Multivariable neural network vibration control based on output feedback

This paper presents a multivariable direct adaptive control concept for vibration suppression in flexible space structures. The adaptive controller is implemented by a combination of forward neural networks. Tuning of the controller gains (neural network synaptic weights) takes place in real-time and is performed by a nonlinear least squares algorithm. The control scheme is based on output rather than state feedback, an approach motivated from the fact that, in most applications, the system state is not readily available. The results are demonstrated by simulation using a high fidelity 6-input 6-output dynamic model of the testbed at the JPL/USAF-PL Large Spacecraft Control Laboratory