Skip to main content
eScholarship
Open Access Publications from the University of California

Repetitive and Iterative Learning Control for Power Converter and Precision Motion Control

  • Author(s): Teng, Kuo-Tai
  • Advisor(s): Tsao, Tsu-Chin
  • et al.
Abstract

This thesis develops learning control algorithms for power converters and precision

motion control. The repetitive control is designed for power converters to provide

zero steady state error and harmonic compensation. A model-based iterative

learning control is designed for linear motor to track given reference profile with

sub-micron RMS error.

The objective of the control design for power converters is to compensate har-

monic distortions in the AC side to enhance power factor of the power converter.

For power inverter, implementations focus on compensating harmonic distortions

in the output AC voltage; For power rectifier, the objective is to compensate

harmonic distortions in the input AC current.

In order to compensate harmonics for power converters, the prototype repeti-

tive control [TTC89] is first being applied to power inverter in fixed frame. How-

ever, the power rectifier is not a linear system. To linearize the system at a more

meaningful equilibrium point, a D-Q transformation is applied. But the original

single-input single-output system become multi-input multi-output system in D-Q

rotating frame, the famous prototype repetitive control design mythology can not

be applied directly.

Repetitive control for multi-input multi-output system is developed for the

control of power converters in D-Q rotating frame. The coupled dynamics in

the multi-input multi-output system is first decoupled by utilizing the Smith-

McMillan decomposition. Then the prototype repetitive control design is applied

to the decoupled single-input single-output system.

In the precision motion control, model-based iterative learning control is pro-

posed to achieve sub-micron RMS tracking error. The learning fiter in iterative

learning control determines the performance in terms of convergence rate and con-

verged error. The ideal learning filter is the inverse of the system being learned.

For non-minimum phase system, direct system inversion would result in an un-

stable filter.

In this thesis, a data-based dynamic inversion method in frequency domain

is proposed. Different inversion filter was investigated in the thesis including

Zero-Phase-Error-Tracking-Controller (ZPETC), Zero-Magnitude-Error-Tacking-

Controller (ZMETC), Direct inversion, data-based phase compensator, and the

proposed data-based frequency domain inversion.

Main Content
Current View