STRUCTURE | Predictive maintenance for transport infrastructure via Multi-Modal Sensing AI

Region:

Flanders

Financed by

AI-driven infrastructure inspection using multi-modal sensing

Transport infrastructure such as bridges, roads and tunnels is ageing, while inspection resources remain limited. STRUCTURE develops AI-driven inspection and predictive maintenance approaches combining multi-modal sensing technologies, such as RGB and infra-red images and Lidar point cloud data, to support safer, more efficient and data-driven infrastructure management.

Target group

The project is relevant for:

Infrastructure owners and operators
Mobility operators
Inspection and maintenance service providers
Public authorities and asset managers

Context

Current infrastructure inspection practices remain largely manual, periodic and fragmented. Inspections often require costly interventions and temporary closures, while results depend heavily on expert interpretation and are not always consistent over time.

At the same time, mobile sensing platforms and advanced sensors generate increasing amounts of data, including visual, thermal, LiDAR and structural information. However, these heterogeneous data sources are rarely integrated effectively, limiting the ability to detect early defects or monitor degradation trends over time.

Existing AI solutions also face important limitations. Many are developed for isolated use cases, lack robustness in operational environments and provide limited explainability, making large-scale deployment difficult.

Objectives & results

STRUCTURE addresses these challenges by developing integrated and interpretable AI methods that transform multi-modal sensing data into actionable maintenance insights.

The project will:

Research methods to combine data from heterogeneous mobile and stationary sensors
Build AI models for defect detection, classification and degradation prediction
Demonstrate scalable inspection workflows in operational environments
Evaluate reliability, robustness and operational impact
Translate research outcomes into practical tools for infrastructure managers

Key results:

Demonstrators combining AI with multi-modal sensing for infrastructure inspection
Curated datasets integrating visual, thermal, LiDAR and structural data
AI models for defect detection and predictive maintenance
Best-practice guidelines for AI deployment in infrastructure monitoring
Workshops and validation activities with industry stakeholders

Approach

The project follows an end-to-end, data-driven approach combining sensing technologies, AI and digital twins across the full inspection and maintenance workflow.

1. Smart data acquisition and edge processing

Deployment of UAV-based inspection systems with real-time data acquisition
Vehicle-mounted sensing using LiDAR and camera systems
RGB and thermal traffic cameras
Development of light-weight models suitable for edge AI analytics

2. Defect detection and characterisation

Development of AI algorithms for crack and defect detection across sensing modalities
Anomaly detection and object tracking
Creation of structured datasets for training and benchmarking

3. Multi-modal data fusion

Fusion of heterogeneous sensing data into unified infrastructure representations
Development of AI models combining geometric, visual and temporal information
Improvement of detection reliability under real-world conditions

4. Digital twins and predictive analytics

Reconstruction and visualisation of infrastructure assets in digital twin environments
Modelling of degradation and defect evolution over time
Integration of traffic and usage data for predictive maintenance

5. Decision support and operational integration