The University of Manchester Manchester, M13 9PL, UK
TWI Technology Centre (Wales) Port Talbot, SA13 1SB, UK
Anthony Peyton and Patricia Scully
The University of Manchester Manchester, M13 9PL, UK
TWI Technology Centre (Wales) Port Talbot, SA13 1SB, UK
WMG, The University of Warwick Coventry, CV4 7AL, UK
Presented at NDT 2016. The 55th annual conference of the British Institute of Non-Destructive Testing. 12-14 Sept. 2016, Nottingham, UK.
The total focussing method has gained popularity in academic and industrial circles, in part due to its ability to generate fully focused and high resolution images, as well as overcome the limited depth of field associated with conventional and phased array inspections. Despite this the inspection of coarse grained and/or highly scattering material proves problematic. In this work two different coherence weighting approaches are described: one based on signal amplitude and another on the phase of the signal. Both approaches are applied to simulated data where both are shown to reduce the effect of grain scatter induced noise in the final cross-sectional image. In addition the sensitivity factor is altered and found to result in a defect indication level 6 dB above all structural noise in the case of the amplitude based coherence approach. The phased based approach resulted in erroneous identification. Both methods were then employed on data obtained from a coarse grained low density, high aluminium, steel cast sample with an artificial defect in the form of a 2.5 mm diameter side drilled hole. Both approaches resulted in different cross-sectional images, however both also allowed the defect signal to be resolved at a level of more than 5 dB above the local structural noise, an improvement over the standard total focussing method approach where the defect signal could not be resolved for the setup used in this study. It is recommended that coherence weighting should be used to compliment the total focussing method when inspecting coarse grained materials.
Coarse grained materials present a challenge to the field of non-destructive testing (NDT)(1,2). Often chosen for their high ductility, corrosion resistance and high toughness, coarse grained materials such as austenitic stainless steels can be found in the petro-chemical, rail and nuclear industries. These materials are difficult to inspect through ultrasonic testing because of the tendency for the ultrasonic waves to be scattered by the grains of the material, making defects difficult to distinguish from structural noise. The ability to monitor structural health during a component’s lifetime reliably or the quality assurance during joining processes is crucial from a safety stand point, and thus the ability to perform NDT on coarse grained materials reliably is an active area of research of importance to several industries including rail(3) and nuclear power generation(4).
The use of linear array probes has become widespread in industry for the inspection of metallic components over the last decade and offers the ability to decrease inspection time, whilst providing enhanced imaging capabilities over conventional inspections using single element probes. Results obtained with linear array probes may be more intuitive to interpret(5).
Sound-velocity inhomogeneities, which decrease the coherence of the received signals, are inherent in coarse grained materials causing a degradation of both spatial and contrast resolutions due to focusing errors. The effect of these in- homogeneities can manifest as imaging artefacts which may be misinterpreted as defects in the material under test. Coherence weighting techniques offer the possibility of reducing the effect of coarse grains on the final ultrasonic images. In this work the amplitude based approach described by Mallart and Fink(6) and later dubbed the Coherence Factor (CF)(7) and the phase based approach developed by Camacho et al.(8,9) will be investigated. These coherence weighting variations have been selected due to their computational efficiency. The Total Focussing Method (TFM) can be implemented in real time(10), and as such a de-noising solution which retains a real time inspection speed is desirable.
The coherence weighting algorithms used in this work produce a weighting map which a cross-sectional image is then multiplied by to produce a final cross sectional image. The weighting value is directly proportional to the level of signal coherence across the elements, and thus represents the focusing quality of the inspection.
1.1 Total Focussing Method
TFM was first described for use in NDT by Holmes et al.(11); it is a synthetic aperture technique of particular interest to academia and industry alike. By using a delay and sum beam-forming operation TFM is able to produce a high resolution
fully focused cross sectional image. TFM makes use of the full matrix of data, acquired by transmitting and receiving on each pair combination of elements in a linear array. Acquiring data in this manner is known as Full Matrix Capture (FMC).
A cross-sectional image is generated via TFM when the FMC data is synthetically focused at every pixel within the region of interest, with the contribution from each A-scan summed. The pixel intensity, ITF M , is described via eq. (1) where tx and rx represent the transmit and receive elements respectively, N denotes the number of elements in the linear array, v is the speed of the ultrasonic wave, the co-ordinates are denoted by x for lateral position and z for vertical position and finally htx,rx is the Hilbert transform(11).
TFM has previously been described as the “Gold Standard” of array based acoustic imaging(12) and work continues to improve the suitability of TFM to industrial applications; for example by developing a calibration procedure(13)or for dealing with complex geometries(14). Developments in the laboratory to adapt TFM to other NDT techniques such as laser based inspections are also an active area of research(15,16).
1.2 Coherence Weighting
The ratio of coherent intensity to total signal intensity can be used as an index of focusing quality, this was first described by Mallart and Fink(6) and later dubbed the Coherence Factor (CF)(7). The pixel intensity, ICF, is described via eq. (2) where i represents the A-scan number, n denotes the total number of A-scans in the data (for FMC this is the square of the number of active array elements) and A(i,t−∆ti) is the amplitude of the A-scan with index value i after time-shifting. β is the weighting coefficient produced by the coherence weighting algorithm being used. We have introduced α as a sensitivity factor, unless otherwise stated for this work α = 1.
