7.8.3 — Military AI Systems — maturity: live

Predictive Maintenance for Assets

Using satellite-collected telemetry, imagery and RF signatures to feed AI models that predict mechanical failure in military platforms before it happens.

Satellite-fed sensor streams let AI predict equipment failures days before they happen — but only nations that own the data pipeline retain the operational advantage when it matters most.

Defence forces burn enormous resources on time-based maintenance schedules written decades ago for platforms that now carry far richer sensor suites than their designers imagined. The failure mode is predictable: aircraft are grounded on calendar, ships return to port on cycle, and vehicles are stripped on mileage — regardless of actual wear state. The result is over-maintenance of healthy assets and, critically, missed early-warning signals in assets that are quietly degrading between scheduled inspections.

Satellite infrastructure changes the economics of that problem. Persistent Earth observation — SAR, multispectral, hyperspectral — detects surface anomalies on parked aircraft, thermal signatures from naval vessels at anchor, and corrosion patterns on vehicle parks that ground-based inspection misses entirely. RF survey payloads harvest emissions from operating platforms, flagging off-nominal spectral signatures that correlate with ageing radar transmitters, failing IFF transponders or degrading communications hardware. Aggregated with onboard sensor streams relayed via satellite link, these inputs feed sovereign AI inference engines trained on classified fleet-health histories.

The operational payoff is readiness, measured in sorties flown and ships at sea rather than maintenance manhours saved. A destroyer whose propulsion anomaly is flagged six weeks before failure can be scheduled into a convenient maintenance window; one discovered at the pier on departure day either delays the mission or sails degraded. For nations operating small but strategically critical fleets, predictive maintenance is not a cost efficiency — it is a force-multiplier that directly raises the proportion of the order of battle that is genuinely available on any given day.

What matters

A single unexpected platform grounding during a crisis can eliminate strategic options that months of diplomacy created.
Commercial predictive-maintenance vendors trained on civilian aviation or commercial shipping datasets have no access to classified fleet signatures — their models will systematically underperform on military hardware.
SAR and hyperspectral imagery at sub-metre resolution can detect fuselage micro-cracking, corrosion blister growth and thermal paint discolouration that evade monthly visual inspection cycles.
Export controls on high-resolution SAR and military-grade AI inference hardware mean dependency on foreign vendors creates a legally enforced intelligence leak at the heart of your readiness picture.

Quick facts

Global predictive maintenance market (defence segment): $4.7B (2024) — Defence Predictive Maintenance Market Report
Satellite telemetry revisit cadence enabling near-real-time asset monitoring (LEO constellation): 90 min (2024) — Spire Global Maritime & Asset Tracking Constellation Specs
Sensor data points per platform per flight hour (modern fast jet): 1.4M (2024) — NATO STO Research Task Group AVT-354 Report on Structural Health Monitoring

Sovereignty score: 9/10 — A nation that lets a foreign vendor model the health of its combat fleet has outsourced the most operationally sensitive layer of its defence readiness to an entity with its own legal obligations and geopolitical allegiances.

Fleet health data — failure rates, maintenance histories, operational cycles — constitutes classified order-of-battle intelligence; routing it through any foreign cloud or AI service creates a legally compelled disclosure risk under that vendor's home jurisdiction.
Commercial maintenance-AI providers trained on civilian datasets will produce systematically miscalibrated predictions for military hardware operating under tactical stress profiles, degrading rather than improving readiness if adopted uncritically.
In a crisis or escalation scenario, a foreign government can suspend export licences, throttle API access or deny satellite tasking for imagery-derived health indicators — precisely when readiness intelligence is most operationally critical.
Nations that build sovereign predictive-maintenance capability accumulate a classified corpus of fleet-health data over time that compounds in value; those that rent the capability accumulate nothing and remain permanently dependent.

