51 analytical inlet module

[nlaing613]

This PR adds the first fully operational version of the inlet_moc module and an associated test case (emami_inlet.py). It is not ready to directly integrate in the same manner as simple_inlet_model.py, but it works nicely for this test case and the code is large. I reason it'd be better to start this sooner than later.

Sources: rotational_solvers.py is adapted from Zucrow and Hoffman, Volume II, Chapter 17. The irrotational shock-net algorithm is documented in SUPIN and Anderson NASA TN D-4960. I extended the irrotational shock algorithm for rotational solvers, where irrotational flow is a special case of rotational flow. This should make the solver more accurate and robust, as shocks violate the irrotationality assumption (Crocco's theorem).

Ruff formatting/checks were also run before opening this PR.

Note: throughout the code, the np.ndarray variable pt is used. This always contains x, y, V, theta, p, and rho. I am open to changing this to a dataclass, but since points are constantly called from the net, repackaging them seems wasteful.

Module Structure

• moc_solution.py

  • Main bookkeeping object.
  • Processes the initial condition and input geometry.
  • Calls for creation of the first characteristic net.

• initial_domain.py

  • Contains helper functions that identify the inlet leading edge.
  • Constructs FlowCells containing the freestream, post-shock state, and initial data line (IDL).

• solver_events.py

  • Identifies incident shock locations in nets.
  • Handles shock reflection.

• planar_inlet.py

  • Geometry class using math module interpolation to return the y location of the two walls, centerbody and cowl, and their slopes at a given x.

• char_net.py

  • Represents a characteristic net consisting of x, y, V, theta, p, and rho.
  • Contains functionality to self-solve after being fed inlet information and an IDL, only for the initial net.

• flow_physics.py

  • Contains a variety of helper functions for perfect-gas and isentropic-flow relations.

• rotational_solvers.py

  • Contains field_point_rot, the solver for the intersection of C+ and C- characteristics, reconstructing the C0 streamline.
  • Contains wall_point_rot, the solver for the intersection of the wall (C0) and two characteristic points.

• utils_moc.py

  • Holds small helper functions that minimize NumPy usage for speed.
  • Includes clipped interpolation, vectorized linear algebra implementations, and a piecewise-linear plus tangent-line intersection function.

Shock Solvers

• shock_solvers/char_shock.py

  • Holds the main shock functions that determine the position and state of upstream net points (net_L) and downstream net points (net_R) that satisfy oblique shock relations and characteristic/compatibility equations.
  • Also contains oblique-shock helper functions and the ShockPoint class.

• shock_solvers/net_logic_helpers.py

  • Assortment of helper functions used to identify valid net points under indexing constraints for interpolation or the next solve.
  • Responsible for clipping coalescing same-family characteristics to retain a valid solution.
  • Constructs valid net boundaries and removes spurious points after resampling or net-shock solves.
  • Returns active point segments for the requested characteristic type, which are fed into net_shock.py.

• shock_solvers/net_shock.py

  • Main marching algorithm that propagates a shock through the upstream net_L.
  • Constructs and resamples the downstream net_R to retain user-specified resolution.

Processing

• processing/processing.py

  • Master plotting and averaging script after the solution terminates.
  • Allows direct and analytical modes, with user-specified sampling resolution for analytical mode.

• processing/triangulated_solution.py

  • Consolidates nets and cells into a triangulated vertex-centered point mesh.
  • Creates an interpolation object that returns primitive variables along a line in y.

• processing/get_streamtube_bounds.py

  • Uses the processing mode and triangulated solution to find external flow-region bounds used in cross-sectional averaging.

• processing/cross_sectional_average.py

  • Uses the triangulated solution, input Cantera gas object, and streamtube bounds to compute stream-thrust averaged quantities.

Plotting

• plot/net_shock.py

  • If plot_during_solve is specified, generates a real-time plot of the net-shock marching process.
  • Mainly used for debugging; open to removal.

• plot/solution.py

  • Master plotting script.
  • Uses the triangulated solution and inlet to generate 2D contour plots of requested variables in the solution domain.
  • Also generates plots of stream-thrust averaged quantities.

• plot/helpers.py

  • Stores axis labels, names, y-ranges, and other plotting/formatting utilities.

Example Case

• examples/moc_testing/emami_inlet.py
  • Runs the NASA Mach 4 case with user-specified x_stop and IDL resolution.
  • Plots two-dimensional primitives plus Mach number and temperature.
  • Plots stream-thrust averaged comparison with refined SU2 Euler solve results.
  • Plots capture bounds computed in get_streamtube_bounds.py versus the analytical function provided in the original paper.

Next Steps

• Wire the inlet solver more directly into StanShock’s geometry infrastructure.

  • In particular, the wall-boundary characteristic solves need access to local wall slope, not just wall y values; thus, the StanShock geometry module could require extension (to return dy_dx).

• Isolate the thermodynamic model.

  • The perfect-gas assumption is currently baked into several characteristic, shock, and helper relations; I'd rather use the StanShock fluid state.

[TimothyEDawson]

Awesome, taking a look at it now! Would you mind if I make minor changes as I review? I'll avoid modifying the functionality and focus on code style, variable names, and type annotations.

[nlaing613]

@TimothyEDawson I don't mind at all, thanks for helping clean up the admittedly lengthy code!

[TimothyEDawson]

@nlaing613 Alrighty, I've got a few things to clean up but overall it's looking pretty good. I did notice the hyshot_inlet example case crashes, so if you want to look into that in the mean time please feel free. If it's going to take a while to fix no worries, we'll merge this in and work on that in a subsequent pull request.

[TimothyEDawson]

I'm good with merging this in as-is and fixing the rest later, but some misc. comments on things to improve going forward:

• I don't typically use the math module when we already use NumPy so extensively. I'd rather just use NumPy exclusively.
• I recommend avoiding unpacking NumPy arrays, like when you unpack x, y, V, theta, p, rho = pt. There are several options for rewriting this sort of logic, but it's hard to recommend one without fully understanding the overall flow of data.
• The logic and type annotations can be substantially simplified by avoiding unnecessary uses of None. Examples:
  • For invalid cases, raise an error instead of returning None. E.g. in char_net.point_mask, is there any valid path through the calling program where a returned None would be valid?
  • In NetShockSolver, are self.net_R or self.net_R_test ever intended to be None? Many type errors disappeared by simply not initializing them with None values, i.e. assuming they will always be of type CharNet.
• I see invalid calls to wall_point_rot in CharNet - looks like you added the wall and family input arguments to the signature but didn't update the calls here?