Ocean based robotic systems are an opportunity to combine the power of acoustic sensing in the water with sophisticated control schemes. Together these bodies of knowledge could create autonomous systems for mapping acoustic fields and localizing underwater sources. However, existing control schemes have often been designed for land and air robots. This creates challenges for applying these algorithms to complex ocean environments. Acoustic fields are strongly frequency dependent, can rarely be realistically modeled analytically, have complex contours where the feature of interest is not always located at the peak pressure, and include many sources of background noise. This work addresses these challenges for control schemes from three categories: feedback and observer control, gradient ascent control and optimal control. In each case the challenges of applying the control scheme to an acoustic field are enumerated and addressed to create a suite of acoustically driven control schemes. For many of these algorithms, the largest issue is the processing and collection of acoustic data, particularly in the face of noise. Two new methods are developed to solve this issue. The first is the use of Principal Component Analysis as a noise filter for acoustic signals, which is shown to address particularly high levels of noise, while providing the frequency dependent sound pressure levels necessary for subsequent processing. The second method addresses the challenge that an analytical expression of the pressure field is often lacking, due to uncertainties and complexities in the environmental parameters. Basis functions are used to address this. Several candidates are considered, but Legendre polynomials are selected for their low error and reasonable processing time. Additionally, a method of intermediate points is used to approximate high frequency pressure fields with low numbers of collected data points. Following this work, the individual control schemes are explored. A method of observer feedback control is proposed to localize sources by linearizing the acoustic fields. A gradient ascent method for localizing sources in real time is proposed which uses Matched Field Processing and Bayesian filters. These modifications allow the gradient ascent algorithm to be compatible with complex acoustic fields. Finally, an optimal control method is proposed using Pontryagin's Maximum Principle to derive trajectories in real time that balance information gain with control energy. This method is shown to efficiently map an acoustic field, either for optimal sensor placement or to localize sources. The contribution of this work is a new collection of control schemes that use acoustic data to localize acoustically complex sources in a realistic noisy environment, and an understanding of the tradeoffs inherent in applying each of these to the acoustic domain.
Identifer | oai:union.ndltd.org:bu.edu/oai:open.bu.edu:2144/30725 |
Date | 03 July 2018 |
Creators | Bogdan, Caitlin |
Contributors | McDaniel, J. Gregory |
Source Sets | Boston University |
Language | en_US |
Detected Language | English |
Type | Thesis/Dissertation |
Page generated in 0.0024 seconds