%0 Thesis
%9 Doctoral
%A Tarttelin Hernandez, P
%B Security & Crime Science/Chemistry
%D 2017
%E Parkin, IP
%E Hailes, SMV
%F discovery:1549841
%I UCL (University College London)
%P 318
%T Modification of n-type and p-type metal oxide semiconductor systems for gas sensing applications
%U https://discovery.ucl.ac.uk/id/eprint/1549841/
%X This thesis investigates the modification of three metal oxide semiconductor gas  sensors with zeolite materials for the purposes of detecting trace concentrations of  gases that have an effect on health, security, safety and the environment.  SnO2, Cr2O3 and Fe2O3 were chosen as the base materials of interest. Zeolites HZSM-  5, Na-A and H-Y were incorporated into the sensing system either as  admixtures with the base material or as coatings on top of it. The aim of  introducing zeolites into the sensing system was to improve the performance of the  otherwise unmodified sensors.  Twenty-two novel zeolite-modified sensor systems are presented for the detection  of a range of hydrocarbons and inorganic gases. Whilst sensors based on SnO2  systems were more responsive to gases, some sensors were also found to provide a  greater degree of variability among repeat tests, particularly at lower operating  temperatures i.e. 300 °C. Cr2O3 sensors modified by admixture with zeolite H-ZSM-  5 were seen to be poorly sensitive to most analytes. Cr2O3 sensors modified by  admixture with zeolite Na-A and by overlayer of zeolite H-Y provided very  promising sensitive and selective results towards toluene gas. Sensors based on  the zeolite modification of Fe2O3 were not found to be promising candidates as gas  sensors at this stage.  Sensors were purposely exposed to gases that had similar molecular structures or  kinetic diameters to assess the true capability of the sensors to discriminate  among analytes. An array of four sensors based on n-type and p-type systems was  subsequently chosen to see whether machine learning classifiers could be used to  accurately discriminate among nine analytes. Using an SVM SMO classifier with a  polykernel function, the model was 94.1% accurate in correctly classifying nine  analytes of interest just after five seconds into the gas injection. Using an RBF  kernel function, the model was 90.2% accurate in correctly classifying the data into  gas type. These are very encouraging results, which highlight the importance of  furthering research in this field; a sensing array based on zeolite-modified metal  oxide semiconductor sensors may benefit a number of research domains by  providing accurate results in a very fast and inexpensive manner.