Gyung-Hun Sim, “외란 관측기를 이용한 매트릭스 컨버터로 구동되는 유도전동기 구동장치의 성능개선,” 아주대학교 공학석사 학위 논문, 2008. > Paper

This thesis concerns the matrix converter as an alternative power converter for induction motor drives. The matrix converter is direct

AC/AC converter with no DC-link. The lack of reactive components in the DC-link is one of the salient advantages of the matrix

converter. Furthermore, the matrix converter features full four-quadrant operation and sinusoidal input currents. The output voltage

is limited to 87% of the input voltage. The matrix converter needs nine bidirectional switches to connect two threephase voltage

systems in all possible combinations. Most scientific work about matrix converters has so far regarded the modulation and control

of the converter. As for an induction motor drive, indirect space vector modulation strategies for the matrix converter are reviewed.

To control the motor drives using matrix converter, direct torque control scheme using space vector modulation method (DTC-SVM)

which enables to minimize torque ripple and obtain unity input power factor, while maintaining constant switching frequency is used.

However, speed control performance is still influenced by the unmodeled uncertainties of the plant such as parameter variations and

external load disturbances. Intensive research of the design of a robust stable speed controller against inherent uncertainties in the

induction motor model has been performed [6-8]. Among them, a soft computing approach using a recurrent fuzzy neural network

(RFNN) is proposed. However, a complicated RFNN structure and too many updated parameters as well as unknown design constants

can lead to a computational burden and infeasible real time implementation techniques. Recently, many kinds of soft computing

methods such as adaptive fuzzy logic, fuzzy neural networks, and recurrent fuzzy neural network have been developed in the field

of AC machine control. The radial-basis function network (RBFN) is widely used as an universal approximator in the area of nonlinear

mapping due to its performance despite a simple structure. The RBFN is architecture of the instar-outstar model and constructed

with input, output and hidden layers of normalized Gaussian activation functions. The RBFN has been introduced as a possible

solution to the real multivariate interpolation problem, because it can be used for universal approximator like fuzzy and neural

systems. However, there must be a reconstruction error if the structure of the RBFN (the number of activation functions in the hidden

layer) is not infinitely rich, and these errors are introduced into the closedloop system and make the convergence time slow, and

that, in the worst case, it can deteriorate the stability. To compensate for the reconstruction error, the method of using additional

sliding-mode like compensating input term is widely used, and its gain is computed with the information of the bounding constant

of the system uncertainty, which is difficult to obtain. In this thesis, a speed controller using the RBFN observer is proposed. The

lumped uncertainties of the induction motor system including parameter variations, external load disturbances and unmodeled

dynamics are approximated by the RBFN, and an additional robust control term is introduced to compensate for the reconstruction

error instead of the rich number of rules and additional updated parameters. Control input and adaptive laws for the weights in the

RBFN and the bounding constant are established so that the whole closed-loop system is stable in the sense of Lyapunov. Simulation

and experimental results are presented to verify the effectiveness and feasibility of the proposed control system.