Using Machine Learning to Overcome Interfering Oxygen Effects in a Graphene Volatile Organic Compound Sensor.

Area of Science:

• Materials Science
• Chemical Sensing
• Machine Learning

Background:

• Graphene sensors offer high sensitivity for detecting volatile organic compounds (VOCs).
• Oxygen interference complicates VOC detection, necessitating understanding of combined gas effects.
• Graphene's inherent sensitivity to oxygen presents a challenge for selective VOC sensing.

Purpose of the Study:

• To investigate the cross-selectivity of graphene varactor sensors to oxygen and various VOCs.
• To develop a machine learning approach for classifying gas mixtures and VOC concentrations.
• To assess the feasibility of accurate VOC sensing in the presence of oxygen and sensor drift.

Main Methods:

• Utilized graphene variable capacitor (varactor) sensors.
• Exposed sensors to controlled mixtures of oxygen (3 concentrations) and VOCs (ethanol, methanol, methyl ethyl ketone; 5 concentrations each).
• Employed a long short-term memory (LSTM) network for classification of sensor response data.

Main Results:

• Sensor responses showed distinct shapes based on relative gas concentrations in mixtures.
• The LSTM model achieved 100% accuracy in classifying VOC type, irrespective of oxygen levels.
• Demonstrated VOC concentration resolution within approximately 200 ppm.
• Successfully classified gas mixtures despite typical graphene sensor drift patterns.

Conclusions:

• Graphene varactor sensors, coupled with LSTM networks, can effectively discriminate VOCs in the presence of oxygen.
• The developed machine learning approach overcomes challenges posed by oxygen interference and sensor drift.
• This method enhances the potential for reliable disease diagnosis and environmental monitoring using graphene-based sensors.