← DeepSpace7 Paper (PDF)

DeepSpace7 · Psyche

Thrusting nearly all the time: swirl torque, wheel momentum and feed-line diagnosis for Psyche

Prometheus7 Research Institute · an independent simulation study, built from NASA's public record of the mission. Read the paper (PDF).

Since 28 September 2026 Psyche has been thrusting nearly 90% of the time. The swirl of the ions leaving its Hall thrusters puts a slight torque on the spacecraft, which the reaction wheels absorb; every few hours the team turns the thruster off, rotates, and fires another thruster at low power to unload them. NASA describes this as the first long-duration measurement of the effect on a Hall thruster in space. This study asks what onboard reasoning can add to that propulsion system: reading each thruster's torque in flight from the wheels, choosing which thruster fires and when to unload from a worst-case forecast so that thrust stays on (where the thrusters' torques differ, the unloads can nearly disappear), and, for the feed line behind it, telling a restricted line from a wrong reading within minutes. Deepspace has the technology for each of these. Every result below comes from simulation, pre-registered and run once.

92.1%
thrust duty kept while wheel momentum never passed 0.90 of capacity, against 87.0–87.9% for fixed unload thresholds
1,920 two-week runs over assumed torques (1–5× an estimate), wheel sizes and power levels
0.5%
median error reading each thruster's swirl torque in flight from the wheels
where thrusters' torques differ in sign, unloads fell from about 47 a week to 1.4
10 / 10
replays of the April 2025 pressure drop: the restricted line named and thrust back on the second line within minutes
600 single-fault runs 99.8% correct, harmful 0

The simulation

Each scenario was flown by the frozen flight code and by a comparison system, on the same simulated spacecraft and the same readings. The page plays the recorded runs back; the decision records are the system's own.

Loading the recorded runs…

Deepspace

    Comparison

      The simulated spacecraft is built from public reports: four Hall thrusters fired one at a time, about 90% duty with a weekly contact pause, the published unload procedure, 36 psi feed pressure and the 2025 drop to about 26 psi. Values with no public number (swirl torque size, wheel capacity, line hardware) are assumed and labelled in the paper, and the studies report results across ranges of them. The comparison systems are simplified baselines, not models of the mission's flight software.

      October 2026, as NASA reported it

      Psyche fires one of four Hall-effect thrusters at a time, up to 250 mN, and since 28 September 2026 has been thrusting nearly continuously, pausing about once a week for contact with Earth; it is expected to thrust nearly 90% of the time until it reaches the asteroid in 2029. It will run at full power for about three months, then gradually throttle toward half power as sunlight falls off. The ions swirl in the thruster's magnetic field, creating a slight torque that the reaction wheels absorb; after several hours of thrusting, the team turns the thruster off, rotates the spacecraft, and fires a different thruster at low power in another direction for several minutes to unload the wheels. About a third of the 1,090 kg of xenon has been used.

      Behind this sits the feed line. On 1 April 2025 the pressure in the line feeding xenon to the thrusters fell from about 36 psi to about 26 psi, and, as designed, the spacecraft powered the thrusters off. The team traced the drop to a part inside a valve on the primary line that was obstructing flow, switched to the identical backup line, and is holding the backup line's valve open. Thrusting resumed in mid-June 2025, and the arrival date held.

      Sources: NASA Psyche blog, 22 October 2026, 29 April 2025, 28 May 2025.

      Swirl torque and the wheels

      The system reads each thruster's torque in flight from the wheels, the same way it reads every other quantity: the model is calibrated from telemetry, never taken on trust. From those readings it forecasts the wheel momentum in the worst case and keeps it under 90% of capacity; folds unloads into pauses that happen anyway, such as the weekly contact; lets a thruster whose torque opposes the momentum take over instead of an unload; reads an unread thruster during an unload, when thrust is off anyway; and allows no thrust restart above 85% of capacity.

      Across 1,920 sealed two-week runs over a grid of assumed values (torque 1 to 5 times a pre-launch estimate, thrusters alike or differing in size or sign, two wheel capacities, full and half power), the wheels never passed 0.90 of capacity. Thrust duty was 92.1%, against 87.0% for a fixed 80% unload threshold and 87.9% at 90%. Where the thrusters' torques are alike, the forecast adds little over a good threshold (within 0.85%). Where they differ, reading each thruster pays: with opposite signs, unloads fell from about 47 a week to 1.4, and duty rose to 98.5%. Whether Psyche's thrusters differ in this way is not public, so all three cases are reported.

      What this means for the unload cost. How much of it onboard reasoning can take back depends on one property of the four thrusters: whether their swirl torques differ. If some swirl the opposite way, thrusters that cancel each other can take turns and the unloads nearly disappear. If they differ only in size, the quietest thruster carries the thrust and about 2% of duty comes back. If all four swirl alike, well-timed unloads are what there is to gain. The measurement that tells these cases apart is the first thing the system makes: each thruster's torque, read in flight from the wheels to within 0.5%.

      The feed line

      A Hall thruster's discharge current follows its xenon flow, so every pressure reading has a second reading to be checked against. When pressure and flow fall together, the line is carrying less than it is asked for. When pressure falls and flow holds, the reading is wrong, and both lines stay in reserve. Measured acceleration is read the same way, against the thrust the discharge should give. Each candidate response (switch line, throttle to what the line carries, stop) is flown ahead in an onboard model built from telemetry, under each reading still open, including an obstruction that is still growing, and the best worst case is chosen. Every step leaves a decision record.

      In 600 sealed single-fault runs the system was 99.8% correct with no harmful action and no damage. In 10 of 10 replays of the 2025 signature it named the primary line and had thrust back on the backup line five minutes after the drop began. When a wrong transducer reading looked like the same event, it kept thrusting with both lines in reserve in 50 of 50. A line that is partly restricted still carries flow, and as the planned throttle falls toward half power it can feed a thruster again: in a second fault on the backup line, with the restricted primary measured by a flow check, the system kept 96.8% of planned thrust with no damage, and its constraint ledger forecast the day a restricted line can feed the planned throttle, never early in 50 of 50 cases (median 40 days on the late, safe side).

      Compound faults and faults with no reading

      In 100 sealed runs of two simultaneous faults (a feed fault with a heater, wheel, star-tracker, temperature-sensor or battery fault) both were remedied in all 100, with no harmful action. Starting with the wheels at 80 to 86% of capacity and a feed line obstructing minutes later, all 30 runs unloaded first and then switched lines, with no restart above 85% and no damage (the unload began before the obstruction, so the rule that holds a restart for an unload was not itself exercised). A wrong pressure reading during an unload was marked invalid in 30 of 30, with both lines kept.

      For a fault none of its readings explains, the system does not guess: it stops thrust, holds attitude, and records the open question with the measurement that would settle it. An intermittent flameout and a slow drift that starves both lines together were each recorded as open questions in 50 of 50 runs, with no harmful action and no damage.

      Limits

      Paper and code

      Flight code SHA-256: 611e4d36e46d0b44ba60387d37380455d76aadf55c50604ffec5298a7ffd0ab2

      Request the code

      The flight code, evaluation scripts, pre-registrations, sealed results and paper are available on request. Each request is reviewed by the institute; once approved, the link you receive here becomes a download.

      An independent research project of the Prometheus7 Research Institute. Mission facts are from NASA's public Psyche blog and cited where used.