Bridge engineering, carried into orbit.
Three disciplines from long-span bridges decide whether a large space structure survives launch, opens to the right shape and holds it: shell stability, membrane form-finding and staged erection. None can be proven at full scale on the ground. All can be verified by analysis, which is what bridges have done for a century.
Which of the methods that let us certify untestable bridges transfer to structures that can never be tested whole — and where, precisely, do they stop?
- Tip error
- 89 mm
- As fraction of span
- 1.04 %
- Rigid-body model
- Check against flexible
- Worst through travel
- 1.10 %
Basis
Declared design in the class of PSE-TM-02: a yoke turning 90° and three panel folds turning 180°, panels 2.5 m, deployed length about 8.5 m. Planar, four hinge lines.
The ideal chain is exact kinematics — each hinge angle interpolates linearly from stowed to deployed. The real chain adds, at every joint, a fixed free-play offset of the stated magnitude with alternating sign, plus a friction lag proportional to how much travel remains, scaled by a per-joint factor so the lines do not move in step. Tip error is the distance between the two tip positions, reported against the 8.5 m deployed span.
The thresholds drawn on the error curve are the measured acceptance criterion reported in PSE-TM-02: a rigid-body deployment analysis of a flexible array agreed with a flexible finite element model to about 1 N·m while tip deflection stayed under roughly 1 % of span, and diverged by an order of magnitude for every doubling beyond 2 %.
Omissions
Kinematics only. No inertia, no damper, no spring torque, no latch transient, no structural flexibility in the panels or the yoke — the links here are rigid and the error shown is joint error, not the elastic deflection that the 1 % criterion was measured against. Free play is applied as a fixed offset rather than a load-dependent deadband, friction as a kinematic lag rather than a torque. Planar, so no out-of-plane or torsional behaviour. This shows the shape of the divergence and where the published threshold sits; it is not a deployment analysis.
A three-panel array wing unfolding on its hinges. The pale chain is ideal kinematics; the crimson chain is the same wing with free play and friction in every joint. Watch the two separate, and watch the separation cross the threshold at which a rigid-body deployment analysis stops being defensible.
The transfer
Long-span bridges are certified, not tested. The methods that make that possible were built over a century for structures that could never be proof-loaded. Three of them transfer, almost unchanged, to structures that can never be tested in one gravity.
Shell & plate stability
The margin against buckling in a thin-walled structure, and the failure mode that governs launch survival for a packaged aperture or boom. Where full-scale test is impossible, the knockdown-factor lineage space inherited from SP-8007 is decades old and conservative to the point of cost. A calibrated stability prediction is the alternative.
Membrane form-finding
Finding the equilibrium shape of a tensioned surface under its own pretension — the analysis behind cable-net roofs, carried to sunshields, solar sails, antenna nets and gossamer reflectors. The same mathematics that shapes a stadium roof shapes a deployed membrane, and the surface accuracy it predicts is the surface accuracy the mission depends on.
Staged erection
Every partial configuration a structure passes through on its way to complete, each analysed for stability, load path and geometry before the next is added. On a bridge it is called erection engineering. In orbit it is deployment and assembly, being rediscovered without the century of literature that already answers most of it.
Verification by analysis is the thread through all three. Each answers a question no rig on Earth can close — will it buckle, will it hold its shape, will it survive the sequence — and answers it the way a bridge does: a documented analytical argument that states its own uncertainty and its own limits, carried under professional liability.
The problem
Gravity is the reason the test cannot be run.
Large space structures are packaged into a fairing and open to many times that size. At full scale, in one gravity, they sag under their own weight by more than the tolerance they must hold. Offload rigs help and introduce errors of their own: on real programmes, the difference between one suspension arrangement and another has exceeded the effect being measured.
One declared 60 m welded plate girder. Move gravity continuously and watch the required depth, the governing limit and the failure mode change. The crossover is the point of the exhibit: stiffness governs under weight, dynamics governs without it.
- Required depth
- 2.36 m
- Steel mass
- 1006 kg/m
- Governing limit
- Deflection
- Crossover
- 4.15 m/s²
- Moves only with the frequency floor, as its square — not with carried mass.
Basis
Declared design, not a reconstruction. Simply supported span 60 m, S355 plate (fy 345 MPa, E 200 GPa, ρ 7850 kg/m³). Welded I-section proportioned from the web depth: web thickness d/140 (minimum 10 mm), flange width d/3 (minimum 300 mm), flange thickness bf/12 (minimum 20 mm).
Four limits are solved for required depth and the largest governs. Deflection 5wL⁴/384EI ≤ L/300. Bending σ = M(h/2)/I ≤ fy with M = wL²/8. First mode (π/2)√(EI/mL⁴) ≥ the stated floor. Fabrication minimum 1.0 m overall. Loads carry a factor of 1.5; the frequency check uses unfactored mass.
At the crossover the deflection and first-mode limits bind at once. Set equal, the mass, the modulus and the second moment cancel, leaving only span, load factor and the frequency floor:
g* = 1.024 · f12 · L / (π2 · γ) = 4.150 m/s² at f1 = 1 Hz, L = 60 m, γ = 1.5
The exception is a floor below the first frequency of the 1.0 m minimum section — 0.93 Hz with no carried mass, 0.35 Hz at 1200 kg/m — where the fabrication minimum takes over first and the crossover moves with carried mass. Above 1.70 Hz, g* lies beyond the slider and the first mode governs across the whole range.
Omissions
No lateral-torsional buckling, no web or flange local stability, no stiffener design, no shear check, no fatigue, no construction stage, no thermal case, no connection design, no resistance factors beyond the single load factor. Section proportioning is a declared rule, not an optimisation. Read the output as a ranking of which limit governs, not as a design.
Earth sizes this girder on deflection. Remove gravity and both strength and deflection fall away, but the first-mode requirement does not move, because stiffness-to-mass carries no g. Below roughly 4 m/s² the dynamic limit takes over and nothing improves after that.
That is the shape of the argument the notes make in three other places. The quantity everyone names is rarely the quantity that governs.
So the structure is qualified in pieces and joined by analysis, and the analysis carries the argument. What it rests on depends on the structure — the stability of a thin section against buckling, the pretension that sets a membrane's shape, the load path through every partial configuration, and for a jointed deployable the stiffness and free play of the joints themselves. These are among the least documented quantities in the field and the ones that decide the answer.
The test cannot close the loop
What is demonstrated on the ground is not what flies. The gap has to be crossed analytically, with the crossing itself justified.
The joint decides the outcome
Stiffness, free play and preload govern deployed shape and dynamics, and are rarely on a drawing, rarely measured, and rarely in the model.
Confidence is usually asserted
Models agree with each other, checks pass, and nobody asks which physics both models are missing. Agreement across a blind spot is not evidence.
Notes
Five technical notes on public data: three studies and two reviews.
- PSE-TM-01
- PSE-TM-02Rev A — 2026-10-15
- PSE-TM-03Rev A — 2026-10-15
- PSE-TM-04
- PSE-TM-05
Field guide
A working review of the literature behind the method.
- NASA SP-8007 — Buckling of Thin-Walled Circular Cylinders
- Pellegrino (ed.) — Deployable Structures
- NASA-STD-5001 / 5017 and ECSS-E-ST-32
About
I design complex and long-span bridges, and have spent twenty years on structures certified by analysis because they cannot be proof-loaded. This is where I test, in public, whether those methods transfer to large space structures.