A hydrogen airship you can build in a shop and then never have to land.
Powered by sunlight, fuel cells and engines so that no single technology failure ends the flight. Sized to be lived aboard by two people for a year, with a stretch goal of five. Able to land on water and operate as a boat.
This is an open engineering notebook, not a concept renderer. Every number on this page was computed by the model at build time, by the same functions the tests call. Nothing here is typed in by hand. Where the model is guessing, it says so.
The hull
Generated from the model's own shape function, not drawn. The surface below is the one whose volume, wetted area and prismatic coefficient every figure on this page was computed from. Rings mark gas cell bulkheads; the blue band is the photovoltaic array, on the actual covered stations the energy balance integrated over.
Fly it
This runs the project's own 6-DOF solver at 100 Hz, not a simplified version for the browser. The same step function the validation gates exercise is called here, so if the vehicle feels wrong there is no second implementation to blame. Expect it to be slow to respond and slow to stop: the displaced air nearly doubles the effective mass in sway and heave.
Click the view, then W and S for thrust, arrow keys for elevator and rudder, Q and E to drop and take on ballast. Loading.
Two things it does that an aeroplane does not
It wallows when stopped and is dead-beat under way. The pitch pendulum has a period around thirty seconds and no aerodynamic damping at zero airspeed, because the fins have no dynamic pressure to work with. Above about 10 m/s the same mode is overdamped. Both are correct, the difference is large, and a control law tuned at one end will misbehave at the other.
The fins are enormous, and they have to be. Setting the fin restoring moment against the Munk moment, the dynamic pressure and the incidence both cancel, leaving a minimum area that is independent of speed and altitude. For this hull it is about 174 m². A first guess of 60 m² diverged just as surely as no fins at all: below the minimum there is no partial credit.
Move a parameter and watch what breaks
Every figure here is recomputed by the same solvers the tests and the reports use. Infeasible regions are shown as infeasible rather than as a small number: a hull that cannot lift its own structure says so, and a wind the vehicle cannot hold against turns the verdict red.
Every figure recomputed by the same solvers the tests and the reports use. A hull length change costs about 300 ms because the solar year is integrated again from scratch, so the panel dims while a result is stale rather than showing a number that is no longer current.
Does it hold up against ships that actually flew?
Unit tests catch regressions. These catch being wrong. The model is fed published geometry for every rigid airship in the reference set and has to reproduce published gross lift within a stated tolerance. Where a source contradicts itself, the fixture records the contradiction rather than resolving it quietly.
| Ship | Gas | Volume | Model | Published | Error | Gate |
|---|---|---|---|---|---|---|
| LZ-129 Hindenburg1936 | hydrogen | 200,000 m³ | 227.9 t | 232.0 t | -1.75% | PASS |
| USS Macon (ZRS-5)1933 | helium | 184,059 m³ | 184.6 t | 182.8 t | +0.99% | PASS |
| USS Akron (ZRS-4)1931 | helium | 184,059 m³ | 184.6 t | 182.8 t | +0.99% | PASS |
| LZ-127 Graf Zeppelin1928 | hydrogen | 75,000 m³ | 85.5 t | — | — | n/a |
Gas purity is first-order, not a refinement
The US Navy quoted Akron and Macon at 95 percent fill with “helium of standard purity”, which was itself about 95 percent, because helium was expensive and the Navy cared about the lift it actually had. Modelled with pure helium, the ship fails its own gate by more than twice the tolerance.
Air leaks inward through the cell film continuously, and lost purity is lost lift, permanently, unless the gas is replaced. On a vehicle whose premise is never landing, the only way to replace it is to make more. That is why onboard electrolysis is load-bearing rather than clever.
Does the loop close?
Regime A is the project's thesis: sunlight in, electrolysis to store, fuel cell to convert back, engines cold, endurance bounded by component life rather than by energy. The balance is run day by day through a year, because an annual average hides the ship that banks a surplus in June and runs a deficit in December.
