Predictive Maintenance

Turning telemetry into scheduled maintenance so faults are fixed on plan, not on failure - the strategic lever on availability.

From reactive to predictive

Maintenance can be reactive (fix on failure), preventive (fix on a fixed schedule), or predictive (fix when telemetry says a component is heading toward failure). Predictive maintenance is the highest-leverage resiliency strategy because it converts unplanned outages - which hit availability hardest - into planned openings scheduled between campaigns. This is central to defending the 0.86-0.995 envelope.

The pipeline

• Acquire: continuous telemetry from magnets, cooling, RF, vacuum, diagnostics
• Feature: extract health indicators (trend, spectral content, residual growth)
• Estimate: remaining-useful-life models per component
• Schedule: place maintenance in the next opening before predicted failure
• Verify: reconcile prediction against what teardown actually finds

def maintenance_decision(rul_hours, next_opening_hours, safety_margin=1.5):
    # act at the opening only if RUL comfortably clears the following opening
    if rul_hours < next_opening_hours * safety_margin:
        return 'service_at_next_opening'
    return 'continue_monitor'

Why it moves the number

Availability A = MTBF / (MTBF + MTTR). Predictive maintenance raises effective MTBF (fewer surprise failures) and lowers effective MTTR (parts and crew staged before the opening). Both push A upward. It cannot close the full 30-100x gap to Tier III alone, but combined with redundancy and fleet learning it is the largest single controllable lever.

The sub-capabilities are remaining-useful-life estimation, anomaly detection, and component-specific health models like first-wall health. Pre-FOAK these models are trained on simulation and component-test data and are labelled as such.