Controller design for a nonlinear servomechanism in the presence of system uncertainties using multilayered neural network (Q5960366)

The nonlinear proportional plus integral controller for the nonlinear servomechanism is designed. It is made up in fact of two controllers: proportional plus integral feedback controller to eliminate the steady-state error due to a step input function and the controller to compensate for nonlinear servo mechanism uncertainties and random disturbances. The nominal system is controlled by the classical linear optimal regulator (LOR). The steady-state output of the plant is designed by means of a feedback neurocontroller. For the transient state output of the plant, the feedforward neurocontroller is used. First, the proportional plus integral optimal control of the linear servo problem is formulated. Such a linear system is taken as the nominal system. The nonlinear model is then formulated as a composite of such a linear nominal system and ad hoc introduced nonlinear uncertainties. As a simulation example of such a servo system, a two-dimensional inverted pendulum mounted on motor-drive is considered. For the simulation, layered neural networks with one hidden layer of 20 neurons and a different number of input \((7;6)\) and output \((5;1)\) neurons for the above-mentioned neurocontrollers were used, respectively. The results show naturally that a nonlinear plant dynamics becomes unstable with an increasing range of added uncertainties. The proposed approach has many limitations when applied to practical problems. The point is that it is almost impossible in general to decompose a nonlinear real plant into a linear part and some additional small nonlinearities. So approaches in the field are going in just the opposite way. The neurocontrollers, realized by neural networks, are used mainly for the adaptive control of nonlinear plants, including servomechanisms, in the case where a dynamics of such plants is not known explicitly.