External cooling technologies such as transpiration cooling and effusion cooling are ideal thermal protection strategies for hot section components. Conventional cooling structures were not capable to adaptively fit non-uniform incoming temperature loads due to the limit in modelling and designing tools. The present study established an optimization workflow to adjust the hole distribution of an effusively cooled porous plate. A Conditional Generative Adversarial Neural Network model was developed to model the high dimensional and non-linear mapping between the surface profile and the surface temperature of a series of effusively cooled plates. Computational Fluid Dynamics was utilized to provide data samples for the training of the model. With careful testing and validation of the trained model, the neural network model was integrated with Genetic Algorithms to search for optimal structures that can uniformly cool the plate to a proper temperature level. Results obtained from the modeling efforts indicated a good capability of the neural network model to reconstruct the cooling effectiveness distribution on the external surface of the porous plates. Integrated with this low cost machine learning model, the GA approach successfully identified several optimized structures which fit well with the thermal loads induced by non-uniform incoming gas temperate. Surface temperature variation of the porous plates was reduced by around 50% as compared to the structure with a regular hole array. These attempts of introducing deep learning to external cooling in the present study were successful and future work could further focus on generalization of the modelling and enhancement of the robustness of the optimization approach.
Read full abstract