Artificial neural network based control of electromechanical systems

In the present paper a neuro-adaptive controller is developed for the nonlinear control of a multivariable motor drive. A hierarchical architecture of the ANN is proposed for the design of the controller, which contains two subnetworks. The first one (NNM) is used for motor identification and the second one (NNC) for adaptive control. Both of these employ error backpropagation rules. It is an advantage of the applied neuro-controller that it does not require a reference model or an inverse model. In the neuro-controller, the identification of the motor is performed by mapping the input-output of the motor, and the adaptation of the controller is obtained by minimising the difference between the reference speed and the mapped speed. A minimum variance self-tuning control is also implemented, and it is compared with the ANN control. The simulation results show that the proposed identification and adaptive control scheme is practically feasible. Details of the practical implementation are discussed, together with important theoretical aspects. The neuro-controller has been successfully implemented in real time using the Texas Instruments TMS320C330 digital signal processor. The minimum variance self-tuning controller has also been implemented on the same DSP.