Odor-source Location Estimation And Search Using A Movable Sensor Array

The research on odor source localization, which involves the knowledge of sensingtechnology, robotics and artificial intelligence, has broad application prospect in thefields such as industrial toxic/harmful gas leak monitoring, fire detection and disasterrescue.The mode of mobile sensor array, i.e., a mobile robot equipped with a fixedtopology of gas sensor array, is adopted in this thesis to solve the odor sourcelocalization. Compared with single gas sensor, the gas sensor array could detect largerrange at the same time, thus the robot could estimate the odor source more efficiently.This thesis focuses on odor plume tracing and odor source declaration, and thefollowing research work has been carried out.1. The state-of-the-art odor source localization has been reviewed, and the existingproblems have been analyzed. The current behavior basedodor-source-localization strategies have been emphasized, including plumefindings, plume tracing and odor source declaration.2. Two plume tracing methods, which are square root unscented Kalman filter(SRUKF) algorithm and nonlinear least square (NLS) algorithm, are proposed.The proposed plume tracing methods are based on the assumption that the windfield is uniform in a small area, and the time-averaged Gaussian-like analyticplume model is adopted. Furthermore, the SRUKF algorithm is improved byfusing multi-sensor information and using multiple random initial points.3. The K-means clustering algorithm is proposed to declare single odor source.The proposed algorithm is based on the fact that, in real indoor environmentswith weak wind, the measured concentration increases with a decrease in thedistance between the mobile sensor array and the source.4. The proposed plume tracing and odor source declaration algorithms have beenvalidated in indoor/outdoor dynamic plume environments and real indoorexperimental environments. In the simulation environments, the spiralalgorithm is adopted to find plume, and the NLS and SRUKF are tested toestimate the source location and control the mobile sensor array to trace theplume. In the real indoor environment, the NLS combined K-means algorithms are used. Both simulation and experiment show that the proposed algorithmshave relatively high success rate and localization accuracy.