End User
UTBM
Description
The service is intended to be evaluated using historical datasets containing both normal operation and leak scenarios. The evaluation framework would separate training and testing data to ensure a robust assessment of detection and localization performance. Performance is expected to be measured using standard classification metrics, including accuracy, precision, recall, F1-score, and false alarm rate for leak detection, along with localization accuracy for identifying the leak position. Visual analysis of sensor signals and detected events would further support validation of the model’s ability to capture abnormal behaviour.
Core Capabilities
Monitoring & Anomaly Detection
Business Need
Service 5 aims to provide an AI-based framework for the detection and localization of hydrogen leaks in hydrogen systems using multi-sensor data. It is intended to support system safety, reliability, and risk mitigation by enabling early identification of abnormal conditions and precise localization of leak events. The service is expected to rely on time-series data collected from hydrogen infrastructures, including signals from pressure sensors, flow meters, acoustic or vibroacoustic sensors, temperature sensors, and other monitoring devices. By learning normal operational behaviour and identifying deviations, it would detect leak occurrences and estimate their location within the system.
It is envisioned to operate under varying conditions and configurations, ensuring robustness across different system scales and environments. The service is also expected to ensure traceability and reproducibility by associating detection results with metadata describing data sources, sensor configurations, and model parameters, providing a reliable foundation for safety monitoring in hydrogen systems.
Operational context from source document:
The service is intended to be developed and evaluated using datasets collected from hydrogen systems, including experimental test benches and operational infrastructures. These datasets are expected to capture both normal and abnormal operating conditions, including leak scenarios under controlled or real-world environments. They would consist of time-series measurements from multiple sensors, such as pressure, flow rate, acoustic or vibroacoustic signals, and temperature, reflecting the physical behaviour of hydrogen within the system.
The service is envisioned to operate within an offline evaluation framework, where historical data are used to simulate detection scenarios and assess performance under realistic conditions. This setup would support validation for applications in safety monitoring, fault detection, and risk management in hydrogen infrastructures.
It is envisioned to operate under varying conditions and configurations, ensuring robustness across different system scales and environments. The service is also expected to ensure traceability and reproducibility by associating detection results with metadata describing data sources, sensor configurations, and model parameters, providing a reliable foundation for safety monitoring in hydrogen systems.
Operational context from source document:
The service is intended to be developed and evaluated using datasets collected from hydrogen systems, including experimental test benches and operational infrastructures. These datasets are expected to capture both normal and abnormal operating conditions, including leak scenarios under controlled or real-world environments. They would consist of time-series measurements from multiple sensors, such as pressure, flow rate, acoustic or vibroacoustic signals, and temperature, reflecting the physical behaviour of hydrogen within the system.
The service is envisioned to operate within an offline evaluation framework, where historical data are used to simulate detection scenarios and assess performance under realistic conditions. This setup would support validation for applications in safety monitoring, fault detection, and risk management in hydrogen infrastructures.
Key Performance Indicators
accuracy
precision
recall
F1
Data Provided
The service is intended to be developed and evaluated using datasets collected from hydrogen systems, including experimental test benches and operational infrastructures.
They would consist of time-series measurements from multiple sensors, such as pressure, flow rate, acoustic or vibroacoustic signals, and temperature, reflecting the physical behaviour of hydrogen within the system.
The service is envisioned to operate within an offline evaluation framework, where historical data are used to simulate detection scenarios and assess performance under realistic conditions.
This setup would support validation for applications in safety monitoring, fault detection, and risk management in hydrogen infrastructures.
The service is intended to be based on a data-driven framework for anomaly detection and localization using multi-sensor data.
The methodology is envisioned as a structured pipeline including data preprocessing, feature extraction, and model development.
Preprocessing would ensure data quality through filtering, synchronisation of sensor signals, and handling of noise or missing values.
Feature extraction is expected to capture relevant patterns from the data, including statistical descriptors, temporal dynamics, and signal transformations.
TEF
TEF H2