System dashboard
Live orbital view
How the layers connect: the Thermal tab sizes the radiator from GPU temperature and IT power (optionally driving the area used everywhere else); the Collision Avoidance tab computes the self-collision workload from that area, the population and the orbits; the External Debris tab adds the environment the fleet does not control. All tabs share one parameter set (Parameters & Assumptions tab).
Parameters
Collisions/yr vs (log-log)
Collision-rate density vs latitude (geography of encounters)
Impact-speed spectrum (collision-weighted)
Two-phase active cooling: GPU temperature drives radiator size drives collision risk
The thermal-collision tradeoff: hotter GPUs, smaller radiators, fewer collisions
Radiator configuration (3D, set in Parameters; drives ADCS inertia)
Thermal system sizing (per satellite)
| Component | Sizing | Value |
|---|---|---|
| Cold plates / evaporators | ΔT junction→coolant at full load | – |
| Two-phase loop (pump, lines, accumulator) | ~0.8 kg per kW rejected + pump power ~1.5% of Q | – |
| Condenser / radiator panels | A × areal density | – |
| Total thermal subsystem mass | – | – |
| Specific mass | per kW of IT power | – |
Two-phase pumped loops keep the transport nearly isothermal, so the radiator runs close to the junction temperature minus two interface drops; that is what makes the T&sup4; lever effective. Raising the junction from 70°C to 105°C cuts the radiator (and the collision cross-section) by roughly a factor 2; the price is paid in silicon reliability, leakage power, and pump margins. Enable "drive A from Thermal tab" in Parameters to propagate this area through the whole model.
Power subsystem: photovoltaics, eclipse storage, and the second big deployed area
Deployed area and collision burden vs IT power per satellite (current links)
Spacecraft (power subsystem highlighted)
The PV array is the second large deployed surface, and in eclipse-prone orbits it exceeds the radiator. The collision analysis nominally counts radiator area only; enabling the PV-area coupling adds the array to the cross-section, which is the conservative reading (Turyshev's 1 MW anchor carries 5.6 m²/kW of PV against 2.5 m²/kW of radiator). Dawn-dusk SSO removes the battery and shrinks the array, which is exactly why that orbit is thermally and electrically attractive; the Collision Avoidance tab prices what it costs.
Attitude Determination and Control: the radiator span sets the control problem
Disturbance torques vs altitude (300 km band, current configuration)
Actuator suite (3D, choose in Parameters)
Design alternatives
| Suite | Mass | Peak power | Notes |
|---|---|---|---|
| 4× reaction wheels + 3× magnetorquers | – | – | Baseline LEO. Wheels absorb the cyclic gravity-gradient momentum; torquers dump the secular part each orbit (dipole A·m²). |
| 4× control-moment gyros + 3× magnetorquers | – | – | For large momentum/agility; pays a fixed mass and complexity premium, wins above ~80 N·m·s. |
| 4× reaction wheels + thruster desaturation | – | – | Replaces torquers where the dipole gets impractical; costs kg propellant per year. |
Sizing logic: the deployed radiator dominates the inertia asymmetry, so gravity gradient is the design driver; T_gg = 3/2 n² ΔI sin(2θ). The cyclic part loads the wheels every orbit (H ≈ 0.637 T_gg T_orb/4); aero, solar pressure, and the residual dipole accumulate as secular momentum that must be dumped. Configuration matters: the four-panel and z-fold layouts shorten the span, cutting ΔI and the whole actuator chain; that is mass the thermal design can trade against. Mass/power figures are SMAD-class regressions for screening, not catalog data. On torque authority: despite the 120 m² radiator the inertia is Hubble-class (~half of Hubble's 87,000 kg·m²), and Hubble flies on reaction wheels. Wheels reject the 13 mN·m disturbance with order-15× torque margin; what large inertia actually costs is agility (tens of minutes per 90° slew, irrelevant for a sun/nadir-fixed compute satellite) and controller bandwidth below the flexible modes of 15 m wings, which this screening model does not cover. At worst-case 45° gravity-gradient attitudes the momentum requirement reaches the top of the wheel class and the recommendation switches to CMGs automatically.
External debris environment (screening estimate, calibrated to public ORDEM/MASTER/ESA figures)
Fleet-level rates vs altitude (300 km band centered at each altitude)
Method: flux = n(h, >d) × v_rel; hits/yr = N × A_eff × flux. For lethal non-trackable (1-10 cm) debris A_eff is the radiator face area A (it cannot be dodged; only damage tolerance helps, hence the critical-area fraction). For trackable objects (>10 cm) the encounter cross-section is σ = shape·A and the events feed the same avoidance chain as internal conjunctions. The density profile is an order-of-magnitude envelope (peak near 750-900 km, traffic bump near 550 km); a bankable assessment requires native ORDEM 3.1 / MASTER-8 runs with the real geometry. The 700-900 km region combines high debris density with very long residence times: least attractive for a large fleet.
