Modeling pneumatic servovalves using neural networks

This paper presents a study where static models of the flow stage of a 3/2 pneumatic servovalve are obtained using artificial neural networks. For simulation purposes, a direct model is proposed in which the mass flow is determined for a given working pressure and command input. For control purposes, an inverse model is proposed in which the command input is determined given the working pressure and the desired mass flow. This approach enables the use of mass flow as the synthesised control action, thus rendering the overall pneumatic system affine. Therefore, control techniques requiring this condition on the system model can be directly used. Under normal working conditions, both servovalve models provide an excellent agreement with experimental results taken from an industrial pneumatic servovalve