Fault-tolerant gas sensor array architecture for artificial olfactory systems

• Main Author: Lotfivandvahed, Nasser
• Format: Thesis
• Language: English
• Published: 2013
• Subjects: Olfactory sensors - Design - Research; Smell
• Online Access: http://psasir.upm.edu.my/id/eprint/91223/1/ITMA%202013%2010%20IR.pdf; http://psasir.upm.edu.my/id/eprint/91223/
• Institution: Universiti Putra Malaysia

During the past two decades, artificial olfactory system has received considerable attention, mainly because of the findings in wide range of applications arising from research in various areas of applied sciences, Artificial olfactory system has been employed in a broad range of industries, including the food, agricultural, biomedical, pharmaceutical, and diverse scientific fields, Most of the efforts to develop artificial olfaction have been in sensor fabrication, sensor structure, and signal processing techniques. I Iowever, major problem in the reported investigations is that the classic structure of the artificial olfactory system is not suitable for critical applications such as medical diagnosis and environment air quality monitoring where reliability and precision arc important. In this thesis, development of the artificial olfactory system based on the fault tolerant architecture was investigated. The proposed architecture can resistant or tolerant failures in the sensor array of artificial olfactory system, that is, fault tolerant. The system is able to continue to function in spite of failures in the sensors of array. The result is a system that can suffer from failure and damage but this harm is not effect on its performance and system is capable to recover faults without direct human intervention. In the proposed architecture, by applying a novel technique which is called 'virtual sensor' the occurred faults are masked and generation of erroneous results is prevented. In this thesis, the ability of the developed architecture based on the virtual sensor to discriminate complex odors and also, the performance of system in faulty situation was studied. For this purpose, from different evaluation experiments various datasets with large number of data from 26,400 to 108,000 were generated. The results demonstrated that the performance of the system based on the proposed architecture in healthy mode, is similar to the classic structure of artificial olfaction. However, in faulty mode, the classifier based on the proposed architecture in comparison with the generic architecture presents 70.77%, 64.47%, 64.45% enhancement in precision, sensitivity, and accuracy rates, respectively. Thus, the proposed architecture can be seen as a considerable improvement in artificial olfactory system.