based on a constant value-rated flux trajectory. In both of

From a transient rotor speed reference showed in Fig. 1, a these two control laws, a saturation model with respect to the rotor
comparative study in both of conventional and optimal regimes flux was taken into account.
was performed. Fig. 2 illustrate the time-varying curve of the The reference variations and estimated motor speed (Fig. 4),
minimum-energy rotor flux. the dynamically optimized flux trajectory as well as the con-
The proposed control scheme was studied by simulation as stant value-rated flux with its estimated one (Fig. 5), quadrature
shown in Fig. 3. A time-varying flux is predicted by using the and the direct stator current (Figs. 6 and 7), energy consump-
expression in Equation (63) with respect to the motor speed tion of the (Fig. 8) obtained from both of the proposed ORFOC
reference injected in the input of the deadbeat rotor flux (optimal rotor flux oriented control) and the RFOC are all
controller. presented.
62 I. Kortas et al. / Energy 89 (2015) 55e64

0.35
optimal rtor flux 140

0.3
120

0.25
100

rotor speed (rd/s)

0.2
80
phir P
O

reference rotor speed

0.15
60 estimated rotor speed

0.1 40

0.05 20

0
0 1 2 3 4 5 6 7 8 9 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
time(s) time(s)

Fig. 2. Optimal rotor flux. Fig. 4. Reference and estimated rotor speed.

In order to be restricted to the transient operation, only the Fig. 8 indicates that major energy-loss savings with dynami-
acceleration (from t ¼ 0 s to t ¼ 1 s) and permanent (from t ¼ 1 s to cally optimized rotor flux trajectories. It is obvious to observe
t ¼ 2 s) modes were taken into account in the motor speed profile, that the RFOC taking into account the magnetic effect and
as shown in Fig. 4. operating with optimal flux leads to the best DSIM energy con-
Fig. 5 present the minimum energy flux trajectory correspond- sumption decreasing. This decrease is more considerable for
ing to the motor speed profile, all the unknown boundary condi- smalls loads.
tions and parameters shown in the optimal rotor flux expression in Energy saving of the optimum rotor flux trajectories is demon-
Equation (63) were determined. strated by the following comparison with the constant value-rated
In Fig. 6 the flux current remains constant during motor flux trajectory summed up in Table 1.
speed increase. We see that the flux current delivered by the
ORFOC registers a significant decrease compared to the one
5. Conclusion
delivered by the conventional RFOC. This means that the pre-
sented method saves energy. Fig. 7 prove that the torque current
This paper presents an optimal control strategy using energy
have the same evolution compared to the electromagnetic
minimization. The criterion used in this optimal control is
torque.
subjected to dynamic equations of a DSIM reduced-order model.

Inverter Inverter
current+MLI current+MLI
(Stator1) (Stator2)
Speed
MADE

DC
AC

InversePark
InversePark
Trans
Trans

Estimeted

Park Trans Park Trans

Flux control

18

Fig. 3. ORFOC schema of the MADE.

 I. Kortas et al. / Energy 89 (2015) 55e64 63

0.35
a x 10
4

0.3 2 energy under RFOC

energy under optimal RFOC
0.25
reference rotor flux
1.8
estimated rotor flux
rotor flux (Wb)

0.2

1.6
0.15

energy (J)
0.1
1.4

0.05 1.2

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 1
time(s)

Fig. 5. Reference and estimated rotor flux.

0.7 0.75 0.8 0.85 0.9 0.95
time(s)

50
flux current under RFOC
b x 10
4

45 flux current under optimal RFOC 4.3

energy under RFOC
40 energy under optimal RFOC
4.2

35
4.1

30
Isd (A)

4
25
energy (J)

3.9
20
3.8
15
3.7
10
3.6
5
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
time(s) 3.5

Fig. 6. Direct current evolution from both of the proposed ORFOC and the RFOC. 3.4
1.64 1.66 1.68 1.7 1.72 1.74
time(s)

Fig. 8. a: Energy consumption evolution from both of the proposed ORFOC and the
RFOC for large loads. b: Energy consumption evolution from both of the proposed
ORFOC and the RFOC for small loads.
50
torque current under RFOC
45 torque current under optimal RFOC

40 The rotor flux and the motor speed are considered as state
variables. A closed-cycle mechanical process operation is pre-
35
sented by a transient and constant regime. Based on the optimal
30 control theory, this approach provides a cost function given as
weighted sum of a DSIM energy-power model. In order to
Isq (A)

25 obtain a minimum-energy rotor flux trajectory, the presented

task is to minimize this cost function constrained to a two dy-
20
namic equations of the rotor flux and the motor speed. The
15 optimal control problem is then treated by the HJB equation and
a minimum energy rotor flux solution is successfully deter-
10 mined in an analytical form. This time-varying solution is
implemented in an RFOC in which a deadbeat rotor flux
5
controller and a saturated model with respect to the rotor flux
0 are introduced.
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
This optimal rotor flux oriented control is compared to the one
time(s)
using the conventional RFOC. This comparative study proves that
Fig. 7. Quadrature current evolution from both of the proposed ORFOC and the RFOC. optimal control allowed much lower energy consumption and a
better efficiency than the conventional RFOC.
64 I. Kortas et al. / Energy 89 (2015) 55e64

Table 1
Compared energy consumption of 20 KW DSIM under rotor flux oriented controls.

