Related Experiment Video
Updated: Mar 6, 2026
10:42
Infrared Degenerate Four-wave Mixing with Upconversion Detection for Quantitative Gas Sensing
Published on: March 22, 2019
6.6K
Selective Sensing of Hydrogen and Ammonia Using a Single CMOS-Compatible Sensor and Transfer Learning Methods
Anwesha Mukherjee1, Mohd Salman Siddiqui1, Idan ShemTov1
1Department of Physical Electronics, School of Electrical Engineering, Tel Aviv University, Ramat Aviv 69978, Israel.
ACS Sensors
|March 4, 2026
Summary
This study introduces a novel palladium nanoparticle-decorated sensor for selectively detecting hydrogen (H2) and ammonia (NH3) mixtures. Advanced machine learning achieved high accuracy in distinguishing gases and their blends for enhanced safety.
Area of Science:
- Materials Science
- Nanotechnology
- Chemical Engineering
Background:
- High-performance gas sensors are vital for the hydrogen economy, especially for detecting flammable hydrogen (H2) and toxic ammonia (NH3).
- Existing sensors struggle with cross-sensitivity between H2 and NH3, hindering accurate detection of individual gases and their fuel mixtures.
- Accurate monitoring is critical for safety, preventing accidents, and optimizing the use of H2/NH3 blends as zero-carbon fuels.
Purpose of the Study:
- To develop a selective sensor for simultaneous detection of H2, NH3, and their mixtures.
- To leverage machine learning for enhanced gas differentiation and classification accuracy.
- To demonstrate a cost-effective, miniaturized sensing solution without requiring sensor arrays.
Main Methods:
- Fabrication of a palladium (Pd) nanoparticle-decorated electrostatically formed nanowire (Pd-EFN) sensor using CMOS-compatible processes.
- Utilizing multigate electrostatic control for enhanced sensitivity and selectivity.
- Applying supervised machine learning (SVM, AdaBoost, Random Forest, etc.) and transfer learning (VGG-19) on sensor response data.
Main Results:
- The Pd-EFN sensor exhibited a highly reversible response with distinct "electrostatic fingerprints" for gas differentiation.
- Machine learning achieved up to 94% accuracy in distinguishing H2 vs. NH3 and H2 vs. (NH3 + H2).
- Transfer learning further improved classification accuracy to ~97% for individual gases and ~87% for mixtures, enhancing discernment of H2/NH3/(NH3 + H2).
Conclusions:
- A single Pd-EFN sensor, combined with machine learning, can selectively identify H2, NH3, and their mixtures with high accuracy.
- This approach overcomes cross-sensitivity issues and eliminates the need for complex sensor arrays.
- The developed technology offers a promising pathway for advanced, miniaturized, and cost-effective gas sensing platforms for hydrogen safety and environmental monitoring.
