Research On Intelligent Sliding Mode Variable Structure Control For Multi-link Robots

Adaptability and robustness are the basic control characteristic for the trajectory tracking control of multi-link robots. Sliding mode variable structure control (SMVSC) is an efficient approach because of its strong robustness against disturbances and variation of parameters. However, its application and development are limited by the chattering of the SMVSC systems. In order to alleviate chattering, merging SMVSC with intelligent control are efficient approaches, such as fuzzy sliding mode control, neural sliding mode control, fuzzy neural sliding mode control based on genetic algorithm, and so on. These approaches are studied and developed in this paper:1. Various control structures for multi-link robots are studied and a parallel control structure is presented in this paper. The parallel control structure has two merits: (1) It can take full advantage of the known information about robots, which gives the equivalent control and reduces the learning time of intelligent control for the uncertainty. (2) The advanced techniques, such as embedded systems and computer communication, are applied to increase the response speed of the control systems. The control requirement can be distributed to several microprocessors that are connected each other by control networks. Therefore, every microprocessor has only simple task, fast learning algorithm and control speed. Also, the network control structure has virtues of both centralized control structure and distributed control structure. If one of the control subsystems is damaged, the other control subsystems can continue to work and the reliability of whole control system is developed. This virtue of distributed systems is convenient to detect and remove system trouble. The parallel working ability can develop robustness of the whole system.2. Designing approaches and general approximate characteristic of fuzzy systems are studied. Two novel control approaches are advanced: one is a fast direct adaptive fuzzy sliding mode control for multi-link robots with model errors; another is a fast indirect adaptive fuzzy sliding mode control for multi-link robots with uncertainty. The fuzzy systems are applied to approximate the unknown parameters instead of estimating them in advance. Therefore, the controller can adaptively adjust its parameters with respect to parameter changes of the controlled system. The chattering of sliding mode control is alleviated without sacrificing the system robustness. The stability of the controller is proved by using Lyapunov direct method. The errors trajectory of the control systems is analyzed in detail. It is proved that the system errors are only relative with the approaching errors of fuzzy systems and independent of the disturbances. So the control systems have small stable errors and strong robustness.3. Designing and learning approaches of neural networks are studied. A novel global neural sliding mode control based on radial basic function is proposed for multi-link robots with model errors and uncertain disturbances. The neural networks can learn the upper limit of model errors and uncertain disturbances. So the chattering of sliding mode control is reduced. The cost function of neural networks is deduced according to Lyapunov stable theory, which can ensure that the control systems are stable.4. The fuzzy neural networks which have both merits of fuzzy control and neural networks control are studied. A sliding mode controller based on self organizing fuzzy neural networks is designed. The controller combines the structure learning and parameters learning. The fuzzy rules are selected by competing method according to the system required accuracy. The controller parameters are adjusted by gradient descent algorithm. So the chattering of sliding mode control is eliminated. The cost function of neural networks is deduced according to Lyapunov stable theory, which ensures that the control systems are stable.5. Genetic algorithm is studied and applied to a fuzzy neural sliding mode controller for multi-link robots without known information. The designed steps of the control system are presented. At first, the controlled system mathematic model must be identified by a neural network. Then, genetic algorithm is applied to train the fuzzy neural networks out systems according to the system mathematic model. At last, the fuzzy neural networks are applied to real time systems which adaptively adjust parameters in systems by gradient descent algorithm with respect to the controlled system parameters. So the chattering of sliding mode control is eliminated without sacrificing its robustness. In order to avoid the slow convergence and the immature convergence of genetic algorithm, an adaptive genetic algorithm is proposed, which can realize to optimize the whole parameters range, develop the optimizing efficiency, and be suitable to complex control systems.