| This thesis describes a portable electronic nose for detecting and classifying ten different volatile organic compounds. The hardware for this instrument consisted of a sensor array composed of four commercial Taguchi Gas Sensors, a MSP430 microcontroller, and a sample delivery system. Using the portable electronic nose, the dynamic sensor responses from the Taguchi Gas Sensor-array were extracted by the microcontroller. Different pattern responses from the sensors generated "fingerprints" for each of the ten chemicals. Peak features were extracted from sensor responses of each chemical. The software part of the electronic nose performed data analysis, data visualization by Principle Component Analysis (PCA), and pattern recognition by decision tree analysis. The decision tree employed for gas classification was a J48 classifier based on a C4.5 algorithm. The weights from the trained decision tree for chemical identification were programmed into the microcontroller, making the portable electronic nose a standalone device (independent from PC). PCA plots showed that 96.66% information of the sensor responses were captured by the first two principal components and the regions representing the ten chemicals were well separated. Furthermore, the decision tree classifier yielded 78.67% of correct classification. |
