M.S. (Master of Science)
Department of Electrical Engineering
Fuzzy systems||Adaptive control systems||Robots--Control systems||Robotics||Manipulators (Mechanism)
This thesis provides an original design idea for a novel control system structure as well as the control algorithm, which combines a neural network with fuzzy logic, and succeeds in dynamical compensation for variant uncertainties, which includes both structured and unstructured uncertainties. A new fuzzy reasoning method is derived in the neural mechanism and implemented with CMAC (Cerebellar Model Articulation Controller), which outperforms the normal fuzzy controller by reducing the computational complexity and providing learning ability which conventional fuzzy systems don?t have. The whole control system is proven to be stable both theoretically and practically. The stability proof is given in this thesis in detail. The simulation test results confirm that the system can intelligently and stably track the desired position in both set point tracking and dynamic tracking in presence of various uncertainties, such as changing payload, various frictions, unknown disturbance, etc. while all conventional fuzzy control systems fail to maintain good performance. This thesis is organized as follows: In Chapter 1, the background knowledge about robotic dynamics and related properties is reviewed as the basis for the control system design. A Computed Torque Controller is also briefly described as a typical conventional control scheme in order to provide a comparison reference for the designed system. In Chapter 2, the basic concepts of the fuzzy control logic are introduced, including fuzzy membership function, fuzzy inference engine, defuzzification, etc. It is also shown that these fuzzy logic systems are universal approximators capable of approximating any nonlinear function over a compact set to arbitrary accuracy. In Chapter 3, the design idea, structure, and implementation algorithms of a novel Neural Fuzzy Control System are presented and derived in full length. A new term ? Neural Fuzzy Inference? is created in this chapter to explain the method of using a Neural Network (the controller I) to implement fuzzy reasoning in order to provide a natural mechanism for membership function optimization and rule conflict guarding. The selflearning algorithm of the controller II in the control system structure is also derived and described in detail. The whole Neural Fuzzy System aims to provide outstanding performance in controlling the non-mathematically modeled multi-link robotic manipulator and overcome both the structured and unstructured uncertainties. The stability proof of this designed neural fuzzy control system is also provided in this chapter. Finally, in Chapter 4, the designed neural network control system is used to carry out six simulation cases, and these results of the simulations are reported and analyzed. For the purpose of comparison, each simulation case is repeated for the conventional control system under the same circumstance, such as disturbance, friction, payload changing, etc. Finally, the conclusion of this thesis is drawn out of the simulation results and analysis.
Peng, Limin, "Design and implementation of a novel neural fuzzy control system for a multi-link robotic manipulator" (1999). Graduate Research Theses & Dissertations. 2089.
ix, 110 pages
Northern Illinois University
Rights Statement 2