Documente Academic
Documente Profesional
Documente Cultură
AbstractIn this paper a Model Predictive Control (MPC) Throttle Control (ETC) and Hybrid Electric Motor control
method for torque control of a Permanent Magnet Synchronous in automotive applications. The ability to enforce constraints
Motor (PMSM) is presented. The proposed approach takes into using MPC could benefit motor control performance and
account constraints on voltages and currents, and allows the use component protection.
of modulation techniques that eliminate the side effects caused
by the direct transistor actuation performed by Model Predictive To avoid the computational load required by online MPC
Direct Torque Control (MP-DTC) approaches. The optimization two main techniques have been widely adopted in recent years:
problem resulting from the proposed MPC formulation is solved explicit MPC [4] and Finite Control Set MPC (FCS-MPC) [5].
online, in contrast with what is done in explicit MPC, where In explicit MPC the optimization problem is pre-solved offline
the optimal control law is obtained offline by multiparametric via multiparametric Quadratic Programming (mpQP), and the
optimization. The performance of the proposed control strategy
is evaluated in Processor-In-the-Loop (PIL) experiments, carried
MPC solution turns out to be an easy-to-implement PieceWise
out on a low cost Digital Signal Processor (DSP) commonly used in Affine (PWA) function of the parameters, for example via a
motion control. Results show that the proposed approach is able binary search tree evaluation [6]. The search time, in this case,
to improve the torque dynamics with respect to a conventional is logarithmic and the memory occupancy is polynomial in the
controller, and that an embedded implementation is feasible in number of PWA regions, thus the explicit controller is well
terms of required computational power and memory. suited for small problems only. The explicit MPC approach
has been successfully applied to motion control, see, e.g., [7]
I. I NTRODUCTION [11].
Permanent Magnet Synchronous Motors (PMSMs) are The FCS-MPC was initially applied in the field of power
widely used in applications where high-performance, fast converters and drives. It exploits the discrete nature of those
torque response and high power density are required, such as in systems by manipulating the inverter switch positions directly.
robotics and machine tool drives. The standard control method The approach results in a combinatorial optimization problem
for both PMSMs and induction machines is Field Oriented that is solved on line by enumeration and provides a sequence
Control (FOC) with a cascade structure, where three linear of switch positions over the considered prediction horizon.
controllers (usually PI) are employed, one for tracking the This technique has received considerable attention in recent
reference position/speed and two for the regulation of currents years and has been successfully implemented in motion control
in the direct-quadrature (d, q) reference frame. When Pulse [12][17]. FCS-MPC has been shown to significantly reduce
Width Modulation (PWM) is used, this control architecture is the switching losses and the harmonic distortions when com-
often referred to as PI-PWM. pared to standard methodologies. Unfortunately, such results
are only achieved with high control frequency (e.g., 40 kHz in
In recent years Model Predictive Control (MPC) has gained [15]), as this direct actuation couples the switching frequency
considerable attention in the field of power converters and with the controller sampling time. As a consequence, even
electrical drives [1][3]. The main idea of MPC is to obtain though the optimization problem is relatively simple to solve,
the control actions by solving at each sampling time a finite- a powerful digital controller is required to guarantee improve-
horizon optimal control problem based on a given prediction ments. Furthermore, the involved computational load scales
model of the controlled process and an estimation of its current exponentially with the MPC prediction horizon, due to the
state. At the cost of a relatively high on-line computational combinatorial nature of the problem. A review of the strategies
burden, it provides a coordinated operation of the available developed to achieve a long prediction horizon while keeping
actuators to track multiple references and satisfy bounds on the control problem tractable is presented in [18].
inputs, outputs, and states of the process. As such, it is
well suited to handle multivariable and constrained control Other MPC approaches that have been proposed for this
problems. The application of MPC in the power electronics kind of problem do not model the discrete nature of the system,
and electrical drives field is mainly motivated by two facts: the and can be formulated as convex optimization problems. In
mathematical models of these systems, needed by MPC, are this field, such strategies are referred to as Continuous Control
well known, and several constraints have to be considered in Set MPC (CCS-MPC). Several comparisons have been carried
the design of the controller. In particular, voltage and current out between FCS-MPC and CCS-MPC, and they have been
constraints must be respected due to the physical limits of shown to provide similar performances [19][21]. However,
the system and to safety reasons. MPC could be applied CCS-MPC has several advantages over FCS-MPC: it works
to other areas of electric motor control such as Electronic well with longer sampling intervals, it allows for decoupling
2309
Inverter B=
R Ts
0
RT
eAc d Bc , G = 0 s eAc d Gc , C = Cc , where
+
R
1
PWM PMSM 0 0
Ac = L Bc = L 1
- R
0 0
L L
abc "
0
#
abc
dq Gc = Cc =
1 0
dq 0 Kt
id iq L
ud - Kt s are the matrices of the continuous-time system. The LTI
PI + model (4) is used as the prediction model in the MPC problem.
