Discrete linear control enhanced by adaptive neural networks in application to a HDD-servo-system

The performance of a linear, discrete high performance track following controller in a hard disk drive is improved for its disturbance rejection by the introduction of a discrete non-linear, adaptive neural network (NN) element. The neural network element is deemed to be particularly effective for rejection of bias forces and friction. Theoretical, simulation and experimental results have been obtained. It is shown theoretically that a NN-element is effective in counteracting these non-linear, system-specific, model-dependent disturbances. The disturbance, i.e. the bias and friction force, is assumed to be unknown, with the exception that the disturbance is known to be matched to the plant actuator input range and the disturbance is an (unknown) continuous function of the plant output measurements. For a non-linear simulation model and a laboratory HDD-servo system, it is shown that the NN-control element improves performance and appears particularly effective for a reasonably small number of NN-nodes.