Overview
WP7, led by the University of Essex delivers an Explainable Machine Learning algorithms integrated into a Decision Support System. The system uses Deep Convolutional Neural Networks to analyse visual and ultrasonic pipeline inspection data and detect overlapping anomalies such as cracks, rust and mold in steel, concrete and plastic pipes. The DSS is an interactive web platform that maps the water network, flags likely defect locations and provides end-users with estimated repair-cost and environmental-impact analytics. To build trust, it generates pixel-level heatmaps showing that AI predictions are based on real structural damage rather than background noise. For precise localization, it uses a diffusion-based segmentation method that captures spatial relationships and refines masks through progressive denoising, reducing fragmentation common in traditional models. This integrated solution supports continuous infrastructure monitoring with low cost and downtime than manual inspection.
Our result
WP7, led by the University of Essex (UESS), delivers a key exploitable result: Explainable Machine Learning algorithms integrated into a Decision Support System (DSS). The system uses Deep Convolutional Neural Networks to analyse visual and ultrasonic pipeline inspection data and detect overlapping anomalies such as cracks, rust and mold in steel, concrete and plastic pipes. The DSS is an interactive web platform that maps the water network, flags likely defect locations and provides end-users with estimated repair-cost and environmental-impact analytics. To build trust, it generates pixel-level heatmaps showing that AI predictions are based on real structural damage rather than background noise. For precise localization, it uses a diffusion-based segmentation method that captures spatial relationships and refines masks through progressive denoising, reducing fragmentation common in traditional models. This integrated solution supports continuous, accurate infrastructure monitoring with lower cost and downtime than manual inspection.
Needs and target audiences
We are looking for pilot customers, strategic partners and investors to support validation, market uptake and commercial exploitation of our Explainable AI and DSS solution for infrastructure inspection.
Current stage
Prototype developed and validated within TUBERS. Next steps include further testing with real-world data, pilot validation with end-users, tighter workflow integration, performance optimisation and preparation for commercial exploitation.
Unique value proposition
A trustworthy AI-based inspection and decision-support solution that fuses visual and ultrasonic data to accurately detect pipeline defects, explain its predictions, and support faster, lower-cost and lower-downtime maintenance decisions.
Keywords
- inspection & maintenance
- pipelines
- explainable machine learning
- ultrasonic testing
- pipeline inspection
- drinking water
- snake robot
Research project acronyms
- TUBERS
Research projects
- Scalable and Modular robotic tools for pipeline inspection and repair
Project identifiers
- 101070115
Contributors
- TECH HIVE LABS ASTIKI MI KERDOSKOPIKI ETAIREIA
Source: Horizon Results Platform