Session: 04-13 Combustion Modeling I

Paper Number: 123963

123963 - Generative Design of a Gas Turbine Combustor Using Invertible Neural Networks 

The need to shift to carbon free powergeneration requires a high invest in the development of gas turbine combustion systems for the whole power range. Combustion systems capable to burn fuels with 100% H2  content need a complete redesign of existing designs due the high reactivity of H2/air mixtures. The application of generative learning show great promise for accelerating the design process and for scaling of design features among different engine frames.  Generative learning has caught much attention in image and text generation. In this paper we will show how generative learning can be utilized for generating gas turbine combustors.

Generative learning trains the model to predict geometric parameters of the combustor design dependent on performance labels as combustion system pressure drop, fuel/air unmixedness and thermoacoustic stability. Since this is an inverse design approach Inverse Neural Networks which have emerged in image generation in the last years have been selected as the AI method. The paper will outline the parametrization of the design and the opportunities and limits of the generative design method.  The inverse neural network creates a mapping of the desgn parameter distribution to the performance label distribution. This mapping is invertible and hence can be used in both directions and hence provides a forward prediction of performance labels and an inverse prediction of the design. The method has been applied to a single jet flame combustor on a generic level.  

 

Presenting Author: Patrick Krueger Institute of Mathematics TU Berlin

Presenting Author Biography: Patrick Krueger holds a Master Degree in Mathematics from Bergische Universität Wuppertal. He is working on a project on Generative AI.

Authors:

Patrick Krueger Institute of Mathematics TU Berlin
Hanno Gottschalk Institute of Mathematics TU Berlin
Bastian Werdelmann SIEMENS Energy
Werner Krebs SIEMENS Energy