Ph.D. Proposal: Bichuan Mo

Tue Jan 11 2022 11:00 AM
MK 317 & BlueJeans
"Jet in Cross Flow with Cylindrical Well and Simplex Swirl Injector: Large eddy Simulation and Reduced Order Modeling for Design"

You're invited to attend


Ph.D. Proposal


Bichuan Mo

(Advisor: Prof. Vigor Yang)


"Jet in Cross Flow with Cylindrical Well and Simplex Swirl Injector: Large eddy Simulation and Reduced Order Modeling for Design"



Tuesday, January 11
11:00 am
MK Building 317

Jet in crossflow is a common technique for fuel injection in propulsive devices. The present research deals with the flow dynamics of jet-in-crossflow with a cylindrical well. Detailed understanding of flow structures helps us understand where and how to use a specific jet-in-crossflow configuration in practice. In the first part of the thesis research, we study the flow dynamics of the situation with cylindrical wells by means of large eddy simulation. The effects of the well radius and depth are examined systematically. Results show that the coupling of well induced vortex and windward rolling vortices modifies the dynamics and structures of the jet flow. The windward rolling vortices become larger and even breakdown under certain conditions, and the generation frequency decreases, both of which depend on the strength of interaction between well vortex and the jet. The mixing mechanism subsequently changes from the mode of counter rotating vortex pair (CVP) mixing to windward rolling vortices mixing, thus enhancing the mixing in the near field. Spatial mixing deficiency (SMD) and temporal mixing deficiency (TMD) are calculated to quantify the mixing effects.  Results confirm that mixing is improved in the near field while remaining similar in the far field.  Compared to the well radius, the well depth exercises greater influence on the interaction between the jet and well vorticial flows, and thus is a more important design parameter.

To design a specific kind of injector, using high fidelity simulations to survey the design space is infeasible. Each high-fidelity simulation takes weeks to run using hundreds to thousands of processors, and the design space grows exponentially with the number of design parameters. Emerging deep learning techniques are promising tools to accelerate the exploration of design space by building reduced order models in the intrinsic low dimensional state space of the problem. In the second part of the thesis research, we propose to develop reduced order modeling of simplex swirl injector dynamics using an autoencoder. The autoencoder can find the nonlinear coordinate transform and reduce the problem to its intrinsic low dimension structure. Deep neural network can be trained to learn the input-output map between the input design parameters and the intrinsic parameters in the reduced state space. The new design parameters can then be mapped to the reduced state space parameters and use decoder to generate a new flowfield for the new design point. Alternatively, if the reduced state space is understandable and have some physical meanings, dynamical system, symmetry or ordinary/partial differential equations can be applied to study the evolution analytically/numerically in this reduced state space. This work relies on the capability of nonlinear dimension reduction of the autoencoder, the universal approximation ability of the deep neural networks, and the dynamical system, differential equations analytical/numerical tools.



  • Prof. Vigor Yang - School of Aerospace Engineering (advisor)
  • Prof. Joseph Oefelein - School of Aerospace Engineering
  • Prof. Graeme Kennedy - School of Aerospace Engineering


MK 317 & BlueJeans