In-Situ Concentration Measurement of Blended Hydrogen Gas Using Sensor Fusion Enhanced by Machine Learning Model

Abstract Blending hydrogen (H2) into natural gas via existing pipelines presents a practical method for H2 transportation. However, accurately measuring the H2 concentration in these blends is crucial for safety, operational efficiency, cost-effectiveness, and energy content tracking. This study proposes an effective sensor fusion system for in-site measurement of blended H2 concentration by combining ultrasonic and thermal conductivity (TC) sensors. To test this system, these sensors were installed in a pipeline along with pressure, temperature, and humidity sensors to compensate for environmental factors, while mass flow controllers regulated gas composition. Calibration across the full range of H2 concentrations confirmed the suitability of ultrasonic and TC sensors for H2 measurement. A dataset comprising 185 sensing data points under various environmental conditions was collected to train a machine-learning model for in-site H2 measurement. The evaluation of the model demonstrated higher accuracy in H2 measurement through sensor fusion compared to individual models, with the optimized model exhibiting excellent performance, achieving a coefficient of determination (r2) of 0.98 between actual and measured values. While this study was conducted under limited environmental conditions, it is anticipated that accurate H2 concentration measurement will be feasible in more diverse environments with additional sensor data collected in the future.