Vector control of a permanent-magnet synchronous motor requires the electrical speed and position information of the rotor. Generally, this state values information can be obtained through position sensors such as encoder and resolver. However, these position seosnors incur higher fabrication costs and need maintenance, To solve the problem, various sensorless control methods which can implement the vector control of the motor drive without position sensors have bee studied,
A Kalman filter theory to estimate state values of a motor drive without position sensor is a popular approach. This filter theory is a method to estimate optimal state values by using the Kalman gain based on error covariance recursively. However, sometimes the extended Kalman filter has a rough estimation because of the difference between a real nonlinear and linearization model. Because the high-order term of the function is neglected in the process of linearization.
Recently, a nonlinear Kalman filter except for a linearization process was proposed by Julier and Uhlman. The error between a real and linear model of the conventional Kalman filter has brought about the advent of an unscented Kalman filter. This filter algorithm combined an unscented transformation to remove the linearization process that may occur in the extended Kalman filter. This new nonlinear state estimation model is better able to strategize to estimate the mean and error covariance using properly represented values such as sigma points and weights through the unscented transformation.
Despite such a merit, the heavy computation load of this Kalman filter algorithms is a major drawback. As a result, a square-root form and a reducediorder model are widely used to solve this problem.
A bond between the square-root concept and Kalman filter is needed to mitigate the performance degradation from round-off error in the early stage of microprocess development The square-root Kalman ?lter is often used for numerical stability and can reduce memory and word length. Better numerical stability can be achieved when the error covariance matrix presents a square-root configuration.
Also, to solve the heavy code size problem, a variety of Kalman filter with reducediorder forms have been proposed. however, the previous published paper using the backielectromotive force needs an additional algorithm to estimate the electrical rotor position and speed and state estimation is not proper below 10% of rated mechanical speed.
This paper proposes the unscented Kalman filter with a Potter square*root algorithm and a parallel reduced*()rder form for state estimation of a motor drive without position sensor. The design and implementation of this filter for vector control of a permanentimagnet synchronous motor under the speed sensorless environment is investigated in this paper.
To evaluate the performance of proposed method, the designed filter With Potter algorithm was applied to the motor drive control without position seonsor of 8-pole 1[hp] permanent-magnet synchronous motor. This motor is linked to 1[hp] direct current motor for load experiment using the 2OOO[PPR] incremental encoder and resolution is four using the 150[MHZ] Texas Instruments TMS32OF2B335. The switching method and frequency are 2.5[kHz] symmetrical space vector pulse width modulation. Sampling period of the vecctor current and speed control has been selected as 200[㎲] and 1[ms], respectively. The real values such as mechanical speed, electrical rotor position and currents are measured and observed by oscilloscope.
The designed filter can reduce code computation time, and the estimation performance is analogous to the conventional Kalman filter simultaneously. The code
size of the proposed model using general unscented transformation was greatly reduced by approximately 43.11% and 38[㎲] maintaining the performance.
Simulation and experimental results such as forward and reverse rotation in high and lowispeeds(2.5~100% of rated speed), load Variation(0504->70?>50->0 and 0?>100->0 of rated torque), detuning parameters(90% of stator resistance and inductance) and noise environment are presented. The proposed model in this paper is Viable for the sensorless speed control of a motor having vector control, and the estimation performance is not significantly different relative to the previously presented Kalman filter model in spite of the reduced computation load The experimental results demonstrate that the designed filter is viable and has powerful performance