1. The biots that kept the ship

Arthur C. Clarke’s Rendezvous with Rama returns as the recalled work because Rama’s maintenance is handled by biots — simple, specialized machines that clean, repair, and tend the vast cylinder. They do not think; they observe and act. The desktop’s predictive maintenance layer could be similar: a set of specialized watchers, each tuned to one kind of degradation, that trigger predefined responses.

This entry asks what predictive maintenance needs from the desktop.

2. Sensors that earn their keep

Predictive maintenance starts with data. The desktop would need:

  • Electrical: voltage, current, and temperature at the board, cell, and attachment level.
  • Mechanical: vibration, motor current, position error, and cycle count for moving parts.
  • Thermal: temperature maps across cells, radiators, and interfaces.
  • Storage: corrected/uncorrected memory errors, bad-block counts, and write-cycle counts.
  • Power: solar-array current and voltage, battery voltage and temperature, charge/discharge cycles.

Not every component needs every sensor. The NASA small-satellite health monitoring guide warns that sensors must buy their way on board through cost avoidance.

3. Compute and storage budget

Predictive models need somewhere to run and somewhere to store history. Options include:

  • Onboard: real-time, low-latency, but limited by radiation-hardened compute and power.
  • Ground: unlimited compute, but delayed and dependent on downlink.
  • Hybrid: onboard trend detection plus ground model refinement.

For the desktop, a hybrid approach is attractive. The platform runs lightweight anomaly detection and change-point algorithms onboard. It downlinks compressed feature histories. Ground systems run heavier RUL models and upload updated thresholds or models.

4. Operational integration

Predictions are useless unless operations can respond. The desktop would need:

  • Maintenance windows: times when a cell can be taken offline without losing mission function.
  • Redundancy policies: which components have hot spares, warm spares, or no spares.
  • Work orders: clear procedures for autonomous or ground-initiated swaps.
  • Data retention: enough history to train and validate models across missions.

5. What this changes

  • Predictive maintenance is an architecture decision, not just an algorithm choice.
  • The desktop must design in sensors, compute, storage, and operational procedures together.
  • The next entry will close the predictive maintenance wondering arc and record the decision.