Source Characterization Via Autonomous Aircraft
Open Access
- Author:
- Kuroki, Yuki
- Graduate Program:
- Meteorology
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- April 02, 2008
- Committee Members:
- George Spencer Young, Thesis Advisor/Co-Advisor
- Keywords:
- Guassian Plume
Gaussian Puff
Expertsystem
Autonomous Aircraft
Genetic Algorithm - Abstract:
- Given numerous concentration observations it is straightforward to find the source of a toxic atmospheric release, to produce a prediction of contaminant impacted area, and to plan evacuation. Human and fiscal resources, however, limit the number of fixed sensors to levels that may be inadequate for these tasks. This study explores the possibility of supplementing a sparse fixed sensor network with unmanned aerial vehicles (UAV) to increase accuracy and to save cost. First we use the Gaussian plume model to simulate how a toxic gas disperses in the atmosphere. Then we back calculate the source location, strength and the wind direction with a Genetic Algorithm (GA). Testing and tuning the GA was conducted to ascertain the best population size, mutation rate and number of generations for achieving a rapid and accurate back calculation. We have developed an ensemble method for improving accuracy by conducting multiple GA runs, each with its own UAV and solution population, and using a median-based consensus for the final answer. Testing conducted with simulated concentrations thresholded to resemble real sensor data and with random weather and source locations suggests the real world applicability and utility of our method. A new navigational intelligence of the expert system type was developed for the simulated UAVs, allowing them to collect the data required for GA-based source characterization given only one fixed sensor. Thus we require development of a new navigational strategy for the UAV to use the actual observations from the fixed sensor and the UAV itself to direct the aircraft in gathering concentration data. Different expert systems were required for plume and puff situations. The results demonstrate that our expert system approach works for both plume and puff releases if given the observations from one fixed sensor and from the simulated UAV itself. These model runs demonstrate that the coupling of autonomous aircraft navigation with a GA has the potential for application to source characterization in real world problems.