Merge pull request OpenMDAO#651 from Kenneth-T-Moore/phase_info_revamp

Methods for level 2 revamp part 1.

aviary/core/AviaryGroup.py

@@ -0,0 +1,124 @@
+import openmdao.api as om
+from openmdao.utils.mpi import MPI
+
+from aviary.utils.aviary_values import AviaryValues
+from aviary.variable_info.enums import EquationsOfMotion
+from aviary.variable_info.variables import Aircraft, Mission, Dynamic, Settings
+
+HEIGHT_ENERGY = EquationsOfMotion.HEIGHT_ENERGY
+
+
+class AviaryGroup(om.Group):
+    """
+    A standard OpenMDAO group that handles Aviary's promotions in the configure
+    method. This assures that we only call set_input_defaults on variables
+    that are present in the model.
+    """
+
+    def initialize(self):
+        """declare options"""
+        self.options.declare(
+            'aviary_options', types=AviaryValues,
+            desc='collection of Aircraft/Mission specific options')
+        self.options.declare(
+            'aviary_metadata', types=dict,
+            desc='metadata dictionary of the full aviary problem.')
+        self.options.declare(
+            'phase_info', types=dict,
+            desc='phase-specific settings.')
+        self.builder = []
+
+    def configure(self):
+        """
+        Configure the Aviary group
+        """
+        aviary_options = self.options['aviary_options']
+        aviary_metadata = self.options['aviary_metadata']
+
+        # Find promoted name of every input in the model.
+        all_prom_inputs = []
+
+        # We can call list_inputs on the subsystems.
+        for system in self.system_iter(recurse=False):
+            var_abs = system.list_inputs(out_stream=None, val=False)
+            var_prom = [v['prom_name'] for k, v in var_abs]
+            all_prom_inputs.extend(var_prom)
+
+            # Calls to promotes aren't handled until this group resolves.
+            # Here, we address anything promoted with an alias in AviaryProblem.
+            input_meta = system._var_promotes['input']
+            var_prom = [v[0][1] for v in input_meta if isinstance(v[0], tuple)]
+            all_prom_inputs.extend(var_prom)
+            var_prom = [v[0] for v in input_meta if not isinstance(v[0], tuple)]
+            all_prom_inputs.extend(var_prom)
+
+        if MPI and self.comm.size > 1:
+            # Under MPI, promotion info only lives on rank 0, so broadcast.
+            all_prom_inputs = self.comm.bcast(all_prom_inputs, root=0)
+
+        for key in aviary_metadata:
+
+            if ':' not in key or key.startswith('dynamic:'):
+                continue
+
+            if aviary_metadata[key]['option']:
+                continue
+
+            # Skip anything that is not presently an input.
+            if key not in all_prom_inputs:
+                continue
+
+            if key in aviary_options:
+                val, units = aviary_options.get_item(key)
+            else:
+                val = aviary_metadata[key]['default_value']
+                units = aviary_metadata[key]['units']
+
+                if val is None:
+                    # optional, but no default value
+                    continue
+
+            self.set_input_defaults(key, val=val, units=units)
+
+        # try to get all the possible EOMs from the Enums rather than specifically calling the names here
+        # This will require some modifications to the enums
+        mission_method = aviary_options.get_val(
+            Settings.EQUATIONS_OF_MOTION)
+
+        # Temporarily add extra stuff here, probably patched soon
+        if mission_method is HEIGHT_ENERGY:
+            phase_info = self.options['phase_info']
+
+            # Set a more appropriate solver for dymos when the phases are linked.
+            if MPI and isinstance(self.traj.phases.linear_solver, om.PETScKrylov):
+
+                # When any phase is connected with input_initial = True, dymos puts
+                # a jacobi solver in the phases group. This is necessary in case
+                # the phases are cyclic. However, this causes some problems
+                # with the newton solvers in Aviary, exacerbating issues with
+                # solver tolerances at multiple levels. Since Aviary's phases
+                # are basically in series, the jacobi solver is a much better
+                # choice and should be able to handle it in a couple of
+                # iterations.
+                self.traj.phases.linear_solver = om.LinearBlockJac(maxiter=5)
+
+            # Due to recent changes in dymos, there is now a solver in any phase
+            # that has connected initial states. It is not clear that this solver
+            # is necessary except in certain corner cases that do not apply to the
+            # Aviary trajectory. In our case, this solver merely addresses a lag
+            # in the state input component. Since this solver can cause some
+            # numerical problems, and can slow things down, we need to move it down
+            # into the state interp component.
+            # TODO: Future updates to dymos may make this unneccesary.
+            for phase in self.traj.phases.system_iter(recurse=False):
+
+                # Don't move the solvers if we are using solve segements.
+                if phase_info[phase.name]['user_options'].get('solve_for_distance'):
+                    continue
+
+                phase.nonlinear_solver = om.NonlinearRunOnce()
+                phase.linear_solver = om.LinearRunOnce()
+                if isinstance(phase.indep_states, om.ImplicitComponent):
+                    phase.indep_states.nonlinear_solver = \
+                        om.NewtonSolver(solve_subsystems=True)
+                    phase.indep_states.linear_solver = om.DirectSolver(rhs_checking=True)

