Active Casing Controlled with a Partial-Update Neural Network

Abstract

An active casing made of appropriately controlled vibrating plates can be used to reduce noise propagating from the mechanism enclosed in the casing. Since a practical vibrating casing can behave in a nonlinear way, the performance quality strongly depends on the ability of control filters to compensate for the nonlinearity. The classical approach to nonlinear active control, e.g. based on the Volterra filters, can deal with harmonics generated by the nonlinearity. However, when a complex structure is considered, neural networks have a higher potential. Although, they are much more computationally demanding, for some cases they can be simplified and still provide acceptable performance.In this paper, results of control obtained for a real casing with multiple actuators exciting each wall are presented and discussed.