Active control of combustion instabilities on a Rijke tube using neural networks

The active control of combustion instabilities with feedback is a promising new tool. In this paper, we first describe a Rijke tube that presents, for some operating conditions, instabilities with pressure levels up to 145 dB/Hz. To control these instabilities, we have developed an internal model control scheme for nonlinear systems that uses two artificial neural networks. The first one, the internal model (IM), approximates the system forward dynamic. The second one, the controller calculates the control input. The controller's parameters are updated adaptively in real time. The IM was first trained to reproduce the burner response (given by the pressure or heat-release measurements! to open loop excitation. It was then used in the control loop to predict the response of the burner to the control action. The adaptive control algorithm used this prediction tl, update the controller's parameters. The developed controller is able to attenuate the instabilities in real time for fixed or variable operating conditions: pressure level attenuation down to -40 dB/Hz has been obtained.