New Methods For Ultrasonic NDE Of Difficult Materials

Project: Research

Project Details


This project will investigate a range of methods for improved ultrasonic NDE of difficult materials. The approach will involve a combination of modelling, novel transducer design and array signal processing. The project has 5 industrial partners, covering three main application areas: aerospace, nuclear and petrochemical. For each industrial application, array performance will be evaluated on real industry samples using a variety of signal processing techniques, including spatial and frequency compounding, allied with advanced beam forming methods. In addition, time-frequency signal processing and spatial filtering techniques will be integrated to enhance the defect detection/imaging in highly scattering materials. This work is underpinned by the materials characterisation, University of Manchester, which will feed into the simulation and processing work packages.

The key objectives at the start of the project were:
• To create models for evaluation of multi-frequency, steered array performance in different materials
• To design, manufacture and test a set of wide band, ‘low frequency’ array transducers
• To evaluate different beam steering algorithms in the laboratory
• To evaluate different clutter reduction techniques in the laboratory
• To perform a comprehensive set of laboratory trials on challenging industrial sample microstructures.

Layman's description

This is an EPSRC/RCNDE targeted project to investigate new ultrasonic array techniques appropriate for application in highly scattering or attenuating materials. The industrial partners have a wide range of materials of interest. The project has four main research themes: experimental data collection of both NDE data and material microstructure characterisation; wideband ultrasonic transducer design and development; ultrasonic array system modelling; and signal/image processing through frequency domain techniques and advanced beam forming.

Key findings

In many engineering applications, materials that exhibit heterogeneous or otherwise acoustically scattering microstructure are employed, examples include austenitic steels and alloys, concrete and fibre reinforced composites. In ultrasonic non-Destructive Evaluation (NDE) of such media, defect target signal is obscured by clutter echoes, caused by numerous, relatively small (relative to the ultrasonic wavelengths) stationary reflectors, which form part of the internal microstructure of the material. The extent of this clutter can be significant and even defects that are larger than these randomly scattering regions can be difficult obscured. This type of time-invariant clutter cannot be reduced by the standard time averaging or correlation techniques that are used to reduce time varying random electrical noise. Accordingly, defect identification invariably involves a compromise between achievable resolution, which is determined partly by wavelength in the material, and the noise arising from scattering in the propagation medium. This project has investigated a range of methods for improved ultrasonic NDE of difficult materials, encompassing analytical modelling, experimental evaluation of industrially relevant samples, novel transducer design and array signal processing methods.
Industrial test samples and standard test blocks were experimentally evaluated in the CUE Facility for Innovation and Research in Structural Testing (FIRST) over a range of array operating frequencies. In addition, some industrial samples were characterised using both Electron Backscatter Diffraction and Spatially Resolved Acoustic Spectroscopy. An efficient methodology was developed to translate the microstructure data from these techniques into finite element models, using the PZFlex platform. Both the experimental and simulation approaches contributed to a database of ultrasound datasets, on which the analytical modelling and signal processing developments were evaluated.
Mathematical modelling methodologies utilised in this work to enhance the interpretation of features from array data acquired in high clutter noise structures included: the Factorisation Method; the Fractional Fourier Transform; and the Born Approximation. The ability of putative inversion algorithms to recover crack properties was demonstrated on experimental data using the Factorisation Method, with the Born Approximation used as the basis for objective crack sizing. Importantly, this was developed for use with features within low SNR images.
Array transducer bandwidths in excess of 100% were manufactured within this project using graded matching layer techniques. This enhanced bandwidth at the transducer front-end was shown to enhance target detection through the developed frequency domain signal processing techniques developed.
Both temporal and spectral signal processing techniques have been implemented to minimise clutter noise, whilst maintaining high quality feature details through array processing NDE datasets. In the spectral domain, adaptive frequency compounding, using Best Linear Unbiased Estimation, has experimentally demonstrated improvements speckle noise suppression in Inconel (31dB) and steel (40dB). In the temporal domain, two new algorithms have been developed: Spatially Averaged Sub-Aperture Correlation Imaging and Correlation for Adaptively Focused Imaging. Both approaches use cross-correlation to suppress incoherent backscatter signal components.
As the project developed it became apparent that the developed algorithms and techniques would require a software framework to enable a more thorough evaluation of their relative performance metrics. Towards this aim, a software platform, cueART, was developed to operate on GP-GPU hardware architectures. Thus, a number of the techniques were converted from Matlab to CUDA to improve their computational efficiency by taking advantage of the parallelization available through the GP-GPU units.
Effective start/end date1/08/1131/07/14


  • EPSRC (Engineering and Physical Sciences Research Council): £263,701.00


Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.