Where the energy goes, annualised
The finding: energy is not the binding constraint
It is not close to being one. Lift makeup, the term that sounds like it should dominate a hydrogen airship, is 1.8% of demand. Station keeping is 83% and it is cubic in wind speed, so the real question this vehicle faces is not whether it can power itself but what weather it can live in. Whether it can carry its own structure is phase 3, and that is where the answer is likely to get hard.
| Design point | Length | Volume | Gross lift | Leak | Hold | Max wind | Regime A |
|---|---|---|---|---|---|---|---|
| Baseline | 90 m | 15,803 m³ | 18.0 t | 1.38%/a | 8 m/s | 12.6 m/s | CLOSES |
| Minimum viable | 65 m | 5,953 m³ | 6.8 t | 1.57%/a | 6 m/s | 14.9 m/s | CLOSES |
| Stretch | 120 m | 37,458 m³ | 42.7 t | 1.12%/a | 12 m/s | 11.5 m/s | FAILS |
The stretch ship does not close, and it is left that way rather than tuned until it passes. It fails on day 354 by 16 percent while showing a comfortable 12.5 percent annual surplus, which is exactly the trap a day-by-day balance exists to catch. The cause is the battery, not the hull: on the shortest day the pack saturates and the overflow spills into the hydrogen path at about a third round-trip efficiency, so those kilowatt hours cost three times as much collection.
Can it be built?
Empty weight scaled from the Hindenburg, across the range of structural scaling exponents the historical record cannot distinguish between. This is deliberately a family of curves: one curve would be a claim the evidence does not support, and the two ends disagree about whether bigger ships are better or worse.
| Envelope volume | n = 1.13 | n = 1.00 | n = 0.90 | n = 0.80 | n = 0.67 |
|---|---|---|---|---|---|
| 5,953 m³ | 33% | 52% | 74% | 105% ✕ | 167% ✕ |
| 15,803 m³ | 37% | 52% | 67% | 86% | 121% ✕ |
| 37,458 m³ | 42% | 52% | 61% | 72% | 90% |
| 80,000 m³ | 46% | 52% | 57% | 62% | 70% |
| 200,000 m³ | 52% | 52% | 52% | 52% | 52% |
✕ marks a hull that cannot lift its own empty weight. All exponents agree at 200,000 m³ because that is the Hindenburg, where the scaling is anchored.
Undecided, and the record cannot settle it
Fitting all eight rigids with published figures gives an exponent of 1.13 at R² = 0.94, which would mean the baseline closes comfortably and that mass fraction gets worse with size, not better. Restrict to the five best-sourced ships, whose volumes span only 1.41 to 1, and the fit collapses to 0.16 at R² = 0.45. The scatter from gas choice, structural material and national design philosophy is about 30 percentage points, which swamps any size trend over that range.
At the theoretical square-cube value the baseline ship cannot lift its own empty weight. A model that quietly picked the favourable end would report a comfortable design where the truth is a coin flip.
| Every rigid with published figures | Gas | Structure | Empty weight fraction |
|---|---|---|---|
| R-38 / ZR-21921 | hydrogen | duralumin | 45.0% |
| LZ-129 Hindenburg1936 | hydrogen | duralumin | 51.8% |
| LZ-126 / USS Los Angeles1924 | helium | duralumin | 59.0% |
| USS Macon / ZRS-51933 | helium | duralumin | 60.1% |
| USS Akron / ZRS-41931 | helium | duralumin | 62.1% |
| USS Shenandoah / ZR-11923 | helium | duralumin | 63.1% |
| R1001929 | hydrogen | duralumin | 67.4% |
| R1011929 | hydrogen | stainless steel | 76.9% |
Structural material moves the fraction by 9.5 points at constant size, year and specification: R100 in duralumin against R101 in stainless steel, both built to the same Air Ministry requirement in the same year. That is larger than any size effect in the dataset. Three of these eight entries were wrong in the first version of this table, all in the direction that flattered the historical fleet.
