Detection of Combustion Instability of Gas Turbine Combustor using Convolutional Autoencoder Model

This paper presents a method for detecting the system instability in gas turbine combustor. The proposed approach is designed as the convolutional autoencoder technique so that it offers strong attractivity even if it has a very little data. Additionally, given that it is a solution to enhance the learning effect in this system, it also provides convenience of use to practicing engineers. From these benefits, the detection rate of the system instability in the proposed method is improved while in operation, which is compared with that in both a root-mean-square and a zero-crossing approaches as well-known statistic methods.