mdolab / openaerostruct Goto Github PK

Not all of the new VLM components have derivative unit tests yet.
Copy the form from other tests for these components

Move this to Python: https://github.com/mdolab/OpenAeroStruct/blob/master/openaerostruct/transfer/displacement_transfer.py#L116

Might be able to use Dr. Hwang's version as a base: https://github.com/hwangjt/OpenAeroStruct/blob/v2/openaerostruct_v2/aerostruct/disp_transfer_group.py

One of the thickness values doesn't change during the optimization

Cloning from master branch, and adding the following conditions (before the call to OAS_prob.run():

OAS_prob.prob['alpha'] = -5.4772415561327445
OAS_prob.prob['wing.thickness_cp'] = np.array([ 0.14340327207903325, 0.3639890838470242, 0.2653579033716302])
OAS_prob.prob['wing.twist_cp'] = np.array([ -8.541536161392285, 5.030094299899314, -4.063335448175808])

This returns a solution with

Fuelburn: -74795.1319562
wing_perf.failure -0.315245694701
wing_perf.thickness_intersects [-0.08346824 -0.28824473]

That is, the thickness doesn't intersect, the wing doesn't fail, but the fuelburn is negative.

We can vectorize the assembly of the matrix and also use dot and cross products from OM library

Given that the new implementation (vectorized, fully Python) implementation works well and matches all previous tests, we now need to remove all of the old code

Some objects that need to be recorded are not pickleable and throws
RuntimeWarning: Trying to record options which cannot be pickled on system with name: twist_bsp. Use the 'options_excludes' recording option on system objects to avoid attempting to record options which cannot be pickled. Skipping recording options for this system.

Produce and visualize sectional cl curves in both the streamwise and chordwise directions

Copy physics from Sham's wingbox branch: https://github.com/shamsheersc19/OpenAeroStruct/tree/mpt_wingbox for the Korn equation

Getting plot_wing to work when run_model() is called instead of run_driver(), for example to view a user-specified mesh.

Currently it's a dense Jacobian formulation for the partial derivatives.
Switch to using a sparse formulation; should be a quick chane.

Korn equation, etc, from https://github.com/shamsheersc19/OpenAeroStruct/tree/mpt_wingbox

Better vectorize the matrix assembly and also get derivatives on python level

Translate this part to Python:
https://github.com/mdolab/OpenAeroStruct/blob/master/openaerostruct/structures/vonmises_tube.py#L104

Can probably use Dr. Hwang's version from here as a base: https://github.com/hwangjt/OpenAeroStruct/blob/v2/openaerostruct_v2/structures/components/fea_vonmises_comp.py

This would be for engines, batteries, etc.
Will need to consider how to add these to the structural model correctly (which nodes, etc)

In analyzing wing-tail configurations it may be useful to set the tail twist to a scalar, such that it's only used for trimming. This throws an error in plot_wing:

  File "/home/neil/hg/OpenAeroStruct/openaerostruct/utils/plot_wing.py", line 314, in load_db
    new_twist.append(np.hstack((twist[0], twist[0][::-1][1:])))
IndexError: invalid index to scalar variable.

Use balance comps as needed to make a trimmed drag polar

Get analytic derivs for this: https://github.com/mdolab/OpenAeroStruct/blob/master/openaerostruct/functionals/moment_coefficient.py#L141

Write this part in Python: https://github.com/mdolab/OpenAeroStruct/blob/master/openaerostruct/transfer/load_transfer.py#L108

Might be able to use this as a base:
https://github.com/hwangjt/OpenAeroStruct/blob/v2/openaerostruct_v2/aerostruct/load_transfer_group.py

Currently, the amount of fuel burn for the 2.5g structural case is too high because it's supposing that we're flying the entire mission at that loading.
In reality, we should use the 1g fuel burn case for the mass added in all cases.

OAS does have many tests, but there are some key files and lines not being tested fully.
Look at the reports at https://coveralls.io/github/mdolab/OpenAeroStruct and find files that don't have good coverage (they have a low % covered).
Add tests mimicking existing tests to cover these files.

