Descripció del projecte
The development of unattended gas sensing networks is highly demanded as these could help meeting the requirements of a ubiquitous and automated detection of volatile chemicals. Such detection scenarios range from the release of highly (neuro)-toxic, volatile compounds in closed spaces such as shopping areas, cinemas, theatres or transportation infrastructures under a terrorist attack; the continuous monitoring of ambient pollutants such as NOx, SO2, CO2, CO, NH3 in large areas with high granularity (i.e., with a high density of detection nodes) to the monitoring of volatile organic compounds released during day-to-day activities at home for supporting the independent living of the elderly or long-term patients.
Such gas sensing networks demand inexpensive gas sensors with improved long-term stability and high selectivity to target gases. In addition, gas sensors should be low-power, as many sensing nodes may not have access to power supply and, thus, should be battery operated.
Currently existing inexpensive sensors (metal oxide chemiresistors or low cost electrochems) do not meet the high selectivity, long-term stability and low-power specifications. In addition, one of the main problems when facing a continuous monitoring approach is the occurrence of missing data due to sensor malfunction, or to communications failure, distance between sensor nodes and gateway, insufficient power in the battery, sensor synchronization, weak Wi-Fi signal strength, corrupted readings, etc.
Considering all of the above, this thesis aims to the development of new chemoresistive sensors with enhanced stability, selectivity and ultralow power consumption. To this end, new gas sensitive materials will be synthesized and fully characterised (morphology, composition and gas sensing properties). The aerosol-assisted chemical vapour deposition and the atmospheric pressure chemical vapour deposition methods will be considered for achieving three-dimensional assemblies (sponge-like) of two-dimensional transition metal dichalcogenides (2D TMDs) (WS2, WSe2, WTe2, MoS2, MoSe2, etc.) doped with clusters of noble metals (Au, Pt, Pd, ….). Preliminary results achieved by URV on 2D TMDs support the stable, highly sensitive and low-power operation of these nanomaterials.
The thesis will identify the most promising gas sensors and will integrate these in newly designed and implemented sensing nodes for developing and deploying a wireless gas sensing network able to implement sensor driving, readout, and data communication to the cloud. Furthermore, Artificial Intelligence algorithms, such as recurrent neural networks and deep learning approaches (for e.g., convolutional neural network), Edge- and Cloud-based computational paradigms, as well as centralized versus distributed/collaborative processing will be explored to further enhance selectivity and reduce possible long-term drift via identifying and using response features from the sensors. Finally, Complete Case Analysis (CCA) and both Single- and Multiple-Imputation methods backed with advanced Machine and Deep Learning techniques such as K-Nearest Neighbours, Random Forest, Variational Autoencoders, Denoising Autoencoders, among others, will be investigated to implement data prediction strategies to handle, in an efficient way, the issues of missing data in long continuous data acquisition exercises, in which unexpected communication breaks or temporary sensor malfunction may occur.
