Study on non-intrusive monitoring method of flow rate distribution within parallel cooling channels based on GAN-like model

Abstract The prediction of flow distribution in regenerative cooling channels of scramjet can provide valuable reference information for flow regulation. The non-intrusive monitoring method based on deep learning is a promising approach. In this work, a generative adversarial networks-like (GAN-like) model is proposed, where the generator and discriminator are employed for temperature field reconstruction and flow distribution prediction respectively. The generator utilizes the sensor data to reconstruct the temperature field of the combustor outer wall, while the discriminator employs the generated temperature field to forecast the flow distribution within the parallel channels. The trained GAN-like model exhibits a commendable capability in predicting temperature field features and flow distribution states under the current dataset. The generator attains remarkable proficiency in reconstruction, evidenced by a structural similarity index surpassing 0.95 and a correlation coefficient exceeding 0.96. Additionally, it showcases an unforeseen aptitude at the boundary location. The discriminator exhibits stable precision in flow rate prediction, as indicated by an absolute error below 0.02 g/s and a relative error lower than 3%.