| Description |
This thesis focuses on characterizing and modeling three gas sensor technologies and applying an inversion-based feedforward dynamics compensator to improve aerial robotbased chemical plume mapping and swarm-based source localization. With recent advances in the design and deployment of unmanned aerial vehicles (UAVs), fast and effective autonomous robotics-based chemical plume mapping and localization are critical to minimize impact, cost, and recovery time following natural disasters, accidents, and malicious attacks. Additionally, the majority of multi-rotor hover-capable UAVs, such as quadcopters, are operationally constrained in terms of power and flight times, thus a thorough understanding of sensor performance and characteristics and leveraging such information for implementation helps to maximize performance. Two commercially available gas sensors, such as a metal oxide (MOX) gas sensor and a non-dispersive infrared (NDIR) gas sensor, and a newly developed micro-electro-mechanical systems (MEMS) technology-based gas sensor are characterized, specifically focused on the determination of the rise and recovery behaviors. The dynamics of each sensor are then modeled using transfer function models, where model parameters are informed by the characterization results. To account for sensor dynamics in the chemical mapping process, the models are inverted for feedforward compensation. A simulator is developed using the sensor models to evaluate gas distribution mapping performance, with and without model-inverse feedforward compensation. As part of this assessment, a maximum scanning velocity for each sensor is determined that is inversely proportional to the sensor time constant and used as a reference velocity in testing. Results of the simulated mapping show a significant reduction in root-mean-squared (RMS) mapping error, as well as improved qualitative map accuracy, with the compensation technique applied. For example, results show a 55.8% average reduction in RMS error over all test cases, and the enabling of mapping velocities greater than five times the theoretical maximum velocity associated with uncompensated sensor operations, while maintaining acceptable mapping quality. Finally, the performance of the sensors in supporting multi-agent plume tracking in simulated scenarios utilizing a particle swarm optimization (PSO) algorithm is evaluated for both the compensated and uncompensated cases. Results show that model-based feedforward compensation increases the performance of the PSO algorithm in terms of successful convergence of the agents to the true location of the highest concentration of the source. For example, in the static Gaussian plume model case, a 64% average improvement was observed. For a dynamic Quick Urban & Industrial Complex (QUIC) dispersion modeling case, a 39% average improvement was observed. In summary, compensating for sensor dynamics improves concentration mapping accuracy and swarm-based plume tracking performance, as well as enabling increased mapping and plume source tracking speeds, thereby reducing the time required by an agent to effectively cover an area of interest. These improvements are particularly advantageous for systems with constrained operational times, such as the multi-rotor hover-capable UAV chemical sensing systems. |
