What the software does.

Optimize one parametric structural model four ways: sizing (gauges and cross-sections), shape (sketch-level geometry moves under geometric constraints), combined size + shape, andtopology (generative internal member layout). A nonlinearly-constrained SQP driver handles the smooth problems, and all gradients are analytic.

Chain the modes with optimization strategies, a compact sequencing language: rough-size, regenerate the layout, size again, then polish with combined size + shape. Exploration and refinement happen in one run instead of a chain of manual hand-offs.

Let the topology engine propose the layout instead of perturbing a template. Topological derivatives score where a new member would earn its mass. Voronoi seeding lays out candidates, with anisotropic cells that stretch along principal stress directions and pack more densely where buckling is critical. Edge-rewriting operators then merge, split, and erase members as the design converges. What comes out is an internal structure grown from the load paths, sized afterward like any conventional design.

FIG 01Generative member layout

Frame membersStructured skin mesh

Constrain the responses that drive airframe design, and get an analytic sensitivity for every one of them. Analysis runs on an in-house finite-element solver built for optimization: MITC shell elements, T-beam sections including laminates, bars and springs, isotropic and orthotropic materials. Buckling and modal problems solve as generalized eigenvalue problems, and finite-difference checks are confined to the test suite.

Mass

objective or constraint, per zone or total

Compliance

global stiffness objective

Von Mises stress

strength constraints across skins, webs, and caps

Buckling

global eigenvalue buckling

Panel buckling

local panel-level buckling, solved per panel

Displacement

deflection limits at selected locations

Natural frequency

modal constraints for dynamics and frequency-placement requirements

Thermal compliance

conduction dissipation, the thermal analogue of compliance, for heat-path efficiency

FIG 02Von Mises stress · wing skinspsi

UPPER SKIN

LOWER SKIN

40k110k180k

FIG 03Buckling field · wing skinsnormalized

UPPER SKIN

LOWER SKIN

00.51.0

Carry heat into the structural solve when temperature drives the design. Steady-state conduction couples one-way into the structural analysis, producing free thermal strains, thermal loads, and thermal stresses, and a thermal-compliance response joins the constraint set. In our documented wing example, adding the thermal load raised compliance by roughly 29% and drove the optimizer to a different sized design than the mechanical-only case.

FIG 04Temperature · wing skinsdeg

UPPER SKIN

LOWER SKIN

100200300

Author models and studies from Python, or write them as plain JSON, and validate everything before it solves. The rule the API is built around: if it passes validation, it runs in the solver and means what you said. A typo'd key or a dangling reference surfaces as an error while you're still authoring, when it costs seconds to fix.

import pgd

project = pgd.Project("lar-wing")
wing = project.model("wing")

al = wing.materials.isotropic("aluminum", E=1.0e7, nu=0.3, rho=2.59e-4)
vs = wing.sketch.points([(120.0, 0.0), (-120.0, 0.0), (70.0, -110.0), (100.0, -110.0)])
es = wing.sketch.connect([(vs[1], vs[0]), (vs[1], vs[2]), (vs[2], vs[3]), (vs[3], vs[0])],
                         labels=["root", "le", "tip", "te"])
wing.sketch.faces([tuple(es)], thickness=(0.5, 0.5))
wing.mesh(size=4)

opt = project.optimization("develop", model=wing)
mass = opt.responses.mass(scale=1.5)
comp = opt.responses.compliance(scale=4e5)
sz = (opt.problem("sz_Co", mode="size", responses=(mass, comp))
        .minimize(comp).constrain(mass))

opt.strategy.sequence(sz.repeat(10))
opt.write("inputs/develop")   # validates first; raises on errors

Take results to the viewers you already trust: Kadabra, the interactive 3D viewer, for inspecting designs and optimization history; ParaView for full field output; Jupyter for previews while authoring. Apgd validate CLI drops into CI with machine-readable reports and stable finding codes.

Model input

Plain JSON documents, schema-validated: materials, zones, loads, constraints, sketch. Authored by hand or from the Python API.

Nastran bulk data

Import existing BDF decks; export optimized models back to BDF for your downstream analysis chain.

Field output

HDF5 and XDMF/ParaView export. Stress, displacement, temperature, and buckling fields open directly in ParaView.

Meshing

Built-in gmsh-based meshing with structured, quad-dominant control, so you don’t need a separate preprocessor.

Check our claims the way you would check a stress report. Every analytic sensitivity is verified against finite differences in the test suite. The finite-element library carries shell and stiffened-panel validation cases, and element results are regression-tested against Nastran-format reference solutions. We are preparing a public benchmark set with reference solutions and error figures.

Run PGD wherever your data is allowed to live. It ships as desktop and command-line software that runs entirely on your hardware, whether that is a workstation or an air-gapped cluster node. There is no cloud dependency and no telemetry, so your geometry never leaves your network. If your work is export-controlled, nothing about the setup has to change.