| Summary: | 碩士 === 逢甲大學 === 電機工程學系 === 107 === The aim of this thesis is to design a motor drive based on the combination of neural network and genetic algorithm. The designed drive system can adjust the control parameters that suffer to the variations of temperature or the load changes. This results in the better performance compared with that of the traditional PID control motor drive.
A Radial Basis Function Neural Network (RBFNN) based PID control is adopted in the control loop of the designed motor drive such that it can adjust the gain parameters in real time according to the learning characteristic of the neural network. The initial parameters of the neural network are unknown. In this thesis, the Matlab software is used to perform the genetic algorithm (GA) and the RBFNN network to modify the radial basis function neural network parameters. When the output meets expectations, Put its parameters into the motor control system. In addition, a development environment provided by Microsoft Visual Studio 2017, was employed to develop a user interface by using C# language.
For the hardware system design, this thesis adopts the microcontroller TM4C123GH6PGE as the core controller which is manufactured by Texas Instruments Incorporated (TI).A current sensor is fed to the ADC converter of the microcontroller. The value of current is used to perform current feedback control using the sensed current value. A Digital to Analog Converter (DAC) circuit is also designed to display the current value on the oscilloscope, which is convenient for the designer to observe the change.
At the beginning of the experiment, the Matlab/Simulink simulation software is used to analyze the feasibility of the three-phase six-switch control for the brushless DC motor. After confirming, the motor driver designed in this thesis is manufactured. Then, the operation effect of the motor under different control modes is detected. The control modes are (1) open loop control, (2)speed PID control, (3) speed current PID control, (4) RBFNN speed PID control, ( 5)RBFNN speed and current PID control, (6) offline gene algorithm combined with RBFNN speed PID control and (7) offline gene algorithm combined with RBFNN speed and current PID control, a total of 7 control methods. Two different rotor detection methods are used under speed PID control, sensorless circuit and Hall element, and comparing the difference between the two-phase conduction and the three-phase conduction modes. In addition to the feasibility of testing the control method, the remaining experiments also placed two different control modes in the same speed and current calculation to compare the differences between the two, such as PID control and RBFNN PID control, and whether there is a RBFNN PID control using optimized neural parameters.
The experimental results show that the control system using the general RBFNN PID needs a relatively long response time. This verifies that the the better parameters obtained from offline genetic algorithm in the neural network. It also shows that the motor does not need to take lots of learning time after the start and can quickly reach the target.
|
|---|
