Optimization

The optimizer: what may change, what must hold, and what is being minimized.

All three come out of the case file. A variable becomes free by gaining an optimize: entry where it is already declared, so a study that frees one more variable differs by three lines; constraints and the objective are their own top-level blocks, mirroring the add_constraint/add_objective calls in OpenConcept’s own optimizing examples.

Everything is checked before anything expensive starts. A cheap probe of the box is built first, and every free variable is confirmed to be an independent variable the box accepts, the objective and every constraint to be a quantity it publishes. A misspelled name is a message naming it with suggestions, not an OpenMDAO error thrown out of setup after a minute of building.

Converging before driving is not optional. The box is a Newton-solved implicit system that needs a continuation ladder to reach a long-range mission from a cold start; the driver’s first function evaluation must begin from a converged aircraft, and every later one begins from its predecessor.

exception cdadt.optimization.OptimizationError

Bases: Exception

Raised when an optimization is not well posed, before any of it is run.

class cdadt.optimization.OptimizationOutcome(baseline, optimum, constraints, objective, sense, failed)

Bases: object

The baseline, the optimum, and whether the constraints hold at it.

Parameters:

• baseline (SizingResults) – The converged design before optimization.

• optimum (SizingResults) – The converged design after it.

• constraints (list of ConstraintResult) – Every constraint evaluated at the optimum.

• objective (str) – Name of the objective, as the case file gave it.

• sense (str) – "minimize" or "maximize".

• failed (bool) – Whether the driver reported failure.

property baseline: SizingResults

The converged design before optimization.

property optimum: SizingResults

The converged design after optimization.

property constraints: list[ConstraintResult]

Every constraint, evaluated at the optimum.

property objective: str

Name of the objective.

property sense: str

"minimize" or "maximize".

property succeeded: bool

Whether the driver converged and every constraint is satisfied.

Both halves matter. A driver that reports success having stopped on its iteration limit inside an infeasible region has not solved the problem, and a report that calls that an optimum is wrong.

property violated: list[ConstraintResult]

The constraints that are not satisfied at the optimum.

property active: list[ConstraintResult]

The constraints the optimum sits on, and which therefore shaped the design.

changes()

Return name -> (baseline, optimum, percent change) for every scalar result.

Return type:

dict[str, tuple[float, float, float]]

class cdadt.optimization.Optimizer(analysis)

Bases: object

Optimize an aircraft against the constraints its case file declares.

Parameters:

analysis (SizingAnalysis) – The sizing analysis to optimize. Its case file must declare an objective and at least one variable carrying an optimize: entry.

Raises:

OptimizationError – If the case declares no objective, or if a free variable, the objective or a constraint does not address something the black box publishes.

property analysis: SizingAnalysis

The sizing analysis being optimized.

property basis: CertificationBasis

The constraints being enforced.

property design_variables: dict[str, str]

Return case-file name -> black-box path for every freed variable.

The two differ whenever a mission value is freed: cruise|h0 in a case file is mission.cruise|h0 inside the box. Resolved once, against a cheap probe, because a design variable has to be declared before the real model exists to be asked.

property objective_path: str

Return the black-box path of the objective.

property objective_units: str | None

Return the units the objective is optimized in.

prepare(verbose=False, driver=True)

Declare the study on the box and converge the baseline design.

Parameters:

• verbose (bool, optional) – Print the continuation steps and the driver’s own progress. Default False.

• driver (bool, optional) – Attach the driver the case file asks for. Passing False leaves the same declared, converged problem without one, which is what a total-derivative check needs: the derivatives an optimizer would step on, before it has stepped.

Returns:

SizingResults – The converged baseline, read before anything has been optimized.

Return type:

SizingResults

run(verbose=False)

Declare, converge the baseline, then drive.

Parameters:

verbose (bool, optional) – Print the continuation steps and the driver’s own progress. Default False.

Return type:

OptimizationOutcome

report(outcome)

Return the baseline-to-optimum comparison and the traceability matrix.

Parameters:

outcome (OptimizationOutcome)

Return type:

str