End User
UTBM
Description
The Data-Driven Predictive Modelling service provides an AI-based framework for predicting the performance evolution and degradation of hydrogen fuel cell systems using time-series operational data. Based on LSTM networks with self-attention mechanisms, the service supports condition monitoring, predictive maintenance, and lifecycle optimisation by delivering reliable forecasts of stack voltage as the primary performance health indicator.
Core Capabilities
Monitoring & Anomaly Detection
Predictive & Prescriptive Analytics
Business Need
Hydrogen fuel cell systems experience progressive voltage degradation over their operational lifetime, but predicting when performance will fall below acceptable thresholds requires accurate long-term forecasting from operational data. System operators need condition monitoring and predictive maintenance capabilities that account for varying operating regimes and support evidence-based maintenance planning — particularly for safety-critical applications in transport and distributed energy.
Key Performance Indicators
Mean Squared Error (MSE), RMSE, MAE, and R² across forecast horizons
Accuracy across both minimal-sensing (voltage-only) and multi-sensing configurations
Reproducibility and traceability of predictions against documented model configurations
Data Provided
Time-indexed stack voltage predictions over predefined horizons (short-term and long-term degradation)
Inputs: fuel cell voltage time-series; optionally temperature, pressure, and reactant flow rates
TEF
TEF H2