uq idref At every time step k, the optimal control move for the
- ref - ref inner loop is obtained by solving the following MPC control
PI + PI + problem:
Figure 1. Drive control system with the standard PI-FOC N
X 1 u 1
NX
min kQ(yk+i|k r(k))k22 + kRuk+j|k k22 +
u
Inverter i=0 j=0
2310
Table I. MP-TC PARAMETERS
uq iq
Prediction horizon N 5
Control horizon Nu 2
Vmax Imax
Voltage limit Vmax 36/ 3 V
U X Current limit Imax 0.8 A
ud id 0.1 0
Output weights Q and P
0 0.5
0.05 0
Input increments weights R
0 0.05
Sampling Time Ts 0.3ms
2311
4000 0.03
ref PI MPC
3500
0.02
3000
2500 0.01
(rpm)
(Nm)
2000
0
1500
1000 -0.01
500
-0.02
0 l
ref PI
PI
-500 -0.03
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
time (s) time (s)
Figure 4. Rotor speed tracking with PI-FOC and MP-TC Figure 5. Mechanical and load torque with PI-FOC
where the optimal control action is not simply the solution of 0.03
the unconstrained control problem, so to stress the QP solver
and better evaluate its performance. The load-torque profile, 0.02
the reference torque and the actual mechanical torque with PI-
FOC and MP-TC are shown in Figures 5 and 6, respectively. 0.01
-0.03
troller. Performance of MP-TC and PI-FOC are quantitatively 0 0.1 0.2 0.3 0.4 0.5
compared in terms of integral square error (ISE). The proposed time (s)
MP-TC grants an improvement of 2.3% and 4.2% in speed and
torque tracking, respectively. Figure 7 shows a detail of torque Figure 6. Mechanical and load torque with MP-TC
tracking after a set-point change. It is worth noting that both
the controllers are able to handle the physical constraints of the
system. From the three-phase current plots shown in Figure 8, MPC in terms of required computational power and memory
it is clear that the drive operates close to the output constraints has been demonstrated through Processor-in-the-Loop tests
during transient phases. Figure 9 shows the times tsol needed to on a low power DSP. PIL experimental results showed that
compute the optimal control action at every sampling instant. MPC can provide performance improvements with respect
These times have been measured on the DSP with an internal to standard controllers. Future work includes extending the
clock. The results are very promising: even when the system proposed strategy to consider the speed loop and to handle
is controlled near the system constraints, the limit of 0.3ms flux weakening. Furthermore, the effect of winding resistance
imposed by the sampling interval is always respected. increase due to temperature rise will be considered to assess the
robustness of the controller. Applications of motor control by
The controller is also minimal in terms of memory al-
location. The data needed by the MPC are stored in about
2.5KB out of the 34KB of memory provided by the DSP. The
memory occupied by the code is not considered in this amount,
but it is negligible with respect to the space required by the
-0.015
QP data. Thus, all the control algorithm fits into the volatile
(Nm)
2312
[8] S. Bolognani, S. Bolognani, L. Peretti, and M. Zigliotto, Com-
0.5 bined speed and current model predictive control with inherent field-
weakening features for PMSM drives, in Electrotechnical Conference,
i (A)
0 2008. MELECON 2008. The 14th IEEE Mediterranean, May 2008, pp.
472478.
-0.5 [9] , Design and implementation of model predictive control for
electrical motor drives, IEEE Transactions on Industrial Electronics,
0 0.1 0.2 0.3 0.4 0.5 vol. 56, no. 6, pp. 19251936, Jun. 2009.
time (s) [10] S. Bolognani, R. Kennel, S. Kuehl, and G. Paccagnella, Speed and
current model predictive control of an IPM synchronous motor drive,
in Electric Machines Drives Conference (IEMDC), 2011 IEEE Interna-
0.5 tional, May 2011, pp. 15971602.
i (A)
2313