Time-varying prediction filter for structural noise reduction in ultrasonic NDE

Area of Science:

• Materials Science
• Non-Destructive Testing (NDT)
• Signal Processing

Background:

• Highly scattering materials exhibit frequency-dependent attenuation, where high frequencies are more attenuated than low frequencies.
• Structural noise in ultrasonic testing (UT) signals can be modeled as a non-stationary random process.
• Existing flaw enhancement techniques often overlook the frequency dependency of UT signals and assume material homogeneity.

Purpose of the Study:

• To present a novel UT signal processing technique that exploits the non-stationary characteristics of incoming signals.
• To improve the visibility of flaws in highly scattering materials by addressing frequency-dependent attenuation and noise.
• To develop a method that differentiates between structural noise and actual material flaws.

Main Methods:

• The proposed technique models structural noise using a linear and time-varying parametric model.
• It analyzes the prediction error generated by this model when processing UT echoes.
• A low prediction error indicates the presence of only structural noise, while a high prediction error signifies a flaw.

Main Results:

• Experiments conducted on stainless steel demonstrated the effectiveness of the proposed method.
• The technique achieved excellent signal-to-noise ratio (SNR) enhancement.
• The method successfully identified flaws by detecting non-predictable alterations in the material structure.

Conclusions:

• The developed UT technique successfully exploits the non-stationary nature of signals for improved flaw detection.
• The prediction error analysis provides a robust way to distinguish flaws from structural noise in scattering materials.
• This approach offers significant advancements in NDT for materials with high scattering properties.