Approach for drift representation and extraction in gas sensors signals by sample entropy

Abstract

Humans and animals perceive the surrounding environment using the physiological mechanisms of perception, commonly called senses (GUERRINI et al., 2017). Bio-inspired by the biological olfactory system, the development of artificial devices that combine chemical sensors array with pattern recognition techniques, commonly termed as “electronic nose” (E‐Nose), have been used for recognition of Volatile Organic Compounds (VOCs). The gas sensors' response may contain some disturbances (noise and drift) composed of multiple frequencies,affecting signal processing tasks' performance. The present thesis focused on analyzing the drift behavior in signals from gas sensors used in artificial olfactory devices. For this purpose, one extensive database was used, reported in the literature as a real database with severe drift issues. An exploratory analysis was performed over that database using discrete Wavelet transform, observing the presence of drift, noise perturbance, and the existence of outliers, making it more challenging to treat that database. Additionally, it was estimated the influence of drifts based on Sample Entropy to establish the dynamics caused in the signals of E-Nose. Finally, it was generated several work scenarios using synthetic measurements generator. I was sought to explore the effect of drifts on different portions of signals from electronic nose systems, analyzing the performance of the rapid detection method for electronic nose systems using artificial data.