Why use neural networks in control system? Conventional feedback controllers may have limitations due to design methods which involve the creation of a mathematical model to describe the dynamics of the plant to be control. In practice, the model of the plant may not represent the exact representative model of the plant due to complexity and uncertainty in the plant. As such, a controller needs to be more robust and have a wider region of operability by enhancing it with learning capabilities.
The use of neural networks in control systems can be seen as the natural step in meeting this new control methodology. Neural networks are a relatively new style of computing, and essentially they are large collections of neurons, which function by transforming particular types of input into some other type of output. When these neurons are connected together in a network, they are able to iteratively learn about the nature of the inputs, thus giving a convergent output to some desired output. This convergence is particularly useful in the area of control, where complex and nonlinear systems cannot be modeled easily, and a controller which learnt the peculiarities of the system as it operated would be very desirable. The neural network judged most suitable to control such applications was Cerebellum Model Articulation Controller (CMAC).
In this research, it is aim to