Year

2005

Degree Name

Doctor of Philosophy

Department

School of Information Technology and Computer Science - Faculty of Informatics

Abstract

Object recognition systems based on ultrasonic sensing have significant drawbacks in generality, resolution and speed. The objective of our research was the development of more efficient technique(s) for ultrasonic based object recognition through the investigation of models of acoustic backscatter, with particular emphasis on the work of Albert Freedman. The �image pulse� model developed by Freedman calculates the echoes generated from convex objects in an underwater environment after insonification with a narrowband transient signal. The primary prediction of this model is that echoes are generated at those points along a scattering body where there are step discontinuities in the derivatives, with respect to range, of the solid angle subtended at the transducer by the scatterer, the amplitudes of the echoes being a linear combination of the magnitudes of said discontinuities. We extended this model for use in an air environment using non-coincident transmitters and receivers and conducted experiments to measure the amplitudes of the echoes from a range of radially symmetric convex objects, at distances up to 1.4m, after insonification with a Polaroid transducer. These amplitudes were compared to those predicted by the model, with the results for the cones highlighting the limitations of the theory at modelling the echoes from the geometrical shadow boundaries of objects. The results for the spherical objects were significantly better however, with an average error of less than 5%, suggesting that the model should be reasonably accurate at calculating the echoes from convex objects with smoothly varying surfaces. The extended forward model was then inverted to produce an inverse model that would calculate the geometrical parameters of a radially symmetric scattering body from an analysis of the echoes received after insonification of these bodies with ultrasonic pulses at two discrete frequencies. A quantitative verification of this inverse model with various scattering bodies proved elusive, with a low correlation between experiment and theory, due to matrix instability and difficulties in obtaining data of sufficient accuracy. However, qualitative trends in the data indicate that the model is essentially correct, though very sensitive to measurement precision and media characteristics, and there is good reason to believe that further work under more controlled laboratory conditions and/or a different medium would verify the model�s validity quantitatively. Finally, the inverse model was tested to see whether it could find a practical application despite its quantitative limitations. In many industries, quality control involves distinguishing between those items that are physically damaged and those that are not, a task that the inverse model may be able to address. Using glass bulbs as the test subjects, some with simulated physical damage and some without, we tested the ability of the inverse model to distinguish between these two classes of objects. In all cases, the model clearly separated the items with simulated damage from those without. The inverse model should be of interest to workers in the field of industrial quality control because of its potential to lead to the development of real-time inspection systems for production lines that could perform with a higher efficiency than the visual inspection procedures currently being employed.

02Whole.pdf (2130 kB)

Share

COinS
 

Unless otherwise indicated, the views expressed in this thesis are those of the author and do not necessarily represent the views of the University of Wollongong.