The electronic nose is a device developed to mimic the human olfactory system. Despite raising interest from applications in the field of medicine, quality control, environmental control and security, such devices remain inferior to their biological counterparts. As the biological system is explored further, new discoveries generate new ways of thinking in creating electronic nose devices. This has led to a large variety of sensors and devices, all of which produce data that requires processing. The data are processed to extract information that can be used to classify or quantify the input to the electronic nose. However, as the devices have advanced, the data processing techniques have remained relatively static, refinements of established statistical methods. Recently, investigation into the phenomenon of nasal chromatography has brought about the development of a new class of electronic nose device; the artificial olfactory mucosa. Taking advantage of a retentive effect, inspired by the aqueous mucous layer covering the olfactory epithelium, this new device produces data whose spatio-temporal properties have not been seen in the field of chemical sensing before. Thus there is a need to develop new processing approaches to obtain the information being produced by these new devices. In this thesis, a new processing approach is presented, centred on the use of convolution to produce characteristic signals which contain information arising from a sensor space that is separated both spatially and temporally, realised in the form of multiple sensor arrays separated by retentive columns or channels. This combined signal is then used to extract an information rich feature set that can be passed on to classifiers or quantifiers to make practical use of the data. This method is simulated on data collected during the development of the artificial olfactory mucosa to validate its use, and then applied to several sets of real world data, collected from a variety of devices; from current e-nose technologies to newly developed artificial olfactory mucosa devices. The simulations put the device in very noisy conditions and the processing approach deals well with a high level of noise in most circumstances, its performance only deteriorating in the presence of extremely high levels of sensor drift. However, it is shown that this method not only has validity when dealing with the advanced devices for which it is intended, but also shows an improvement over standard processing approaches when utilised in conjunction with current technologies. Utilising convolution on data collected from current devices, methods are developed where the characteristic signal can be generated internally from a single array, and when applied, produce improvements over standard processing approaches.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:541162 |
| Date | January 2010 |
| Creators | Taylor, James E. |
| Publisher | University of Warwick |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://wrap.warwick.ac.uk/38501/ |
Page generated in 0.0023 seconds