Resumo:
Nowadays, induction motor drive uses vector-control to get faster answer torque. To
use flux estimation, direct measurement can be used with Hall sensors or another
measurement technique or flux estimation by measurement of the stator voltage and current.
The direct flux measurement is expensive and the process accuracy may not be enough
and yet stator flux estimation process eliminates flux and speed sensors, decreasing cost and
augment system reliability.
For stator flux estimation this work uses the strategy of programmable cascaded low pass filter (PCLPF), implemented by recurrent-neural-network and training with Kalman
Filter. The PCLPF method permits ideal voltage integration, from extremely low frequency to
high frequency field-weakening range. Implementation of the filter, based on neural network
is simpler with good performance and presenting faster performances by means of DSP
(Signal Digital Processor). The use of the Kalman Filter as an RNN training algorhyth has
shown good results as far as data quantity and total training time and concerned.
Besides the measurement of the stator voltage and current, the motor parameters
necessary for flux estimation using the direct vector control oriented through the stator flux, is
the impedance equivalent to the stator winding of with the resistenace is significant.
This work presents the stator resistance estimation, using Extended Kalman Filter,
making torque and stator flux estimation more accurate.
Later on, estimation of other parameters of an induction motor will be conducted, such
as: simultaneous rotor and stator resistance and rotor inductance, by using the concept of EKF
and also the rotor speed and resistance by means of the RNN and the EKF training.
The estimations proposed above have been confirmed by means of the simulations
results.