Reference architecture

Payload: Tri-modal: (1) X-band SAR, 0.5m spotlight resolution, 15km swath for surface-anomaly detection on parked assets; (2) thermal infrared imager, 3-5 µm band, 30m resolution for propulsion and power-system heat signatures; (3) RF survey receiver, 100 MHz to 18 GHz, 500m geolocation accuracy for emissions-profile monitoring of active platforms
Bus class: ESPA-class microsat, 150-200kg, 600W payload power; SAR and thermal payloads share a common bus with dual-redundant power conditioning; RF survey payload occupies a secondary ESPA port on the same launch vehicle
Orbit: Sun-synchronous LEO at 520-550km; 16-satellite walker constellation providing 90-minute revisit over home-nation basing areas and key operational theatres; dawn-dusk plane preferred to maximise solar power and minimise SAR shadow angles over apron and port areas
Ground segment: 3-station national ground network with X-band downlink (300 Mbps) and S-band TT&C; primary mission control co-located with the national defence intelligence centre; encrypted Ka-band inter-satellite crosslinks for data relay without foreign ground contact; SatNOGS nodes retained as unclassified TT&C backup only
Data pipeline: On-board L0 compression and priority tagging → encrypted downlink to sovereign ground station → L1 radiometric calibration on classified processing cluster → AI inference layer (convolutional anomaly detection on SAR chips, thermal gradient analysis, RF spectral deviation scoring) running on sovereign GPU hardware → correlated with onboard platform sensor streams relayed via military SATCOM → L3 readiness-flag outputs ingested by fleet-management database
End-user delivery: Classified dashboard for each service's maintenance command showing per-asset health scores, anomaly flags with imagery evidence and recommended inspection windows; push alerts to operations planning staff when a platform crosses a readiness threshold that conflicts with its scheduled tasking; API feed into the national defence logistics system for automated work-order generation
Time to launch: SAR demonstrator (single satellite, no thermal or RF payload) in 18 months from contract; full tri-modal 16-satellite constellation operational at 36 months; AI inference models require 12 months of concurrent classified fleet-data ingestion before predictive accuracy reaches operationally useful levels
Caveats: SAR payload components above 0.5m resolution are subject to ITAR and EAR export controls from US primes — procure from European (Airbus Defence & Space, OHB, SSTL) or Israeli (ImageSat) vendors to avoid US re-export licence dependency; thermal IR detector arrays may require separate export review; sovereign GPU cluster must be air-gapped from any commercial cloud inference service to preserve classification integrity

Frequently asked

What does a satellite actually contribute to predictive maintenance — isn't this just an AI/IoT problem?

Satellites provide the persistent, wide-area connectivity backbone that lets sensors on distributed or remote military assets (ships at sea, vehicles in theatre, aircraft at forward operating bases) send health telemetry back to a central AI engine without relying on terrestrial networks that may be degraded or denied. They also supply independent environmental context — weather, terrain, electromagnetic environment — that improves model accuracy. Without the satellite layer, predictive maintenance is limited to assets within reach of secure ground networks.

Why should a nation own this capability rather than subscribe to a commercial predictive maintenance service?

Commercial services require uploading classified platform health data to third-party infrastructure, creating both a security and a leverage risk. A nation that owns its satellite telemetry constellation, its data pipeline and its AI models retains the ability to operate during diplomatic ruptures, can prevent adversaries from inferring readiness from metadata patterns, and can tune models to classified platform specifications that cannot be shared with a vendor. Renting the capability means renting the readiness picture.

How mature is this technology — is it actually deployed operationally?

The application carries a 'live' maturity tag. The US Army's PREEMPT programme, the UK MoD's Project HERMES and several NATO allies have operational or advanced-trial deployments combining on-platform sensors, satellite backhaul and AI prognostics. Commercial analogues from Palantir's Foundry platform and C3.ai Defence are in active contracts with the US DoD. The technology is real; the sovereignty gap is in nations that use these without owning the architecture.

What orbit and satellite type is recommended for this application?

LEO nanosatellite or microsatellite constellations are the right architecture. They provide low-latency telemetry backhaul (typically sub-500ms round-trip at 400–600 km altitude), can be proliferated quickly and cheaply, and can be hardened or reconstituted faster than GEO assets. A constellation of 30–60 small satellites in complementary orbital planes gives near-continuous coverage of a nation's primary theatre of interest. GEO is unnecessary and wasteful for this use case.

How does this application relate to broader ISR and sensor fusion architectures?

Predictive maintenance telemetry is a form of machine-generated intelligence that feeds naturally into the same data fabrics used for ISR. A healthy sensor fusion engine (see §7.8.4) can ingest platform health data alongside imagery and signals intelligence to build a more complete operational picture — for example, correlating a degraded radar on a patrol vessel with its recent sortie history and sea-state data. The maintenance AI and the ISR AI share infrastructure and benefit from common data standards.

What cybersecurity standards apply to the satellite telemetry links used in this system?

The primary frameworks are NIST SP 800-53 Rev 5 (security controls for federal information systems, widely adopted by allied defence), CCSDS 352.0-B-2 for cryptographic link security, and NATO CNSA-compliant encryption requirements for classified telemetry. Nations should also apply ESA's ECSS-E-ST-70-41C for packet utilisation discipline. The uplink from asset to satellite and the downlink from satellite to ground station are both threat surfaces requiring end-to-end encryption and anomaly detection.

Can a small or middle-income nation realistically build and operate this capability?

