Artificial neural network: Gas recognition

Abstract

The objective of this paper is to describe development of gas recognition tool based on Artificial Neural Network (ANN). This recognition tool has capability to recognize five different gases: ammonia, acetaldehyde, acetone, ethylene, and ethanol. Developed ANN is trained using data from the UC Irvine Machine Learning Repository database from October, 2013. The implemented system for gas recognition uses following input parameters: concentration of gas (ppmv), flammability, constant pressure (kJ/kgK), constant volume (Kj/kgK), specific heat capacities (cp/cv) and molecular weight (g/mol). Developed neural network consists of 30 neurons distributed in a single hidden layer. For purpose of training 174 samples were used. Testing dataset contained 64 samples, 38 of which were used as a testing set. With 36 samples correctly classified resulting in accuracy and specificity were 97.37%. These results were obtained after adjusting neural network using several different parameters which is explained in this paper.