Frequency-Diverse Geometric-Mean And Arithmetic-Mean Filtering For Ultrasonic Flaw Detection

J. Xin,K. D. Donohue,Nihat M. Bilgutay,X. Li

1991-01-01

Materials Evaluation

Abstract:Split-spectrum processing (SSP) involves ensembles of frequency-diverse signals obtained by passing the broadband ultrasonic A-scan through a bank of bandpass filters. Although applying certain non-linear statistical filters to the SSP ensemble significantly improves flaw detection in large-grained materials, the filter performance is very sensitive to the frequency range of the bandpass filter bank because the expected flaw-to-grain echo power ratio typically varies over the spectrum of a broadband transducer. This paper introduces a new SSP statistical filter for flaw enhancement, referred to as the geometric-mean filter, which exhibits performance comparable to that of other nonlinear SSP filters while maintaining consistent performance over a broader range of frequencies. Thus, this filter requires less tuning of SSP parameters to obtain acceptable performance when processing A-scans from various large-grained material samples. This paper presents a theoretical analysis of the geometric-mean filter that incorporates the nonidentical distributions of the flaw- and grain-echo amplitudes over the spectral region used for processing. The geometric-mean performance is compared to the performance of a linear SSP filter referred to as the arithmetic-mean filter, which has low performance sensitivity relative to other nonlinear SSP filters. Theoretical and experimental results show that the geometric-mean filter provides better flaw-enhancement capabilities and exhibits lower performance sensitivity than the arithmetic-mean filter over a broader variation of the SSP parameters.