Improving Inspection Reliability through Data Fusion of Multi-View Array Data

Lead Research Organisation: University of Bristol
Department Name: Mechanical Engineering

Abstract

The objective of this project is to obtain a step-change improvement in the detection and characterisation of defects in safety-critical components across a range of industries including nuclear power generation and the defence sector. This will be achieved through data-fusion of the multiple views of a component's interior that can be obtained through modern ultrasonic array imaging techniques. Previous work by the team has demonstrated a two-order-of-magnitude improvement in detection performance when data fusion was applied to ultrasonic data obtained from separate scans performed with single-element probes. This was in a case where the expected defects were small, point-like inclusions that scatter roughly uniformly in all directions. The proposed project will develop the data-fusion philosophy for improving defect detection performance from multi-view array data in the much more complex case where the defect morphology cannot be assumed in advance and the scattering pattern may be strongly directional. Therefore, the project will necessarily address the critical challenges of applying data fusion to defect classification and sizing from multi-view array data. Demonstrator software will be produced that will show an image of the test component with indications ranked by the probability of them being produced by a defect; it will then be possible to probe any of these indications to show detailed classification (e.g. crack, void, inclusion etc.) and sizing information. The project is supported by EDF, Hitachi, BAE Systems and AMEC Foster Wheeler.

Publications

10 25 50
publication icon
Bevan RLT (2020) Data Fusion of Multiview Ultrasonic Imaging for Characterization of Large Defects. in IEEE transactions on ultrasonics, ferroelectrics, and frequency control

publication icon
Bevan RLT (2019) Experimental Quantification of Noise in Linear Ultrasonic Imaging. in IEEE transactions on ultrasonics, ferroelectrics, and frequency control

publication icon
Budyn N (2019) A model for multi-view ultrasonic array inspection of small two-dimensional defects. in IEEE transactions on ultrasonics, ferroelectrics, and frequency control

publication icon
Wilcox PD (2018) Quantification of the Effect of Array Element Pitch on Imaging Performance. in IEEE transactions on ultrasonics, ferroelectrics, and frequency control

publication icon
Wilcox PD (2020) Fusion of multi-view ultrasonic data for increased detection performance in non-destructive evaluation. in Proceedings. Mathematical, physical, and engineering sciences

 
Description An extremely efficient technique for modelling the sensitivity of multi-view ultrasonic imaging algorithms to small defects of any shape was developed and published. In parallel, a robust experimental protocol for quantifying noise (both random and structural) in multi-view ultrasonic array images of engineering components has been developed and published. Together, these two tools provide the essential cornerstones for fusing multi-view ultrasonic images. Building on this work, a number of methodologies have been developed for fusing multi-view ultrasonic images to increase defect detection performance. The most effective methodology for improving detection performance in edge cases (small defects at the limits of detectability in any view) is a modified matched filter approach. This improved edge-case performance requires accurate (sub-millimetre) image co-registration, which is not always achievable in practice. Even if co-registration with this level of accuracy cannot be achieved, data fusion using other techniques has been shown to lead to significant performance improvements in the general case (defects clearly visible in at least one view, but the relevant view depends on defect, location and orientation). This has the potential to significantly reduce operator burden and inspection cost as well as improving reliability. We are also looking at characterisation of detected defects. Work on data fusion to improve detection performance has been submitted for publication. For small defects, a Bayesian framework has been developed that compares the measured responses in multiple views to a library of pre-computed responses. From this, a likelihood of each library response matching the measured response is computed to form a map of the most likely defect type and, implicitly, the uncertainty of the characterisation. For larger defects, image based sizing can be used and we have developed a technique for fusing data from multi-view images to improve sizing accuracy. Papers on defect characterisation are in advanced states of preparation.
Exploitation Route We envisage some aspects of work being implemented almost immediately by two of the project partners in their own code, who already have a system capable of capturing the raw ultrasonic data to feed into data fusion algorithms. For wider uptake, we have implemented an initial version of the procedure into our own open-source array data capture and processing software BRAIN to enable others to try them out.
Sectors Aerospace, Defence and Marine,Digital/Communication/Information Technologies (including Software),Energy,Manufacturing, including Industrial Biotechology,Transport