Running the aerostructural test case gives

NL: NLBGS Failed to Converge in 50 iterations

even when the number of iterations is bumped to 100. Might need a little bit of digging.

Additionally, the test finishes with a relative error of E-3 rather than the required E-8. Not sure what's different between local and travis tests.

Want to remove this part and make it Python-only:
https://github.com/mdolab/OpenAeroStruct/blob/master/openaerostruct/aerodynamics/geometry.py#L155

Can use this as a base:
https://github.com/hwangjt/OpenAeroStruct/blob/v2/openaerostruct_v2/aerodynamics/components/mesh/vlm_normals_comp.py

Cloning from master branch, and adding the following conditions (before the call to OAS_prob.run():

OAS_prob.prob['wing.thickness_cp'] = np.array([ 4.5279479027781128e-01,1.9036800154838401e-01,9.8321106961832772e-02])
OAS_prob.prob['alpha'] = -4.0948666786020738e-01
OAS_prob.prob['wing.twist_cp'] = np.array([ -3.2162315517514468e+00,7.1314196111774741e+00,-3.2470369225802571e+00])

Crashes with

Traceback (most recent call last):
  File "run_aerostruct.py", line 132, in <module>
    OAS_prob.run()
  File "/Users/brendan/Documents/SFI/codeother/openaeroversions/aerotest/OpenAeroStruct/run_classes.py", line 587, in run
    self.prob.run_once()
  File "/usr/local/lib/python3.6/site-packages/openmdao/core/problem.py", line 1170, in run_once
    driver.run_once(self)
  File "/usr/local/lib/python3.6/site-packages/openmdao/core/driver.py", line 786, in run_once
    system.solve_nonlinear(metadata=metadata)
  File "/usr/local/lib/python3.6/site-packages/openmdao/core/group.py", line 739, in solve_nonlinear
    self.nl_solver.solve(params, unknowns, resids, self, metadata)
  File "/usr/local/lib/python3.6/site-packages/openmdao/solvers/solver_base.py", line 23, in wrapper
    fn(driver, params, unknowns, resids, system, metadata)
  File "/usr/local/lib/python3.6/site-packages/openmdao/solvers/run_once.py", line 51, in solve
    system.children_solve_nonlinear(local_meta)
  File "/usr/local/lib/python3.6/site-packages/openmdao/core/group.py", line 759, in children_solve_nonlinear
    sub.solve_nonlinear(sub.params, sub.unknowns, sub.resids, metadata)
  File "/usr/local/lib/python3.6/site-packages/openmdao/core/group.py", line 739, in solve_nonlinear
    self.nl_solver.solve(params, unknowns, resids, self, metadata)
  File "/usr/local/lib/python3.6/site-packages/openmdao/solvers/solver_base.py", line 23, in wrapper
    fn(driver, params, unknowns, resids, system, metadata)
  File "/usr/local/lib/python3.6/site-packages/openmdao/solvers/nl_gauss_seidel.py", line 132, in solve
    system.children_solve_nonlinear(local_meta)
  File "/usr/local/lib/python3.6/site-packages/openmdao/core/group.py", line 759, in children_solve_nonlinear
    sub.solve_nonlinear(sub.params, sub.unknowns, sub.resids, metadata)
  File "/usr/local/lib/python3.6/site-packages/openmdao/core/group.py", line 739, in solve_nonlinear
    self.nl_solver.solve(params, unknowns, resids, self, metadata)
  File "/usr/local/lib/python3.6/site-packages/openmdao/solvers/solver_base.py", line 23, in wrapper
    fn(driver, params, unknowns, resids, system, metadata)
  File "/usr/local/lib/python3.6/site-packages/openmdao/solvers/run_once.py", line 51, in solve
    system.children_solve_nonlinear(local_meta)
  File "/usr/local/lib/python3.6/site-packages/openmdao/core/group.py", line 757, in children_solve_nonlinear
    sub._sys_solve_nonlinear(sub.params, sub.unknowns, sub.resids)
  File "/usr/local/lib/python3.6/site-packages/openmdao/core/component.py", line 609, in _sys_solve_nonlinear
    self.solve_nonlinear(params, unknowns, resids)
  File "/Users/brendan/Documents/SFI/codeother/openaeroversions/aerotest/OpenAeroStruct/vlm.py", line 973, in solve_nonlinear
    self.lup = lu_factor(params['AIC'])
  File "/usr/local/lib/python3.6/site-packages/scipy/linalg/decomp_lu.py", line 58, in lu_factor
    a1 = asarray_chkfinite(a)
  File "/usr/local/lib/python3.6/site-packages/numpy/lib/function_base.py", line 1215, in asarray_chkfinite
    "array must not contain infs or NaNs")
ValueError: array must not contain infs or NaNs

