Fuzzy logic and neural network based advanced control and estimation techniques in power electronics and ac drives

This dissertation presents, examines and analyzes advanced control and estimation techniques in power electronics and ac drives by the systematic application of fuzzy logic and neural network in several power electronics systems. Such artificial intelligence tools were investigated for several reasons which will be clarified in the coming chapters, including the fact that fuzzy logic provides a convenient and user-friendly front-end for developing control programs, helps designers concentrate on the functional objectives. In addition, neural networks have the capability of capturing system's behavior by leaming the input/output relationship, thus relieving designers of laborious modeling.

A variable speed wind generation system was developed, where fuzzy logic principles were used for efficiency optimization and for performance enhancement control. A squirrel cage induction generator fed power to a double-sided PWM converter system which pumped power to a utility grid or to an autonomous system. The system has three fuzzy controllers. The controller FLC-1 searches the generator speed on-line so that the aerodynamic efficiency of the wind turbine can be optimized. A second fuzzy controller, FLC-2, programmed the machine flux by on-line search so to optimize the machine-converter system efficiency. A third fuzzy controller, FLC-3, performed robust speed control against turbine oscillatory torque and wind vortex. The project of the hardware, software design, and the experimental evaluation are covered in detail.

Next, a neural network was applied for estimation of feedback signals in an induction motor drive, which has some distinct advantages when compared to DSP hased implementation. A feedforward neural network received the machine terminal signals at the input and calculated flux, torque and unit vectors at the output, which were then used in the control of a direct vector-controlled drive system. The system was operated in the wide torque and speed regions independently, with DSP-based estimator and neural network-based estimator, and were shown to have comparable performances. The neural network estimator had the advantages of faster execution speed, harmonic ripple immunity and fault tolerance characteristics when compared to a DSP-based estimator.

The application of fuzzy logic to the estimation of power electronic waveforms was taken into consideration for distorted line current waves in a TRIAC light dimmer and in a three-phase diode rectifier feeding an inverter-machine load. Fuzzy logic estimation was applied to assess the rms current, fundamental rms current, displacement factor and power factor. Both the rule base and relational approaches were used for estimation of the above parameters. The estimated values were then compared with the actual values, indicating good accuracy. It was demonstrated that fuzzy estimation gave considerably faster response than the conventional hardware or software computation method, which may be very important for dynamically-varying system conditions.

Finally, the development of a speed and flux sensorless vector-controlled induction motor drive was considered, primarily aimed for electric vehicle type applications. The Stator flux oriented drive started at zero speed in indirect vector control mode, transited to direct vector control mode as the speed developed, and then transited back to indirect vector control at zero speed. The vector control used stator flux orientation in both indirect and direct vector control modes with the stator resistance variation compensated by measurement of stator temperature. The problem of integration at low stator frequency was solved by cascaded low-pass filters with programmable time constants. The control strategy of the four-quadrant drive was analyzed, validated by simulation study, and finally evaluated by experimental study on a laboratory 5 hp drive system.