Fast estimation of internal flowfields in scramjet intakes via reduced-order modeling and machine learning

The interface between fluid mechanics and machine learning has ushered in a new avenue of scientific inquiry for complex fluid flow problems. This paper presents the development of a reduced-order predictive framework for the fast and accurate estimation of internal flowfields in two classes of scramjet intakes for hypersonic airbreathing propulsion. Proper orthogonal decomposition is employed as a reduced-order model while the moving least squares-based regression model and the multilayer perceptron-based neural network technique are employed. The samples required for the training process are generated using a sampling strategy, such as Latin hypercube sampling, or obtained as an outcome of multi-objective optimization. The study explores the flowfield estimation capability of this framework for the two test cases, each representing a unique type of scramjet intake. The importance of tuning the user-defined parameters as well as the use of multiple reduced-order bases instead of a global basis are highlighted. It is also demonstrated that the bias involved in the generation of input samples in an optimization problem can potentially be utilized to build a reduced-order predictive framework while using only a moderate number of training samples. This offers the potential to significantly reduce the computational time involved in expensive optimization problems, especially those relying on a population-based approach to identify global optimal solutions. Published under an exclusive license by AIP Publishing.https://doi.org/10.1063/5.0064724

Automated summary

This study combines fluid mechanics and machine learning to develop a new approach for fast and accurate prediction of internal flowfields in hypersonic airbreathing propulsion scramjet intakes. It highlights the importance of tuning parameters and using multiple reduced-order bases, as well as the potential of utilizing bias in building a reduced-order predictive framework in optimization problems.