AI Microservice for Real-Time Transformer Temperature Forecasting

Use Case:

Detection of hidden nuclear material (e.g. uranium) in transport containers; use of simulated and real detector data to predict presence and density of shielding/dense materials

Outcome:

• Successful simulation of 12 million particle events to mimic 10 minutes of detector operation;
• Integration of classification and regression pipelines (SVM, RF, SGD);
• Detection of uranium with 80%+ classification accuracy in high-density center zones;
• Regression models estimate material density across the 3D voxel space;
• Feature importance analysis guides future detector hardware design.

Ecosystem Support:

Technical collaboration with AI experts and domain physicists.

AI Relevance:

• High-impact AI application in critical infrastructure and national security;
• Contributes reusable AI and simulation toolkits for physics-based inspection scenarios;
• Demonstrates the value of interdisciplinary EU collaborations in addressing complex industrial AI challenges.

Summary:

Logicmelt Technologies S.L. and Digafer S.A., two Spanish SMEs, collaborated to develop an AI-powered system capable of detecting nuclear and dense materials—such as uranium and lead—inside transport containers using cosmic radiation detectors.
Traditional inspection techniques suffer from critical blind spots and inefficiencies, particularly in noisy or complex detection environments. By integrating muon tomography, custom simulation tools, and advanced machine learning pipelines, the project created a robust system capable of identifying nuclear material density and type, even in cluttered or shielded environments.

Through a combination of particle physics simulation (via GEANT4), feature engineering, and ensemble AI models (KNN, SVM, Random Forest, SGD), the system successfully achieved voxel-level predictions of nuclear material presence and density. A key innovation was its ability to process 12 million particle events (~10 minutes of real-world measurement) with effective spatial resolution. The project not only produced high-performance AI models but also developed a flexible, open simulation and ML pipeline for future testing and detector design. This showcases how AI can strengthen EU security frameworks, bringing transparency, speed, and intelligence to critical infrastructure inspections.