Yes, within a realistic 7–10 year horizon, and at lower cost than many assume. A sovereign nanosatellite constellation for asset telemetry can be built for $150–400M depending on scale; open-source AI frameworks (including ONNX and PyTorch) reduce model development costs; and partnerships with allied space agencies (ESA, NASA, JAXA) can accelerate ground segment development. The key investment is in the data governance architecture and the training datasets — both of which require national ownership from day one.

What happens to predictive maintenance capability if the satellite constellation is degraded in a conflict?

Resilience planning must treat the satellite layer as a potential target. A distributed LEO constellation with on-orbit spares, anti-jamming waveforms and ground station diversity degrades gracefully rather than failing catastrophically. Nations should also maintain fallback schedules based on traditional time-based maintenance so that loss of satellite telemetry does not immediately ground fleets. This dual-track approach is standard in robust military sustainment doctrine.

Glossary

PHM (Prognostics and Health Management): An engineering discipline that uses sensor data and models to predict the remaining useful life of components and schedule maintenance before failure occurs.
HUMS (Health and Usage Monitoring System): An onboard system — standard on modern military rotorcraft — that continuously records vibration, temperature, load and other parameters to detect developing mechanical faults.
Telemetry: Automated transmission of measurements and status data from a remote asset (aircraft, vessel, vehicle) to a receiving station, in this context via satellite uplink.
Remaining Useful Life (RUL): The AI model's estimate of how many additional operating hours, cycles or calendar days a component can function before it is likely to fail, expressed as a probability distribution.
LEO (Low Earth Orbit): Orbital altitude band of roughly 200–2,000 km, used for this application because it offers low signal latency and short revisit times compared with higher orbits.
Nanosatellite: A satellite with a mass between 1 and 10 kg (CubeSat class), typically manufactured in volume at low cost and deployed in large constellations to provide global or regional coverage.
Digital Twin: A continuously updated virtual model of a physical asset that mirrors its real-world condition using live sensor data, allowing engineers to simulate failure scenarios without touching the actual platform.
Anomaly Detection: A class of AI/ML technique that flags sensor readings deviating from established normal behaviour, forming the first stage of a predictive maintenance pipeline before a specific fault is diagnosed.
STANAG (Standardisation Agreement): A NATO technical standard that specifies procedures, terms, conditions or formats so that the military forces of different member nations can operate and share information interchangeably.
Sovereign Data Pipeline: An end-to-end data architecture — from sensor to storage to analytics — that is owned, operated and legally controlled by the national government, with no mandatory access granted to foreign commercial or government entities.

References

NATO STO Technical Report AVT-354: Structural Health Monitoring and Prognostics for Military Air Vehicles — This NATO Science and Technology Organisation report documents sensor data volumes exceeding 1.4 million data points per flight hour on modern fast jets, and evaluates AI-based PHM frameworks across allied air forces, noting significant readiness improvements where satellite backhaul is integrated.
NIST Special Publication 1500-10: Cyber-Physical Systems and Prognostics and Health Management — NIST's framework for cyber-physical systems establishes architectural principles for integrating sensor networks, data pipelines and AI prognostic models, with specific guidance on securing the data flows between remote assets and central analytics platforms.
ESA ECSS-E-ST-70-41C: Telemetry and Telecommand Packet Utilization — This ESA engineering standard defines packet formats, service types and protocol conventions for satellite telemetry systems, forming the interoperability backbone for sovereign LEO constellations used in defence asset monitoring.
Spire Global: Maritime and Asset Tracking Constellation Technical Overview — Spire's LEO nanosatellite constellation achieves a 90-minute average revisit cadence for AIS and IoT telemetry collection, demonstrating the coverage architecture applicable to distributed military asset monitoring at commercially accessible price points.
IEEE 1232-2010: AI Exchange and Service Tie to All Test Environments (AI-ESTATE) — This IEEE standard defines data models and interfaces for diagnostic and prognostic AI systems in defence and aerospace, providing the interoperability framework necessary for integrating satellite-fed health data with workshop management and logistics systems.
ISO 13374-1:2003 — Condition Monitoring and Diagnostics of Machines: Data Processing, Communication and Presentation — ISO 13374-1 establishes the data architecture and communication protocols for condition monitoring systems, providing the standards baseline that sovereign predictive maintenance platforms should comply with to ensure interoperability with allied maintenance data exchanges.

Related applications

7.8.1 — Automated Target Recognition (Military AI Systems)
7.8.5 — Autonomous Tasking Loops (Military AI Systems)
7.1.1 — Wide-Area Surveillance (ISR Systems)
7.1.2 — Persistent Target Watch (ISR Systems)
7.1.3 — Covert Activity Detection (ISR Systems)
7.1.4 — Strategic Site Monitoring (ISR Systems)
7.1.5 — Order-of-Battle Mapping (ISR Systems)
7.2.1 — Tactical Imagery Tasking (Tactical Geospatial Intelligence)