Adaptive Integral Position Control Using RBF Neural Networks for
Brushless DC Linear Motor Drive
Abstract
The paper presents an adaptive integral positioncontroller usingRBF (Radial Basis Function) neural networks (NNs)for a brushless DC linear motor. By assuming that the upper boundsof the nonlinear friction and force ripple, an RBF NN is used forapproximating the friction, the force ripple and the load; an adaptivebackstepping control with integral action is then proposed to achieveposition tracking of the linear motor. The parameter adjustmentrules for the overall controller are derived via the Lyapunov stabilitytheory. Based on the LaSalle-Yoshizawa lemma, the proposedcontroller is proven asymptotically stable. Experimental results areconducted to show the efficacy and usefulness of the proposedcontrol method.
Keyword: Adaptive control, backstepping, brushless, neuralnetworks, Radial Basis Function, DC linear Motor.I. ITRODUCTION
Over the past decades, linear motors have been widelyused for high-speed and high-accuracy applications,such as precision X −Y tables, fast manipulators,semiconductor manufacturing equipment, computernumerical control (CNC) machine tools and so on. BrushlessDC linear motor is one of the most useful linear motors forindustrial automation and precise motion drive. Generallyspeaking, brushless DC linear motor has the advantages ofsilence, simple mechanism, no dust, and high-speed andaccurate motion ability. In addition, such a linear motor hasseveral unsolved technical difficulties, as compared toconventionally rotary DC servomotors, in its nonlinear andtime-varying characteristics, such as nonlinear friction,ripple force generated bymagnetic pole and end-effects of themotor.Precise motion control for brushless DC linear motorshas attracted much attention in both academia and industry.Friction compensation is one of the key issues on highprecision motion control of such motors. The well-known LuGre friction model (Z-model) was proposed by Canudas et al.[1], which could describe many friction’s behaviors and wasuseful for various control tasks.Adaptive tracking control with neural networks for themotor with the friction has been investigated by severalresearchers. Based on the LuGre model, Caundas and Ge [2]proposed an observer-based adaptive friction compensationcontroller to deal with system’s position / velocitydependency characteristics; they used a multi-layer GaussianRBF neuralnetwork to approximate the unknown function α (x, x