DEVELOPMENT OF E-NOSE FOR DETECTION OF HAZARDOUS GASES USING MACHINE LEARNING APPROACH

Subject Area: Artificial Intelligence

This paper presents the development of an e-nose for the detection of hazardous gases using a machine learning approach. The aim was to develop an intelligent system for accurate detection of harmful gases contained in the atmosphere of mining sites, which are majorly Carbon (ii) Oxide (CO) and Methane (MH 4 ), mostly released into the atmosphere during mining activities and have resulted to the death of many workers at the mining site over the years. This issue was solved using a machine learning approach. The methodology employed data collection of hazardous gasses using MQ-7 and MQ-2 sensors, and feature extraction using the dynamic moment technique. Feed-Forward Neural Network (FFNN) developed with the tanh activation function and back-propagation training algorithm was adopted and used to train the features to generate the gas detection algorithm. The gas detection algorithm was implemented with Simulink and evaluated. The mean square error and the regression results were obtained, analyzed, and validated via tenfold cross-validation. The MSE and regression values were 2.5113e-07Mu and 0.9944 which implies that the system developed was reliable and efficient for the detection of said hazardous gas found in mining sites.