Neural Network Compound Predictor For Spirits in an Electronic Nose
An electronic nose was designed to investigate the digital response from MOS sensors for high concentration Ethanol spirits. Hardware was custom designed, and fabricated to classify spirits with 12 sensors for further analysis with a neural network. Electronic noses have been used for quality control with various foods and beverages in the past. Methods for classification and post processing techniques have varied from numerous publications. The popular methods used in the industry have been presented and their corresponding effectiveness has been analyzed. A systematic procedure has been created which can be used to recreate results for future sample collections. Seven different sensors were used to record time dependant chemical responses in a controlled environment. The sample responses were further used to acquire 7 unique features defining each sensor’s response. These 49 features were used as the inputs into a 45 hidden node, 3 node output, back propagating neural network. This network was able to classify 5µL samples of mixed concentrations of Ethanol, Ethyl Acetate and Isopropyl Alcohol to within 12.5% accuracy.