Neuro-Identification of Some Commonly Used Volatile Organic Compounds Using Electronic Nose

Abstract

Electronic nose system comprising of three Figaro sensors (TGS2602, TGS832 and TGS816)  arrayed in a chamber interfaced through  PICO ADC11/10 card to a computer system loaded with artificial neural network (ANN) was used to identify three  volatile organic compounds (Formaldehyde, Acetone and Chloroform). The back propagated ANN had four layers having positive linear, logsigmodal, logsigmidal  and tansigmodal transfer functions  with 10, 20, 20, 1 neurons, respectively. 60% of the acquired data was used for training and 20% each for testing and validation.TGS832 had the highest average sensitivity (1.3639 volts) while TGS816 had the least (0.0420 volts) for formaldehyde with similar trend for chloroform and acetone. Sensors’ sensitivities were significantly different from the control at p < 0.05. Mean square error of 0.0006, 0.0001 and 0.0003(R²:0.996, 0.997 and 0.996) were obtained for the ANN training of formaldehyde, chloroform and acetone respectively. Validation run gave correct identification of the VOCs.

Key Words: Sensors, Neurons, Normalization, Electronic nose, Data card