Time (s) 0.6 1 1.2 1.7 1.9 2

D energy (J) þ50.4 321.2 343.2 277 350.2 434.7

%D energy þ0.0067% 1.8% 1.46% 0.72% 1.1% 1.07%

Acknowledgments [2] Lorenz, RD., Yang SM. Efficiency-optimized flux trajectories for closed cycle
operation of field oriented induction machine drives, IEEE industry applica-
tion society Annual Meeting, 1988 Conference, vol. 1, 457e762.
Thank you to the editors and two anonymous reviewers for their [3] Lorenz RD, Yang SM. AC induction servo sizing for motion control applications
detailed and helpful comments, which considerably clarified and via loss minimizing real-time flux control. IEEE Trans Ind Appl 1992;28.
strengthened the paper. I want to thank Mr. Mohammed Fawzi and [4] Canudas de Wit C, Ramirez J. Optimal torque control for current-feeded in-
duction motors. IEEE Trans Automat Control 1999;44.
Mr. Anis Mimouni Sakly for their patience, their availability, their [5] Didier G, de Wit CC, Ramirez J. Nonlinear H2 and H∞ optimal controllers for
advice and constant help throughout this work. current-fed induction motors. IEEE Trans Automat Control 1999;44.
[6] Jasem MKh, Jaddu HM. The optimal control of a three-phase induction motor.
In: Proceedings of the 25th IASTED conference, Lanzarote, Spain; 2006.
[7] Chang JH, Kim BK. Minimum-time minimum-loss speed control of induction
Appendix motors under field-oriented control. IEEE Trans Ind Electron 1997;44.
[8] Yang SM, Lin FC. Loss-minimization control of vector-controlled induction
motor drives. J Chin Inst Eng 2003;26:37e45.
[9] Rubaai A, Uribina O, Kankam MD. Design of an adaptative nonlinear
controller-based optimal control theory for a voltage source induction motor
Simulation motor parameters. drive system, industry applications conference. In: Thirty-sixth IAS Annual
Meeting Conference Record of the IEEE, vol. 2; 2001. p. 1279e84.
Rated voltage V 220/230 V Stator self inductance Ls 81.2 mH [10] Chiasson J. Modelling and high-performance control of electric machines, IEEE
press series on power engineering. John Willey & Sons; 2005.
Current Rating IN 19 A Rotor Self Inductance Lr 8.9 mH
[11] Geering HP. Optimal control with engineering application. Springer-Verlog
frequency F 50 Hz Mutual Inductance 26.3 mH Berlin Heidelberg; 2007.
Stator-Rotor M [12] Locatelli A. Optimal control. Birkhauser Verlog; 2001.
rated torque Y 127 Nm Stator self leakage 0.78 mH [13] Buoro D, Pinamonti P, Reini M. Optimization of a distributed cogeneration
inductance ls system with solar district heating. Appl Energy 2014;124:298e308.
Number of pairs 2 Rotor self leakage 0.38 mH [15] Song Z, Hofmann H, Li J, Han X, Ouyang M. Optimization for a hybrid energy
of poles p inductance lr storage system in electric vehicles using dynamic programming approach.
Stator resistance Rs 0.4 Ohm Moment of inertia J 0.6 Kgm2 Appl Energy 2015;139:151e62.
Rotor resistance Rr 0.4 Ohm Viscosity coefficient f 0.02 [16] Ahmadi MH, Ahmadi MA, Mehrpooya M, Hosseinzade H, Feidt M. Thermo-
dynamic and thermoeconomic analysis and optimization of performance of
irreversible four-temperature-level absorption refrigeration. Energy Convers
Manag 2014;88:1051e9.
[17] Ahmadi MH, Ahmadi MA. Thermodynamic analysis and optimization of an
irreversible Ericsson cryogenic refrigerator cycle. Energy Convers Manag
2015;89:147e55.
References [18] Ahmadi MH, Ahmadi MA, Mehrpooya M, Sameti M. Thermo-ecological anal-
ysis and optimization performance of an irreversible three-heat-source ab-
[1] Mimouni MF, Dhifaoui R. Modelling and simulation of double-star induction sorption heat pump. Energy Convers Manage 2015;90:175e83.
machine vector control using copper losses minimization and parameters [19] Sayyaadi H, Ahmadi MH, Dehghani S. Optimal design of a solar-driven heat
estimation. Int J Adapt Control Signal Process 2002;16:1e24. http://dx.doi.org/ engine based on thermal and ecological criteria. J Energy Eng 2014:04014012.
10.1002/acs.767.