01-10-2012, 11:21 AM
Genetic Neuro Controller Based Direct Torque Control for Switched Reluctance Motor Drive.
Genetic Neuro Controller.pdf (Size: 185.83 KB / Downloads: 36)
Abstract
Direct torque control (DTC) of Switched reluctance
motor is known to have simple control structure with
comparable performance. However, the role of optimal
selection of the voltage space vector is one of the weakest points
in a conventional DTC drive. In this paper optimal selection of
voltage space vectors is achieved using GA based neural
network. Simulations results validate the proposed intelligent
system with fast torque and flux response with minimized
torque and flux ripple.
INTRODUCTION
Switched Reluctance Motor (SRM), the doubly salient,
singly excited motor has simple and robust construction.
Although, the induction motor is still the workhorse of the
industries, the promising feature of the high torque to mass
ratio, high torque to inertia ratio, low maintenance, high
specific output and excellent overall performance of SRM
make it an efficient competitor for ac drives. The simplified
converter topology and switching algorithm due to the
unipolar operation avoiding shoot through faults makes SRM
advantageous in applications of aerospace, which require
high reliability. Also it finds wide application in automotive
industries, direct drive machine tools etc [1].
However, significant torque ripple, vibration and acoustic
noise are the main drawbacks of SRM to achieve high
performance. As the control of SR motor is being the recent
trend of research, schemes were developed involving linear
and non-linear models to control torque ripple [2]. But due to
inaccuracy in linear models and complexity involved in nonlinear
control, the Direct Torque Control (DTC) was
proposed which provided simple solution to control the
motor torque and speed and minimized torque ripple.
NEURAL NETWORK DTC FOR SRM
DTC is based on theories of field oriented (FOC) control
and torque vector control. Field Oriented Control uses space
vector theory to optimally control magnetic field orientation.
The DTC principle is to select stator voltage vectors
according to the differences between the reference torque
and stator flux linkage with exact value. Voltage vector are
so chosen to limit the torque and flux errors within hysteresis
bands. The required optimal voltage vectors are obtained
from the position of the stator flux linkage space vector, the
available switching vectors and the required torque and flux
linkage [9]. To drive the control scheme for the SR motor,
the non-uniform torque characteristics will first be examined.
The motor torque output can be found using the motors
electromagnetic equation.
RESULTS AND DISCUSSIONS
A Matlab/Simulink closed loop model was
constructed for the SR motor GA based neuro-DTC control
system. The motor parameters such as torque, phase flux
and position are obtained from the 3Φ SRM. Adaptive
neuro-DTC is used for voltage space vector generation is
constructed. Sampled flux and torque errors, multiplied by
weights, and the output of neuron is optimized to get the best
voltage space vector. Based on the present position of
motor, torque error and flux error the optimal selection of
voltage space vector is done with the help of neuro-GA.
Thus the converter switches and hence the motor is
controlled by DTC scheme.
In this simulation test, the motor reference flux and
torque were maintained at a constant of 0.3Wb and 5Nm
respectively. The torque results in Figs.6-8. shows lower
ripple content and constant amplitude nature for GA based
neuro-DTC control compared to classical DTC.
DTC using Neural Network
The algorithm used to train the neural network is back
propagation with momentum factor. The time taken to train
the neural network using this algorithm is 2000s. The
simulations that have been performed in this paper were
obtained using a trained state selector neural network. The
desired outputs are taken from the outputs of the
conventional DTC. Thus, the training time is basically the
time used in the simulation by the conventional DTC with
the induction motor. All training algorithms were used to
train the 3-5-3 neural-network structure using sigmoids. The
temperature coefficient of all the neurons was fixed to one,
which gives reasonable weight magnitudes.
CONCLUSION
Direct torque neuro controller trained with genetic
algorithm has been evaluated for switched reluctance motor
drive and which have been compared with the conventional
direct torque control technique. Since the conventional DTC
presents some disadvantages such as difficulties in torque
and flux control at very low speed, high current and torque
ripple, variable switching frequency behavior, high noise
level at low speed and lack of direct current control, an
adaptive torque controller is proposed for high performance
applications. In this paper, genetic algorithm based direct
torque neuro controller shows better response. By using this
controller, parameters of switched reluctance motor are also
tuned and parameter variations are much reduced. When
compared to other adaptive controllers precise results have
been obtained using genetic algorithm based direct torque
neuro controller.