AN ELECTRONIC NOSE SYSTEM BASED ON EVOLUTIONARY COMPUTATION AND SIMILARITY MEASURES FOR CLASSIFICATION AND QUANTIFICATION OF GASES

Abstract

Electronic Nose System (ENS) is an electronic device which is a composition of sensors array and artificial intelligence algorithms. ENS is built to mimic the functionality of biological olfaction in order to help humans in the detection and quantification of several chemical compounds present in the environment. In the fast-growing technological world, developing a robust ENS is a crucial need because of its several applications in different industries. The development of miniaturized sensors array has become feasible in the recent decade due to the great advancement in the fabrication process and characterization techniques of sensing materials. However, state-of-the-art artificial intelligence algorithms are mostly investigated for odors identification and quantification. Although the state-of-the-art algorithms perform well in terms of recognition accuracy, a portable electronic nose system remains a challenge due to the computational complexity associated with these algorithms. Moreover, the issue of sensors replacement/recalibration to handle long term sensors drift also restricts the field deployment of an ENS. _x000D_ The focus of this work is to develop machine learning models that are capable of dealing with the issues of computational complexity and long term sensors drift. The advantage of developing machine learning models over other methods is that these models adapt the classifier to the sensors drift without explicitly describing the drift, making an ENS feasible for long term deployment in the field. _x000D_ We have developed machine learning models based on evolutionary computation and similarity measures which are capable of classifying and quantifying different gases. Initially, we have proposed a computationally efficient model based on Recursive Discrete Binary Particle Swarm Optimization (RDBPSO) and Euclidean Distance (ED) to classify different gases, without considering the sensors drift. Afterward, the proposed model is extended by incorporating different similarity measures to further enhance the computational complexity and to test it against the sensors drift. The proposed models are tested for both the classification and the quantification of gases and a significant improvement is observed in terms of saving memory requirement, computational cost, and robustness against drift.

Date of Award	2019
Original language	American English
Awarding Institution	HBKU College of Science and Engineering