Validation against Turyshev, arXiv:2604.27197 (computed at the fixed reference case, independent of the sliders)
Reference case: N = 80,000, A = 120 m², σ = 4A, 500-800 km, mix 43/53/70/97.6°. Benchmark: S. G. Turyshev, "Orbital Data Centers: Spacecraft Constraints and Economic Viability", arXiv:2604.27197 (May 2026), an independent power/thermal/economics closure model built at JPL. The comparison covers the shared thermal and power sizing layer; Turyshev does not model fleet self-collisions.
This model vs Turyshev: our value over his anchor (dots), his published ranges (gray bands), 1.0 = his anchor or range midpoint
Comparison detail
| Quantity | This model (defaults) | Benchmark B | Comment |
|---|---|---|---|
| Radiator area per IT kilowatt | – | 2.5 m²/kW (1 MW anchor) | Same physics; ours runs hotter (T_rad ) and radiates both faces, his anchor is more conservative. |
| Radiator temperature | – | 300-400 K design range | Inside his range. |
| PV area per IT kilowatt (dawn-dusk) | – | 5.64 m²/kW (BOL, 1 MW high-sunlight anchor) | Independent reproduction of his PV sizing. |
| Radiator areal density | – | 2-10 kg/m² | Inside his range. |
| Thermal-subsystem mass per IT kW | – | ~5-25 kg/kW (radiator share of his 29.4 kg/kW PV+storage+radiator floor) | Consistent. |
| Fleet scale context | – | 17-46 GW IT → 170,000-461,000 nodes at 100 kW | The reference fleet sits inside his envelope. |
| Collision-risk treatment | Explicit fleet-level model: rates, geography, spectrum, cascade κ | Aggregated hazard rate λ_MMOD inside the lifetime penalty Π_life | Complementary: this model supplies the physics behind his λ_MMOD and the correlated cascade risk his independent-hazard framework cannot represent. |
Benchmark B does not model fleet self-collisions, so the comparison is on the shared thermal/sizing layer and on consistency of treatment. The two models close the loop on each other: B shows the economics are gated by mass and link costs even before collision physics; this model shows collision physics adds constraints B assumes away.
How to use the parameters
The sidebar on the left is the single source of truth for every tab. Two optional couplings tie the subsystems together: "drive A from Thermal tab" makes the radiator area follow GPU temperature and IT power, and "count PV area in collision cross-section" adds the solar array to the collision target. With both enabled, dragging the GPU temperature slider propagates end to end: junction temperature → radiator area → collision cross-section → fleet collision rate → cascade margin.
Assumptions register
| Assumption | Validity / limitation |
|---|---|
| Dilute gas, Poisson statistics | Mean free path exceeds shell thickness by ~10^7; individual event probabilities small. |
| Circular orbits, no perturbations | J2 only accelerates the RAAN randomization the model already assumes. Eccentricity shifts densities by km; negligible on rates. |
| Randomized RAAN and phase | Exact for ensemble means. Coordinated slotting or LTAN clustering (all dawn-dusk) violates it; not modeled. |
| σ = shape factor × A | 4A = enveloping sphere (conservative); physical band A…4A; all rates linear in σ. |
| Inclination dispersion ±0.5° | Total rate insensitive; only the F_sp × F_vel split depends on it (logarithmically). |
| Cascade branching κ | Warning index, not an evolutionary debris model (use LEGEND/MASTER for that). Avoidance does NOT apply to κ. |
| External debris profile | Order-of-magnitude envelope calibrated to public ORDEM 3.1 / MASTER-8 / ESA figures. NOT a certified run; use the scale slider for the uncertainty band. |
| Thermal model | Steady-state two-phase loop; radiator radiates from both faces to an equivalent sink; transients, shadowing and degradation excluded. |
| Power model | Beta-0 worst-case or dawn-dusk eclipse bounds; PMAD and battery as SMAD-class regressions; degradation via a single EOL fraction. |
| ADCS model | Screening-level torque budget and actuator regressions; flexible-body dynamics, jitter, and slew agility excluded. |
| f_fail | End-to-end fraction of dangerous internal/trackable encounters not mitigated, including dead satellites; lethal non-trackable debris cannot be avoided at all. |