Find in Library

Many systems exist in real control applications whose characteristics are difficult to be mathematically modeled, therefore performing the design of an adequate controller is a computationally complex task using the classical methods. Alternatively, neural networks prove to be a good tool in control systems design which can be used without the need to know the exact model. This paper aims at designing a neuro-controller that combines both supervised and adaptive neuro-control schemes. The supervised scheme mimics the classical PID controller off-line; while the adaptive scheme can adapt to the system uncertainty on-line, which could eliminate the need for an exact model. The objective of the proposed neural control system is to stabilize a robot arm and the resulting robot arm angles. However, an experimental set-up of an inverted pendulum rig mounted on a cart is used as the test-bed. The simulation results prove that the proposed adaptive neuro-control scheme successfully maintained the pendulum in an upright position at steady-state.