Two-dimensional prediction of the superposition film cooling with trench based on conditional generative adversarial network

Abstract

Nowadays, the film cooling strategy is widely adopted to cool down turbine blades and significantly extend their service life. To protect the turbine's structural integrity while improving its efficiency, the effectiveness of air film cooling needs accurate evaluation in engineering. This study investigated a classical flat-plate model with a trench hole under multi-row superposition conditions. However, conventional semi-empirical correlations are hard to predict well. Thus, a Conditional Generative Adversarial Network (CGAN) model was developed to predict the two-dimensional film cooling effectiveness based on the film cooling parameters. The CGAN model consists of a Gated Recurrent Units (GRU) network generator and a Convolutional Neural Network (CNN) discriminator. The depth and width of the trench, compound angle, hole location, and blowing ratio information are mapped to the flow history information. Based on the sequential information, the GRU generator predicted the wall film cooling effectiveness, while the CNN discriminator focused on identifying generated image being real or fake. The adversarial training process significantly improved the accuracy of the generator, and the generator is capable of predicting the wall film cooling effectiveness well.