B.A.R.B.E.C.U.E.S. (Basically Another Really Badly Enhanced Compressible Unstructured Euler Solver) is a 2D CFD code developed in Python that solves the compressible Euler equations. The code is meant to be a testing ground for novel ideas and features to push the limits of traditional CFD codes by doing standard processes (i.e., initialization or convergence) in non-standard ways. More information about such features can be found in the Wiki (implemented "soon"). The simulation is controlled through a JSON file called config.json, and an example can be found in the source code.
The solver also has another utility that can be used to generate meshes called the G.R.I.L.S Mesher (*.GRI Level Set Mesher). The mesher is similar in architecture to the DistMesh (in that it runs the core underlying level-set based method for producing high quality meshes), but has some of its own features and changes. Some examples of how to use GRIFT can be seen in the "mesh_generation.py" script.
NOTE: There appear to be some issues regarding freestream conditions with different initialization methods and getting the flowfield to converge. It is recommended to use either "freestream" or "exp" based initialization methods.
BCs must follow this convention in order for the NUMBA JIT to properly function. The identifiers are:
| Boundary Condition | Flag Number |
|---|---|
| Inviscid Wall | 0 |
| (Supersonic) Exit | 1 |
| Supersonic Outflow | 2 |
| Supersonic Inflow | 3 |
The inviscid flux method must follow this convention in order for the NUMBA JIT to properly function. The identifiers are :
| Flux Method | Flag Number |
|---|---|
| Roe Flux | 1 |
| HLLE Flux | 2 |
If no number is specified, or an invalid number is provided, the solver will default to the Roe Flux.
There are two options for convergence: Smart and Standard. Smart convergence detects when certain solution-dependent variables (ASCs) become asymptotic and terminate the simulation early. Standard convergence uses the L1 residual norm of the Euler equations to determine convergence. The flag for this is “0” for Standard Convergence, and “1” for Smart Convergence. If using smart convergence you need a minimum of one of the following ASCs:
| ASC | Flag Number |
|---|---|
| Drag Coefficient |
0 |
| Lift Coefficient |
1 |
| Pitching Moment Coefficient |
2 |
| Average Total Pressure Recovery Factor (@ (Supersonic) Exits) | 3 |
There are 3 unique initialization methods for initializing the state variables in the cells of the domain.
| Init Method | Effect |
|---|---|
| freestream | All cells have freestream initial condition. |
| weak | The freestream Mach number is halved and then the freestream initial condition is applied. |
| linear | Uses a linear scaling function to scale the freestream Mach number based on each cell's centroid coordinate location. |
| exp | Uses an exponential scaling function to scale the freestream Mach number based on each cell's centroid coordinate location. |
| moc | Propagates characteristic lines from the inflow and reflects them off of the inviscid walls. Uses aforementioned reflections in order to create zones for oblique shock trains which can scale Mach number down according to the location within the shock train. |
V1.5.0.1 Minor optimizations to Roe Flux algorithm, also fixed issue with MATPLOTLIB figures popping up.
V1.5.0 In-code documentation updated, doc-strings fixed, etc. Additionally, GRILS has had compatability added to add the "Exit" outflow boundary condition and has had some other small tweaks to increase information during runtime.
V1.4.0 GRILS Mesher has had its documentation updated and and default values optimized using particle swarm optimization. All the things I wished to accomplish with GRILS have been done and it will be moving into a finished state. V1.5.0 will contain a manual with in-depth information regarding the program, configurations, examples, and information regarding its optimization process. Old initialization methods have been depreciated and removed, the only two remaining methods are the "freestream", "sdf", and "moc" based methods.
V1.3.1.1 Small change to flux.py to avoid extra computations resulting in ~4.5% increase in solver speed. Also fixed *.out files using the wrong value for the pressure drag when computing approximated total drag.
V1.3.1 Additional geometries added to GRILS.
V1.3.0 A new meshing utility available under GRILS, which can be used to generate basic meshes of shapes object in some flow-field. V1.3.0 can only produce flat plates and circles, and a combination of them can be modeled in a flow-field. More complex geometries are going to be added in next version.
To generate a mesh, simply execute the “mesh_generation.py“ script. The config file provided under /src/ is already set up to work with this demo mesh.
NOTE: If simulation is struggling to converge past the first few iterations with a mesh generated by this script, it is recommended to change the initialization condition. The solver runs First-Order with Forward-Euler and no limiter, as such the solver can struggle to "find its feet" while simulating the flow field.
V1.2.8 Fixed the Characteristic-line-like initialization method that initializes the initial state by running characteristic lines from the inflow to outflow and looking at possible location for oblique shock trains in the domain.
V1.2.7 Added additional initialization method (“exp”) that uses exponential scaling for scaling initial freestream state at each cell. Sometimes this is more optimal (fewer iterations til convergence) than other methods. Configured simulation to generate an “Output” directory that contains all output files. Also clarified plotting labels and fixed small error regarding output file where the Lift and Drag coefficient values were swapped.
V1.2.6 Original AMR algorithm (mesh_refinement.refine_interp_uniform()) works, also newer and better refinement algorithm (mesh_refinement.adapt_mesh()) that is fully Numba compatible, so it is both faster and produces better quality meshes.
V1.2.5 Numba JIT integration for continued speed increases. AMR & MOC-based initialization are temporarily broken.
V1.2 Conversion of simulation input parameters to a JSON format and source code reconfiguration.
V1.1 Continued bug fixes, further performance increases via vectorization of operations and further in-code documentation.
V1.0.1 Minor bug fixes, performance optimizations, and in-code documentation.
V1.0. Initial commit of the base code.