What should the engine burn?
Comparing fuels by energy per kilogram is the habit of every other vehicle and it is the wrong metric here. On an airship the scarce resource is not mass, it is lift: every kilogram of fuel aboard is a kilogram of payload that is not, and every cubic metre inside the hull is a cubic metre that is not lifting. Ranked by energy stored per kilogram of lift given up, the order inverts.
| Fuel | MJ/kg | Lift cost | MJ per kg of lift | Water recovery needed |
|---|---|---|---|---|
| Modern air-density gas blend, 46 mol% propane / 54 mol% methane | 46.5 | ~0 | 46550.0 | none |
| Blaugas, as carried by LZ-127 Graf Zeppelin | 48.0 | 0.04 | 1358.9 | none |
| Jet-A with exhaust water recovery | 42.8 | 1.00 | 42.8 | 81% |
| Hydrogen drawn from the lift cells | 120.0 | 13.40 | 9.0 | impossible |
| Hydrogen in 700 bar Type IV storage | 120.0 | 19.40 | 6.2 | 11% |
You cannot burn the lifting gas
“One gas for lift and fuel” is the most attractive idea in the propulsion module and it does not survive the buoyancy budget. Removing 1 kg of hydrogen from a cell removes 1 kg of weight and about 13.4 kg of gross lift, so the ship goes 12.4 kg heavy per kilogram burned, while combustion returns only 8.94 kg of water. No recovery fraction can hold trim, and recovering the water makes it worse rather than better: 21.3 kg heavy instead of 12.4.
A modern buoyancy-neutral blend is trivial to formulate and better than the historical one: 46.1 mol% propane with 53.9 mol% methane is exactly air density, 46.6 MJ/kg, both commodity fuels obtainable anywhere. Blaugas itself was not air density, despite what every popular source says: relative density 0.963, so consuming Graf Zeppelin’s full load made it about 1,316 kg heavier.
Which resource runs out first?
The energy balance said energy does not bind. This steps a day at a time through a multi-year mission tracking gas mass and purity, water, food and consumables, to find out what does. The answer is a legal interval, and the thing everyone expects to bind turns out not to.
| Resource, on its own | Days | Years |
|---|---|---|
| condition inspection | 365 | 1.00 |
| food | 400 | 1.10 |
| food shelf life | 5,479 | 15.00 |
Water was expected to bind. It does not.
A 90 m hull presents about 1,170 m² of plan area to the rain. In the trade wind belt at a metre of annual rainfall, even a poor 40 percent collection efficiency gathers 1,277 kg/day against a net loss of 9.3 kg/day for two people at 85 percent recycling. Catchment covers the loss 137 times over, and still covers it more than fifteen times at the most pessimistic end of every assumption.
The vehicle is water rich. Ballast is free, electrolyzer feedstock is free, and the hygiene allowance that looked like the largest lever in the life support budget is not a lever at all. That makes water a station-choice question rather than an equipment one: parked under a subtropical high instead of in the trade winds, the catchment term collapses and the whole analysis changes.
Diagnostics
The curves the design actually turns on. Shear and bending moment are drawn as two charts sharing an axis rather than one chart with two scales, because newtons and newton metres are not comparable heights and putting them on one plot invites a reading that means nothing.
Power required against airspeed
Drag goes as the square of speed, so power goes as the CUBE. Doubling cruise speed costs eight times the power, and on a vehicle whose energy comes from a fixed area of sunlight that single fact shapes the whole mission concept. This ship is slow because being fast is unaffordable, not because it cannot be made faster.
Hours of station keeping per day, against wind
How long the daily solar budget can hold position against a given wind. There is a speed above which the ship cannot hold station at all and must drift, and finding it is one of the most operationally important numbers the model produces.
