Abstract
A new approach based on convolutional neural network (CNN) technology is proposed for efficient inverse design of nozzle that can be designed according to the required pressure distribution. The regularization term is proposed to improve the smoothness of the nozzle configuration generated by the CNN. Firstly, the details of CNN technology are introduced. Secondly, the details of the numerical method are presented, including code validation and gird resolution study. Thirdly, the nozzle dataset is obtained using the design of experiments method. Then, the regularization for nozzle contour smoothness is considered and the weight of the regularization term is discussed. Finally, the nozzle cases are predicted to show the effectiveness of the proposed model. The results show that the proposed model can obtain the required nozzle configuration accurately, and the mean square error and mean absolute error are and , respectively. Besides, the smoothness of the nozzle configuration can be improved using the proposed regularization term in the training process. The average relative errors between the required and predicted pressure distribution are less than 10% under satisfying smoothness. Therefore, the nozzle configuration can be predicted using the proposed model under the required pressure distribution.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call