Descripció del projecte
The PhD candidate will work on the research and development of the AI algorithms applied to gas sensors to widen the applications that unspecific, low-cost sensors have nowadays. The research will be conducted in two scenarios: activity monitoring in indoor settings and gas leak detection in industrial settings.
A main research line will enable monitoring of Daily Activities using unobtrusive gas sensors. Unlike other systems, gas-based systems are sensitive to large number of activities such as room occupancy, showers, cooking, opening/closing doors, visits to the bathroom, irregular night activity, etc. Actually, the detection of any activity that changes air composition comes naturally with the use of gas sensors. Previous studies have shown that chemical gas sensors can improve room occupancy predictions. For example, an array of polymeric gas sensors was placed in a 200 m 3 room with semi-controlled conditions used by the JPL-NASA to simulate spaceship cabin atmosphere. Several volunteers performed physical activity and different common daily activities. It was possible to predict the level of activity performed in the room and detect the use of ethanol-based medication (Fonollosa et al. 2014). More recently, it was showed that under simple and controlled conditions, all indoor climate parameters are highly correlated with occupant presence (Pedersen 2017). Results showed that room occupancy can be predicted with standalone measures of carbon dioxide or total volatile organic compounds in a test-room. However, when the system was placed in a three-room dorm apartment shared by two persons, performance of standalone sensors decreased significantly and they were coupled to PIR sensors. In this scenario the research will be directed to development of new algorithms trained to monitor activity at home.
Linked to the previous research, the continuous monitoring algorithms will be adapted and redesigned to detect gas leaks and monitor air composition. To improve the conventional methods for industrial gas leak detection using artificial intelligence algorithms, we propose the following: Improvement of the current state of the art inverse gas dispersion algorithms both in precision and in efficiency. We propose to test various AI algorithms over the data; then by finding the most suitable algorithms, modifying current AI methods, or combining them, we can ensure the strength and high performance of our model. The general approach would be to use historical and simulated gas leak data and combine it with site specific meteorological (historical and real time) and topographical data to train the model to identify the strength and weakness of each algorithm.
In summary, the research will develop new algorithms for gas sensors, in particular for low-cost technologies. The algorithms will help to alleviate common limitations of gas sensors such as drift, cross-sensitivity and aging. Different techniques will be explored and standardization methodologies to reduced calibration costs and extend system lifetime will be explored. Using bench-marking algorithms the best set of algorithms will be selected for each target scenario.
