Eventos Anais de eventos
COBEM 2023
27th International Congress of Mechanical Engineering
DATA-DRIVEN FLOW RECONSTRUCTION USING LOW FIDELITY SIMULATION FOR COMPRESSIBLE FLOW PREDICTIONS IN CONVERGENT-DIVERGENT NOZZLES
Submission Author:
Allan Moreira de Carvalho , SP , Brazil
Co-Authors:
Allan Moreira de Carvalho, Daniel Dezan, Wallace Ferreira
Presenter: Allan Moreira de Carvalho
doi://10.26678/ABCM.COBEM2023.COB2023-0080
Abstract
The rise in computing power has made numerical experiments more accessible and efficient for complex problems. Machine learning methods are seen as complementary to traditional methods in processing vast amounts of data. In the field of supersonic nozzle aerothermodynamics, predicting wall heat transfer is crucial. High-fidelity methods, such as RANS, are commonly used. This study implements a surrogate model using dimensional reduction and ANNs/Kriging to map low-fidelity results to the high-fidelity method. Results show that a quasi-1D Euler solver can accurately reconstruct a 2D viscous flow. A comparison of the Neural Network and Kriging methods showed comparable performance, with Kriging being faster and more accurate. The reduction method used POD effectively reduced the data to 10 latent variables with minimal loss of information.
Keywords
machine learning, Heat transfer, Flow Reconstruction
