A New Adaptive Method of Component Maps in Gas Turbine Thermodynamic Model Based on Spatial Transformer Networks

Gas turbine performance degradation is a common phenomenon that affects the efficiency and reliability of gas turbines. To simulate the degraded performance maps of gas turbine compressors, a novel method based on spatial transformer networks (STNs) is proposed in this paper. STNs are a type of neural network module that can learn to perform spatial transformations on input images, such as cropping, scaling, and rotating, which are used to learn the transformation coefficients for mapping the standard performance maps to degraded ones. The training data is generated by applying random affine transformations on the standard maps and sampling pixel points on the transformed maps to simulate the operating conditions of degraded gas turbines. The loss function is defined a s the mean squared error between the predicted and actual degraded maps. The experimental results show that the proposed method can predict the degraded performance maps of gas turbine compressors from a few sampled points, and can be used for fault diagnosis and prognosis. Further related work will be implemented in the near future.