ULTRASONIC SIGNAL PROCESSING: SYSTEM IDENTIFICATION AND PARAMETER ESTIMATION OF REVERBERANT AND INHOMOGENEOUS TARGETS
Abstract
The nondestructive testing of multi-layered targets and targets with inhomogeneous and randomly distributed scatterers as in large grained materials have many important applications. Ultrasonic examination of such targets results in interfering multiple ecohes (reverberation) which complicates their evaluation by conventional techniques. This research consists of the analytical evaluation of backscattered echoes from sample targets coupled with the development of suitable digital signal processing techniques for their characterization. By decoupling the components of the backscattered echoes, an appropriate identification and classification technique is introduced which allows the characterization of the layered structure using detected echoes of significant intensities. Computer simulation was developed to verify the significance of the classification technique. The classification procedure allows the application of signal processing techniques such as subtraction, correlation, spectral analysis, and cepstral analysis. The subtraction technique is applied in order to separate various classes of echoes. This technique necessitates interpolation and synchronization of the digitized data. In this study an appropriate method of interpolation is presented based on the time-shift property of Fourier transforms. Correlation techniques are applied to the backscattered signal in order to improve the visibility of various classes of echoes. The correlation techniques improves the signal-to-noise ratio at the expense of resolution. The presence of the periodicity in the power spectrum can be related to layer thickness which is experimentally verified. Cepstral analysis is also appropriate for the processing of reverberant echoes in order to extract desired features. In this study, cepstral processing is used for separation of echoes, and extraction of the averaged echo waveshape for use in deconvolution. Results demonstrate that the power cepstrum provides good resolution. Various signal processing techniques, in both the time and frequency domains, have been applied to backscattered signals for grain size evaluation. Time domain analysis consists of time averaging, autocorrelation functions, and determination of the probability density function of the backscattered signals. Time averaging demonstrates significant sensitivity to grain size variation and also provides good reproducibility. Autocorrelation functions of the data were not informative since no periodicity exists in solids. It is shown that relative changes in statistical parameters (e.g., mean and standard deviation) of the probability density functions are also feasible for grain size evaluation. Quantitative evaluation of the magnitude spectra of backscattered signals were assessed by moment analysis. Moment values show inadequate sensitivity to grain sizes due to the presence of random peaks and valleys in the spectra. Furthermore, the magnitude spectrum were cepstrally smoothed in order to obtain an estimate of attenuation in the backscattered signals as a function of frequency. Results demonstrate a moderate performance of this technique for grain size evaluation.
Degree
Ph.D.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.