#### Location

Brunswick, ME

#### Start Date

1-1-1992 12:00 AM

#### Description

Ultrasonic flaw detection in large grained materials is limited by the high level of coherent grain noise due to interfering and attenuating random scatterers that often masks the flaw signal, leading to difficulties in its detection. Several techniques have been developed in the past to reduce grain noise and enhance flaw visibility. A nonlinear frequency diverse statistical filtering technique, also called split-spectrum processing (SSP), has been used to enhance flaw detection with considerable success [1, 2]. This technique is illustrated in Fig. 1. The wideband input signal x(t), which in general consists of both the flaw signal and the grain noise, is first transformed into the frequency domain using the fast Fourier transform (FFT). The transformed signal spectrum is then split into N narrowband spectra in the frequency domain using parallel bandpass filters. The narrowband spectra are then transformed back to the time domain using inverse Fourier transform and weighted by factors w1 to wN, where the weighting factors wi are chosen such that the amplitude of each narrowband signal is normalized to unity. The N narrowband signals are subsequently processed using various linear and nonlinear operations. In this paper, we concentrate on the Order Statistic (OS) filter, and examine how the statistical characteristics of the narrowband signals (ie., SNR variations) affect the choice of processing order for the SSP technique.

#### Book Title

Review of Progress in Quantitative Nondestructive Evaluation

#### Volume

11A

#### Chapter

Chapter 3: Interpretive Signal Processing and Image Reconstruction

#### Section

Signal Processing

#### Pages

943-950

#### DOI

10.1007/978-1-4615-3344-3_121

#### Copyright Owner

Springer-Verlag US

#### Copyright Date

January 1992

#### Language

en

#### File Format

application/pdf

Rank Determination of Order Statistic Filters for Ultrasonic Flaw Detection

Brunswick, ME

Ultrasonic flaw detection in large grained materials is limited by the high level of coherent grain noise due to interfering and attenuating random scatterers that often masks the flaw signal, leading to difficulties in its detection. Several techniques have been developed in the past to reduce grain noise and enhance flaw visibility. A nonlinear frequency diverse statistical filtering technique, also called split-spectrum processing (SSP), has been used to enhance flaw detection with considerable success [1, 2]. This technique is illustrated in Fig. 1. The wideband input signal x(t), which in general consists of both the flaw signal and the grain noise, is first transformed into the frequency domain using the fast Fourier transform (FFT). The transformed signal spectrum is then split into N narrowband spectra in the frequency domain using parallel bandpass filters. The narrowband spectra are then transformed back to the time domain using inverse Fourier transform and weighted by factors w1 to wN, where the weighting factors wi are chosen such that the amplitude of each narrowband signal is normalized to unity. The N narrowband signals are subsequently processed using various linear and nonlinear operations. In this paper, we concentrate on the Order Statistic (OS) filter, and examine how the statistical characteristics of the narrowband signals (ie., SNR variations) affect the choice of processing order for the SSP technique.