Case study
Prorail × Avy
total flight distance across both sites
images analysed by AI
average AI detection accuracy
ProRail manages one of the Netherlands' most critical infrastructure networks thousands of kilometres of track and the safety-critical systems that keep them running. Keeping that network compliant, secure, and operational depends on continuous inspection at a scale that manual processes struggle to match.
The problem
ProRail's inspection model is manual, resource-intensive, and increasingly under pressure. Inspectors drive to sites or walk the tracks. Transient incidents can pass undetected. Response to large-scale events like storms is slow and expensive. And safety requirements are tightening while available workforce capacity stays flat.
The specific challenges were:
Scale: The network is too large to inspect consistently with ground personnel
Response time: Manual confirmation of an incident delays response before it begins
Safety: Inspectors on or near live tracks face genuine personal risk on every visit
Prioritisation: Without data, subcontractors are deployed based on assumptions, not evidence
The solution
ProRail and Avy designed a structured BVLOS drone and AI pilot across two operational sites; Maasvlakte in Rotterdam and the Apeldoorn–Stroe rail corridor, to test whether autonomous drone-based surveillance could replace or significantly enhance manual inspection.
Aircraft and systems
The Avy Aera fixed-wing BVLOS drone covered long-range corridor operations, while a Drone-in-a-Box quadcopter system handled detailed inspection within contained yard environments. AI analysis was delivered by Arkensight. Ten flight days were executed across both locations under a range of conditions: daylight, night, sun, and snow.
Use cases validated
LOD compliance monitoring — gates, fire hydrants, emergency roads, vehicles
Security surveillance and copper theft detection
Vegetation and track clearance checks (post-storm)
Level crossing sightline validation
Personnel detection in safety-critical zones
Night flight operations
Two AI modes in one pipeline
A key outcome of the pilot was the validation of two complementary AI deployment modes within a single end-to-end data pipeline: post-flight processing for structured routine inspection reports, and live-stream AI analysis for real-time alerts on urgent events. Both were successfully demonstrated.
The results
1,004 km total flight distance across both sites
13.65 hours total flight time
10,500 images and frames analysed by AI
11 object types detected — including vehicles, persons, gates, fire hydrants, and fence damage
94% average detection accuracy — 96% for persons, 95% for vehicles
8 distinct use cases proven operationally
Real-time live-stream AI detection validated alongside post-flight processing
Night operations confirmed under realistic conditions
What this means for energy infrastructure operators
The ProRail pilot demonstrates that drone-based surveillance combined with AI analysis is not a future capability. It is operational today, across real Dutch railway infrastructure, in real weather conditions, validated against the compliance and safety standards that rail operators actually work to.
For operators managing large, safety-critical networks, the combination of long-range fixed-wing drones for corridor coverage and Drone-in-a-Box systems for contained yard operations offers something manual inspection cannot: continuous, data-driven monitoring that puts your team in front of a screen rather than on a live track, and deploys subcontractors based on validated evidence, not assumptions.