Sketch of support vector regression (SVR). Data point inside (blue bullet) and outside (red square) the tube (gray area). The gray dashed line is the hyperplane (regression plane) predicted by (the non-linear) svrLinear. A large C enlarges the area of the tube (blue area)

Colored surface is the hypersurface (instead of hyperplane since it is non-liner). In my Report I get non-physical (negative) friction velocities predicted by svrLinear when I use too small "slack" (i.e. too large C). Thick blue line: u_τ=0

Developing turbulence models using Machine Learning in Python

• A proposed BSc project in Study Period 3, 2022-2023.
• This project is carried out by 4-6 students in the Spring. The students are supposed to spend 50% of their time on the project and at the same time they take two 7.5HEC courses.

• Description of project
• Here's an example of Python code for Machine Learning for optimizing a k-omega turbuilence model.
• Data files
  • LM_Channel_5200_vel_fluc_prof.dat
  • LM_Channel_5200_mean_prof.dat
  • LM_Channel_5200_RSTE_k_prof.dat

• Prof. Lars Davidson's homepage
  
• MTF073 CFD course
  
• Heat Transfer course
  
• Mechanics of fluids
  
• Turbulence modeling
  
• Compressible flow
• OpenFOAM
  
• PhD course: LES and DES using an in-house Python code
  
• International MSc programme in Applied Mechanics
• PhD course in CFD