AI-RWay

The Project: Innovation Serving Predictive Maintenance

AI-RWay is an end-to-end platform for intelligent railway infrastructure inspection, developed to support the evolution from a Proof of Concept to a fully integrated solution validated in operational environments.

The platform integrates:

• Drone-based video acquisition with geo-referenced metadata

• An automated processing pipeline based on Computer Vision and Machine Learning models

• Structured event and alert generation

• A web interface for operational management and intervention planning

The platform enables automated or semi-automated verification of railway line conditions, supporting the analysis of infrastructure status and the management of timely notifications.

By leveraging Artificial Intelligence and drone-based data acquisition, AI-RWay enables intelligent maintenance, automating inspection and monitoring processes across railway components and enhancing the efficiency, traceability, and responsiveness of maintenance operations.

Covered Use Cases

1️⃣ Vertical Signage Monitoring

The system automatically detects the presence and condition of railway signs, identifying:

• Missing signs

• Degraded or vandalized signs

• Deviations from the reference condition

2️⃣ Object and Obstacle Detection on Track

The Object Detection module identifies anomalous objects along the railway track (e.g., debris, obstacles) through:

• Frame-by-frame analysis with controlled downsampling

• Temporal aggregation of detected events

• Automatic geolocation of alerts

3️⃣ Track Circuit Monitoring (Track Circuit System)

The system identifies and verifies the condition of installed elements along the railway line, with a specific focus on the track circuit system:

• Cable recognition

• Detection of missing or damaged components

• Operational status classification

The dedicated model was trained on proprietary datasets collected under real operational conditions.

Platform Functionalities

ZIRAK, as a project partner, played an active role in the following tasks:

• Design of the overall software architecture, including backend, frontend, and data acquisition and communication methods

• Pre-processing and annotation of drone-acquired data

• Selection of Artificial Intelligence and Computer Vision models

• Development of the entire software platform

• Field management of the demonstrator activities

AI-RWay implements a comprehensive video and alert lifecycle management system, enabling users to:

• Automatically upload drone-acquired videos

• Analyze preprocessed content using dedicated AI algorithms tailored to each use case

• Automatically generate events

• Classify detected issues and assign severity levels

• Schedule maintenance interventions and attach relevant metadata (e.g., severity, notes)

• Visualize geo-referenced events on an interactive map through a dynamic Digital Twin of the infrastructure, continuously updated by AI model outputs

• Efficiently manage historical data and event records

To ensure thorough validation of the platform, the project included the creation of dedicated datasets for the different application use cases. Validation activities were supported by real-world demonstrations conducted on selected sections of the reference railway network, under varying lighting and weather conditions to ensure robust coverage.

These activities also enabled accurate drone georeferencing, further enhancing the reliability and operational applicability of the solution.

Map view

Events management

Rail lines management

Results Achieved

The AI-RWay project concluded with the development and validation of an integrated platform for intelligent railway infrastructure inspection, demonstrating its operational feasibility in real-world scenarios.

Throughout the project, the following components were developed and integrated:

• AI modules for the detection and classification of vertical railway signage

• Object Detection modules for identifying obstacles and anomalous objects along the track

• A monitoring system for the track circuit systems

• A backend infrastructure for managing geo-referenced videos and metadata

• A web application for map-based visualization, alert management, and maintenance planning

The platform was validated through dedicated flight campaigns and end-to-end testing, covering the entire workflow from video upload to automatic event generation.

During the validation phase, the following performance levels were achieved:

• 94% accuracy in object and obstacle detection

• 90% accuracy in sign classification and condition detection

• Up to 99% accuracy in track circuit status recognition

The demonstrated solution enables the transformation of railway inspection from a predominantly manual activity into a digital, repeatable, and traceable process, supporting a structured approach to assisted and predictive maintenance.

Our partners