 
Description Methods implemented in our open-source software, Brain, and expected to be used by partners in industrial trials.
First Year Of Impact 2019
Sector Aerospace, Defence and Marine,Energy,Manufacturing, including Industrial Biotechology
Impact Types Economic

 
Title A model for multi-view ultrasonic array inspection of small two-dimensional defects 
Description Supporting data for paper entitled "A model for multi-view ultrasonic array inspection of small two-dimensional defects" in IEEE Trans. Ferroelect., Ultrason. Freq. Contr., 2019. 
Type Of Material Database/Collection of data 
Year Produced 2019 
Provided To Others? Yes  
 
Title Experimental quantification of noise in linear ultrasonic imaging 
Description Supporting data for the paper entitled "Experimental quantification of noise in linear ultrasonic imaging" 
Type Of Material Database/Collection of data 
Year Produced 2018 
Provided To Others? Yes  
 
Description BAE Subs 
Organisation BAE Systems
Country United Kingdom 
Sector Academic/University 
PI Contribution We provide regular updates to all partners through the project review meetings (twice per year), as well as implementing the techniques developed in the project into our open-source array inspection software, Brain.
Collaborator Contribution In addition to direct cash contributions, BAE support the project through attendance at project review meetings and have supplied datasets obtained from welds in small-bore pipework, obtained by them following our procedure.
Impact Papers, procedures and software.
Start Year 2016
 
Description EDF 
Organisation EDF Energy
Country United Kingdom 
Sector Private 
PI Contribution We provide regular updates to all partners through the project review meetings (twice per year), as well as implementing the techniques developed in the project into our open-source array inspection software, Brain.
Collaborator Contribution In addition to direct cash contributions, EDF support the project through attendance at project review meetings, access to samples and datasets.
Impact Various papers, procedures and software.
Start Year 2016
 
Description Hitachi 
Organisation Hitachi Europe Ltd
Country United Kingdom 
Sector Private 
PI Contribution We provide regular updates to all partners through the project review meetings (twice per year), as well as implementing the techniques developed in the project into our open-source array inspection software, Brain.
Collaborator Contribution In addition to direct cash contributions, Hitachi support the project through attendance at project review meetings and loaned us two stainless steel samples with stress-corrosion cracking. We have scanned these ultrasonically, Hitachi have X-rayed them and will shortly section them.
Impact Papers, procedures and software.
Start Year 2016
 
Description Wood 
Organisation Wood Group
Country United Kingdom 
Sector Private 
PI Contribution We provide regular updates to all partners through the project review meetings (twice per year), as well as implementing the techniques developed in the project into our open-source array inspection software, Brain.
Collaborator Contribution Wood support the project through attendance at project review meetings and providing access to samples and their ultrasonic scanning systems.
Impact Papers, procedures and software.
Start Year 2016
 
Title BRAIN 
Description BRAIN is a flexible ultrasonic array data capture and imaging software suite for NDT applications, based around the concept of Full Matrix Capture (FMC) of array data with imaging performed in post-processing. New imaging tools can be readily added to the core. BRAIN is written in Matlab and can be run either from within Matlab or as a standalone Matlab application. 
Type Of Technology Software 
Year Produced 2010 
Open Source License? Yes  
Impact Numerous collaborating companies are known to use BRAIN for trialling new inspection and imaging algorithms. These include Rolls-Royce, BAE System in the UK, Tenaris in Argentina and potentially many others. New array imaging algorithms developed under various programmes are added to BRAIN as they reach maturity. 
URL https://sourceforge.net/projects/bristol-brain/