aviary/core/PostMissionGroup.py

@@ -0,0 +1,14 @@
+import openmdao.api as om
+from aviary.utils.aviary_values import AviaryValues
+from aviary.utils.functions import promote_aircraft_and_mission_vars
+
+
+class PostMissionGroup(om.Group):
+    """OpenMDAO group that holds all post-mission systems"""
+
+    def configure(self):
+        """
+        Congigure this group for post-mission.
+        Promote aircraft and mission variables.
+        """
+        promote_aircraft_and_mission_vars(self)

aviary/core/PreMissionGroup.py

@@ -0,0 +1,24 @@
+import openmdao.api as om
+from aviary.utils.aviary_values import AviaryValues
+from aviary.utils.functions import promote_aircraft_and_mission_vars
+from aviary.variable_info.functions import override_aviary_vars
+
+
+class PreMissionGroup(om.Group):
+    """OpenMDAO group that holds all pre-mission systems"""
+
+    def configure(self):
+        """
+        Configure this group for pre-mission.
+        Promote aircraft and mission variables.
+        Override output aviary variables.
+        """
+        external_outputs = promote_aircraft_and_mission_vars(self)
+
+        pre_mission = self.core_subsystems
+        override_aviary_vars(
+            pre_mission,
+            pre_mission.options["aviary_options"],
+            external_overrides=external_outputs,
+            manual_overrides=pre_mission.manual_overrides,
+        )

aviary/docs/examples/modified_aircraft.csv

@@ -49,6 +49,7 @@ aircraft:engine:num_fuselage_engines,0,unitless
 aircraft:engine:num_wing_engines,2,unitless
 aircraft:engine:reference_mass,7400,lbm
 aircraft:engine:reference_sls_thrust,28928.1,lbf
+aircraft:engine:scale_factor,1.0,unitless
 aircraft:engine:scale_mass,True,unitless
 aircraft:engine:scale_performance,True,unitless
 aircraft:engine:scaled_sls_thrust,28928.1,lbf

aviary/docs/getting_started/onboarding_level2.ipynb

@@ -46,7 +46,7 @@
     "import openmdao.api as om\n",
     "from aviary.api import Mission\n",
     "from aviary.variable_info.enums import ProblemType as PT, EquationsOfMotion as EOM\n",
-    "from aviary.interface.methods_for_level2 import AviaryProblem\n",
+    "from aviary.interface.methods_for_level2 import AviaryProblem, TwoDOFProblemConfigurator\n",
     "from aviary.utils.doctape import check_contains\n",
     "\n",
     "EOM.HEIGHT_ENERGY;\n",
@@ -56,6 +56,8 @@
     "    dummy_prob = om.Problem()\n",
     "    dummy_prob.mission_method = EOM.TWO_DEGREES_OF_FREEDOM\n",
     "    dummy_prob.problem_type = ptype\n",
+    "    dummy_prob.builder = TwoDOFProblemConfigurator()\n",
+    "    dummy_prob.target_range = 0\n",
     "    AviaryProblem.add_objective(dummy_prob)\n",
     "    dummy_prob.setup()\n",
     "    objectives = dummy_prob.model._responses.keys()\n",
@@ -654,7 +656,7 @@
     "import openmdao.api as om\n",
     "from aviary.api import Mission, Dynamic\n",
     "from aviary.variable_info.enums import EquationsOfMotion as EOM, AnalysisScheme as AS\n",
-    "from aviary.interface.methods_for_level2 import AviaryProblem\n",
+    "from aviary.interface.methods_for_level2 import AviaryProblem, TwoDOFProblemConfigurator\n",
     "from aviary.utils.aviary_values import AviaryValues\n",
     "from aviary.utils.doctape import check_contains\n",
     "\n",
@@ -683,6 +685,8 @@
     "\n",
     "for otype, obj in expected_objective.items():\n",
     "    dummy_prob = dprob()\n",
+    "    dummy_prob.builder = TwoDOFProblemConfigurator()\n",
+    "    dummy_prob.target_range = 0\n",
     "    AviaryProblem.add_objective(dummy_prob, otype)\n",
     "    dummy_prob.setup()\n",
     "    # traj timeseries values are promoted to the top in the real problem\n",
@@ -1006,7 +1010,7 @@
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "latest_env",
+   "display_name": "Python 3 (ipykernel)",
    "language": "python",
    "name": "python3"
   },