Gas sensors are extensively utilized in monitoring air quality, ensuring public safety, and detecting released trace gases in countless industrial fields. Accordingly, the development of highly efficient, sensitive, selective, reliable, low power consump- tion and low-cost gas sensors is in considerable demand. A myriad of gas sensors using traditional metal oxide semiconductor materials have been developed, nev- ertheless, the selectivity and power-consumption of these sensors are still far from satisfactory. Inspired by human olfaction, advanced nanomaterials as well as artifi- cial intelligence technology may provide the solution to these issues. In this work, the pristine graphene-based highly sensitive gas sensors working at room temper- ature for NH3 detection were developed. In combination with machine learning techniques, the selectivity of pristine graphene-based gas sensors is significantly enhanced, which present excellent performance towards odor discrimination.

As a first step, the stabilization mechanism of functionalized graphene in an aqueous dispersion of surfactant is elucidated via all-atom molecular dynamic sim- ulations. The stabilizing role of flavin mononucleotide sodium salt (FMNS) is demon- strated by the potential of mean force calculations for pairs of graphene flakes cov- ered by FMNS molecules. At a high surface coverage, graphene flakes repel each other which leads to the stabilization of graphene dispersions. To achieve approxi- mately the same potential of mean force (PMF) energy barrier of 10 kJ/(mol · nm2), the surface coverage of graphene flakes by FMNS molecules is 44% lower than by sodium cholate (SC) molecules, and 71% lower than by sodium dodecylbenzene- sulfonate (SDBS) molecules, respectively. With this in mind, FMNS functionalized graphene-based gas sensors are then developed, demonstrating excellent sensing performance to NH3 gas. The optimized NH3 sensors demonstrate outstanding per- formance: ultralow limit-of-detection (1.6 ppm), excellent sensitivity (2.8%, 10 ppm; 18.5%, 1000 ppm), reproducibility, reversibility, low power consumption, room tem- perature function, as well as low cost. The roles of FMNS from graphene prepa- ration to NH3 sensing are elucidated via all-atom molecular dynamics simulations
(MDS): (1) stabilizer for the graphene dispersion, (2) p-type dopant for graphene- based sensing element, and (3) active adsorption sites for NH3 gas sensing.

Moreover, in combination with machine learning techniques, biomimetic elec- tronic olfaction based on graphene single channel nanosensors is proposed. The de- veloped prototype exhibits excellent odors (Eucalyptol – Euca, 2-nonanone – 2Nona, Eugenol – Euge, 2-phenylethanol – 2Phe, N2) discrimination and identification perfor- mance at room temperature, maximizing the obtained results from a single nanosen- sor. Upon exposure to binary odor mixture, the response features behave similarly to existing individual odor component, mimicking the overshadowing effect in hu- man olfactory perception. Computational simulations support the experimental re- sults and reveal competing adsorption of odor molecules occur. With this approach, the industrial pollutants (NH3, PH3) were succefully identified at ultra-low concentra- tion (100 ppb – 1000 ppb) with satisfying performance.

The present work represents a novel and reliable strategy to develop highly sensi- tivity, highly selective, and low-cost graphene-based gas sensors towards inorganic gases detection (NH3, PH3) and volatile organic compounds (VOCs) sensing at room temperature. The developed strategy may allow for gas detection, odor recognition of a wide spectrum of odor molecules, as well as detection of volatile organic com- pounds (VOC) in an extensive variety of domains, e.g., environmental monitoring, public security, smart farming, or disease diagnosis (e.g., lung cancer, COVID-19).

Keywords: graphene, liquid phase exfoliation, gas sensor, electronic nose, sensitivity, selectivity, machine learning, molecular dynamic simulation, NH3/PH3, volatile organic compounds