AI-Enhanced Artificial Olfactory Electronic Nose Real-Time Systems Based on Room Temperature Enabled 2D-Gas Sensor Arrays

Artificial olfactory systems combine state-of-the-art electronic nose and A.I. algorithmic developments to mimic the sensory mechanism of a human nose without olfactory fatigue. By innovatively engineering 2D sensory materials (such as monolayer metal sulfides and metal selenides) onto interdigitated arrays and embedding them into electronic nose systems articulates access to different methods of room temperature detection targeting highly volatile organic compound (VOCs) gaseous mixtures. Algorithmic programmed artificial electronic nose systems can devise temporal and topological techniques to capture an “ aromatic fingerprint ” formed by sensory arrays. In this thesis, two proof-of-concept preliminary works that demonstrate the fabrication and evaluation of room temperature and temperature-voltage dependent hydrogen and hydrocarbon selective based gas sensory arrays based on ultrathin (4 nm) heterojunction structured metal oxides (based on Ta2O5, WO3) are presented. Study of the gas sensor mechanisms in particular, device physics (of 1D-Au contacted ZnO nanorods) and their electronic band structure allows further optimisation of initialisation conditions. Further, experimentation and exploitation of 2D materials (particularly SnS2, SnSe2, WS2 and WSe2) to create of 2D electronic nose systems that extract multiple gaseous data from a diverse number of gas mixtures is the next step from current gas sensing technologies. For the first time, Meta-learning algorithms are introduced to evolve gas sensing with greater capacity in detection, recognition, tracking and profiling to sample and extract VOCs. Research into 2D sensory arrays delves to achieve three forms of gas selectivity (i) mixture detection, (ii) similarity detection and (iii) specific wine detection has been presented. In this thesis, eight novel sensory parameter benchmarks have been devised; covering: (i) sensitivity, (ii) selectivity, (iii) suitability (response time) and (iv) integrability (initialisation conditions), (v) power requirements and (vi) pattern recognition (ability to differentiate) together with prediction (vii) accuracy and (viii) precision. This innovative set of metrics form the basis for evaluating electronic nose performance has been shown. In this thesis, a meta-learning enhanced olfactory electronic noses (MEOEN) system has been developed with advanced meta-learning algorithms. By training the graph convolution neural network with electronic nose and GCMS data captured from wine based VOCs, this research demonstrates the first proof-of-concept artificial electronic nose system to conduct red wine sampling function with recognition and prediction capabilities. Results specifically show an accuracy and precision of 98.74 % and 98.49 %, respectively. Findings in this research demonstrates a major innovation for artificial olfactory systems as it is systemically trained for red wine sampling. It is the first proof-of-concept research that shows gas sensory capability to sample and profile red wine detailing critical advantages over GC-MS analytical methods. Impact of this thesis, marks an important progression in gas sensing and also artificial olfactory systems research to build the next A.I. mechanisms and frameworks for gas recognition and prediction in the domain of the upcoming digital era.