Shear force along the hull
Buoyancy is distributed in proportion to cross-sectional area and weight is distributed wherever the heavy things are. Those two do not match, and the running difference is shear. The steps are the gondola, the engines and the fin roots.
Bending moment along the hull
The primary structural output, and what every laminate schedule downstream gets sized against. Warm above the line is hogging, ends down and middle up; cool below is sagging. This ship does both: it hogs forward of the gondola and sags aft of it, and it does so in still air at exact global equilibrium, because buoyancy and weight are never distributed the same way.
Where the model is guessing
Values nobody has published, with the range and what measurement would resolve each. A number without a source fails the build here, so anything genuinely unknown has to be declared rather than quietly invented. This list is the project's research queue.
Defect-dominated, as above.
Resolved by: Bench measurement.
Nitrogen is a much larger molecule and permeates far more slowly, but the ratio across a metallised barrier is dominated by pinhole defects rather than by bulk solubility, and defect density is a manufacturing variable nobody publishes.
Resolved by: Measure on the same coupon as the hydrogen test. Inward nitrogen flux is what sets the purity decay rate and therefore the electrolyzer duty cycle.
Same back-computation, same lack of primary data.
Resolved by: Not worth resolving.
Strongly dependent on ambient temperature, humidification state, and whether the stack is kept warm.
Resolved by: Measure on the selected system at the coldest expected condition.
Entirely a behavioural and equipment choice. A submarine crew manages on 12; a household figure is over 100. The range spans a factor of five and it is the largest single lever on the water budget.
Resolved by: A decision, not a measurement. Choose the washing equipment and the standard of living, then this number follows.
Depends entirely on encapsulation. A bare cell on a thin polymer backing is at the low end; anything with a glass or thick fluoropolymer front sheet is at the high end and probably not viable here.
Resolved by: Select a specific flexible module and weigh it, including adhesive and wiring. This is a purchasing decision, not a research problem.
Marine and aerospace practice both specify a glass barrier but quote thickness as a number of plies rather than in millimetres, and ply thickness varies with the cloth. Two plies of 200 gsm glass is the common recommendation.
Resolved by: Fix the cloth areal weight and the layup, then measure a cured coupon. A conservative build simply uses more plies: the mass penalty is negligible and the failure mode is not.
Varies enormously across the band and with season. The low end is the eastern subtropical ocean, the high end approaches the ITCZ.
Resolved by: Reanalysis precipitation data for the specific station, which the mission module will ingest in phase 5.
Typical rather than best-in-class. Derived from the sub-1 L/(m2 day) helium figure quoted for multi-ply polyester and Tedlar airship envelopes, scaled for hydrogen.
Resolved by: Bench measurement on a candidate coupon.
Back-computed from the "several percent per month" loss rate reported for Zeppelin-era cells rather than from any direct measurement. The material was cattle intestine laminated to cotton and its performance varied with humidity, age and how well it had been maintained.
Resolved by: Not worth resolving. This entry exists to bracket the model at the bad end and to check that the permeation physics spans three orders of magnitude.
Build order
Each phase has a validation gate. Nothing downstream of a failing gate is trustworthy, so a phase has to pass before the next opens.
- 1FoundationUnits, atmosphere, gas properties, buoyancyGATES PASS
- 2Does it close?Permeation, electrolysis, fuel cell, solar, water balanceGATES PASS
- 3Can it be built?Structure, buckling, mass fraction versus sizeIN PROGRESS
- 4Does it fly?Aerodynamics, propulsors, 6-DOF with added massGATES PASS
- 4bThe powertrain decisionFuel choice, TBO consumables, dissimilar redundancyIN PROGRESS
- 5Can it be lived in?Habitat, life support, thermal, the year-long missionIN PROGRESS
- 6Will it kill me?Hydrogen safety, lightning, icing, failure injection, regulationIN PROGRESS
- 7The siteDesign explorer, flight simulator, mission playerIN PROGRESS
- 8Build documentationFrame drawings, laminate schedules, bill of materialsPENDING