Adaptive Neural Network Feedforward Control for Dynamically Substructured Systems

The potential applications of dynamically substructured systems (DSSs) with both numerical and physical substructures can be found in diverse dynamics testing fields. In this paper, an adaptive feedforward controller based on a neural network (NN) is proposed to improve the DSS testing performance. To facilitate the NN compensation design, a modified DSS framework is developed so that the DSS control can be considered as a regulation problem with disturbance rejection. Then an adaptive NN feedforward compensation technique is proposed to cope with uncertainties and nonlinearities in the DSS physical substructure. The proposed NN technique generalizes the existing results in the literature, and it does not require any information of the plant model and disturbance model, which significantly simplifies its application on DSS. In particular, we propose a novel adaptive law for the NN online learning, where appropriate NN weight error information is derived and used to achieve improved performance. Real-time experimental results on a mechanical test rig demonstrate the improved performance by using the NN compensation strategy and the new adaptation law.