Currently the derivatives for these components are computed using cs.
We may need to split up these components to make computing the analytic derivatives easier.
Evaluate the complexity and feasibility of reorganizing these components.

Currently this component only has analytic derivatives for the non-symmetric wing; need to add analytic derivatives for symmetric wing case

Ensure that all tests pass and compare the timings to see the hit that we take for using Python only

Currently throws

Exception in Tkinter callback
Traceback (most recent call last):
  File "/usr/lib/python2.7/lib-tk/Tkinter.py", line 1540, in __call__
    return self.func(*args)
  File "/home/neil/hg/OpenAeroStruct/openaerostruct/utils/plot_wing.py", line 654, in auto_ref
    self.check_length()
  File "/home/neil/hg/OpenAeroStruct/openaerostruct/utils/plot_wing.py", line 631, in check_length
    self.num_iters = int(db.keys()[-1].split('|')[-1])
IndexError: list index out of range

Currently does not work but also does not raise any errors.

Follow previous-created resources to obtain the induced drag using Trefftz plane analysis. Example of resource: http://docs.desktop.aero/appliedaero/potential3d/InducedDrag.html

Numpy's assert_almost_equal doesn't fail with NANs while OM's does so switch to those throughout

Certain tasks, such as writing the mesh to .xyz format for Tecplot, have helper scripts floating around. It would be good to collect/consolidate them, add documentation, and ship them with OAS.

Not sure how important this is, but I received a ComplexWarning:

/home/neil/hg/OpenAeroStruct/openaerostruct/geometry/utils.py:74: ComplexWarning: Casting complex values to real discards the imaginary part
  row[:] = np.einsum("ikj, ij -> ik", mats, row - quarter_chord)

The Jacobian stays fixed for this component, but the derivatives are currently computed in compute_partials, which is unnecessary.

Set up the Jacobian once in setup() and set the derivatives there.

We are unnecessarily allocating some Jacobian information when we declare all partials this way.

Do a find all in the project and change all instances of these to use the correct declare, such as declare_partials('A', 'thickness'), etc.

If the component requires further work or sparsification to make it good, create a new issue for that.

The docs were kind of stitched together from a variety of ideas at different times.
Go through these docs holistically and update / rearrange as necessary.

We currently have analytic derivs for this comp, but they are dense and use multiple Python for loops.

Instead, we want to use a sparse formulation and use vectorized math to get the derivatives.

Most of the code ported over doesn't have documentation; go through and add explanations for each component

This component will probably need to be broken up into multiple smaller components

https://github.com/mdolab/OpenAeroStruct/blob/master/openaerostruct/geometry/geometry_mesh.py#L158

Certain codes raise Error which does not seem to be a defined python Error, which raises NameError when raised. Found in geometry/utils.py but potentially in other locations too.

We've introduced a large number of VLM components and they're flat in the aerodynamics folder.
They can be better named and organized.

For the workshop:

• Need simple script to show how to check derivs of a component
• Use view_mat to help visualize jacobians
• Have sparse and dense examples

Currently using CS