The Study Of Fuzzy Control Strategies For Robot Manipulators

The intelligent control of robot has been receiving greater attention in theory circle and engineering practice in recent years. In this paper, the theory of fuzzy control was applied as key instrument to study the nonlinearty and uncertainty of robot. The main research of this paper is concluded as follows:Robot is a very complex non-linear multi-input multi-output system characterized with time-variant strong-coupling non-linear dynamics. Due to the inaccuracy of modeling and measuring, worsen by variable load and external disturbance, it is almost impossible to derive an accurate dynamics model of a robot system. In this paper, a real-time online modeling approach for T-S fuzzy model of non-linear system was proposed, which can identify the pattern of the model with an online clustering algorithm based on subtractive clustering and refine the parameters of the model with recursive least square method. Real time data collected was put into a FIFO buffer before they are filtered and fed to the process of identification, thereby rapid real time modeling can be implemented.The uncertain robot system can be decomposed into two parts: nominal part with perfect knowledge of dynamic model and uncertain part with uncertainties. Norminal part can be controlled with Computed Torque Control (CTC) while uncertain part can be compensated in two ways. One way is to adjust the compensation amplitude of variable structure controller through fuzzy reasoning according to the undertainty, to solve the dithering brought by variable structure control. The other way is to compensate the uncertainty and external disturbance of the system by fuzzy neural network, where we proposed a method of providing accurate online training signal for the fuzzy neural network. To achieve a rapid and accurate control, variable structure control is adopted in the initial phase of learning or when the deviation is considerably large, and transit between them with a smooth function.For the uncertain robot, two type of fuzzy adaptive controllers can be applied. One is indirect fuzzy adaptive control with fuzzy logic system against the unknown model function of the uncertain robot, appended with a variable structure control with saturated functions, to ensure the system stablity and reduce the dithering. The other is direct fuzzy adaptive control with fuzzy logic system in Cartesian coordinate system, appended with a robust H∞controller as compensator to attenuating uncertainties,to ensure global stability and H∞performance index.Unlike Mamdani's fuzzy reasoning, Distance-Type Fuzzy Reasoning Method (DTFR) is based on the distance between fuzzy sets, the weight of the consequent in the conclusion is calculated according to the distance between the antecedent of the rule and the fact. In this paper, we presented a concept of knowledge radius with direction, with this concept, we can select more effective rules for reasoning, which can considerably improve the speed and precision of reasoning. In solving the inverse kinematics problem of Robotic manipulators, simulation of a 2-DOF rigid Robotic manipulators shows that this algorithm is faster and more accurate compared to Adaptive Neuro-Fuzzy Inference Systems (ANFIS).Two real-time control schemes were applied to X-Y table. One is adaptive control based on online real-time fuzzy modeling using Simulink RTW (Real-Time Workshop) function. The other is client/server architecture for distributed control using TCP/IP communication. The server software is written by C++ program language which can control the robotic manipulators by using API function. The client software is programmed by Simulink, which can apply the anticipant control arithmetic. UDP protocol is used between the server and the client software.