The Sign Coherence Factor (SCF) is a form of Phase Coherence Imaging (PCI) introduced by Camacho et al.(8). For a well focused image there will be little variation in the phase of echoes from a reflector once the signals have been time- shifted according to the delay-and-sum procedure. Thus by looking at the standard deviation of the polarity of the data contributing to the desired pixel, the intensity of said pixel, ISCF , can be defined as per eqs. (4) and (5).
Here N is the number of elements being used for inspection, α is the sensitivity factor, bi is a bitwise operator; if the sign bit of the received echo (A(i,t−∆ti)) is greater than or equal to 0 then bi = +1 otherwise bi = −1.
2. Experimental Set-up
This work employed the CIVA simulation package, developed at CEA (French Atomic Energy Commission). CIVA is a commonly used simulation tool within industry and has been extensively verified(17). Structural noise in coarse grained material is simulated in CIVA by assuming the crystallographic structure of the material is homogeneous and is populated by a set of randomly distributed scatterers(18). The spatial density of the scatterers is assumed to be uniform within a
Table 1: CIVA Structural Noise Parameters
|Reflector Density (Points / mm3)||σAmplitude|
Table 2: Parameters used for CIVA simulation
|Number of Elements
given density and the distribution of the amplitude of reflectivity of the scatterers is assumed to follow a zero mean Gaussian distribution. A coarse grained material may be described using only the scatter density and the standard deviation of reflectivity distribution (σAmplitude). These two parameters are not considered to be predictable, and instead must be found empirically.
In this work the values used for scatter density and σAmplitude are those found by Avramidis et al. to be representative of an austenitic steel rail sample(19) and are shown in table 1. All simulations are performed using the probe parameters detailed in table 2. The simulation scenario we shall look at is that of a specular reflector, represented in this case by a 3 mm side drilled hole (SDH) at a depth of 24 mm from the top surface of the sample. A diagram of the simulation set up is presented in fig. 1.
2.2 As-cast Coarse Grained Sample
A coarse grained low density, high aluminium, steel cast sample was produced and provided by WMG at the University of Warwick for inspection purposes. The sample contains a columnar grain structure with grain sizes in the millimetre range. Figure 2 is a micrograph of the as-cast steel block from the inspection surface. An artificial defect in the form of a 2.5 mm diameter SDH was inserted into the sample at a depth of 27 mm from the top surface. A 64 element probe manufactured by Vermon – specifications are presented in table 3 – was controlled by a Micropulse 5PA manufactured by Peak NDT, Derby, UK.
Table 3: Parameters used for inspection of coarse grained material
|Number of Elements
By simulating a 3 mm SDH in the material the performance of the imaging algorithms can be compared for an ideal and repeatable scenario. In this work we first explore the effect of applying CF and SCF to a TFM inspection and then consider altering the sensitivity parameter α.
Figure 3 shows three cross-sectional images of CIVA simulated data with a 20 dB dynamic range. Figure 3a shows a conventional TFM reconstruction. Note there are several areas of high intensity and the background noise is also appreciable. Figures 3b and 3c show TFM images which have undergone weighting via CF and SCF respectively. Note that the SCF processing has employed a 7-by-7 pixel moving average filter as recommended by Camacho et al. to mitigate discontinuities which manifest when the phases of the RF signal are grouped around ±π (9). Figures 3b and 3c both suppress the background noise, however several areas of high intensity still exist, resulting in the inability to identify the response from the side drilled hole.
3.1.1 Altering α
As shown in fig. 3 directly applying a coherence weighting factor is not always sufficient to resolve a defect indication from the structural noise. Camacho et al. describe a sensitivity factor for the SCF algorithm(8), this is represented by α in eq. (5), we have applied this sensitivity factor to the CF algorithm as per eq. (2).
Changing the value of α corresponds to varying the strictness of coherence needed for favourable weighting, with a high value of α being less tolerant to partially coherent signals than a lower value of α. The optimal value of α will depend on several factors including the material being inspected as well as the chosen method for inspection and is not absolute. Figure 4 shows several cross-sectional images with various values of α constructed from a common data set. As the value of α increases the indications due to structural noise decrease. With the condition α = 4 (fig. 4d) the majority of the structural noise is below the imaging threshold aside from a small cluster of indications at a depth of approximately 15 mm, however, by increasing the value of α to α = 6 the side drilled hole indication is over 6 dB above every other indication, making it easily resolvable
Figure 5 also shows the several cross-sectional images with various values of α constructed from the same common data set for the case of SCF. As the value of α increases the indications due to structural noise decrease. Once again, with the condition α = 4 (fig. 5d) the majority of the structural noise is below the imaging threshold aside from a small cluster of indications at a depth of approximately 15 mm. However, note that these indications are now the strongest on the image, and unlike the CF results shown in fig. 4 an increase in the value of α does not resolve the indication from the SDH. In fig. 5f the indication due to the SDH is 13 dB below the peak indication.
3.2 As-cast Coarse Grained Sample
Figure 6 shows a comparison of cross sectional images generated using a common data set acquired from the inspection of a coarse grained low density, high aluminium, steel cast sample as described in section 2.2. The cross sections are presented with a dynamic range of -20 dB. The effect of the CF and SCF weighting algorithms in figs. 6b and 6c respectively is significant when compared to the conventional TFM image fig. 6a. Figure 6a contains multiple indications of equally strong magnitude, making the indication from the SDH impossible to isolate from the structural noise. The effect of the structural noise is greatly reduced when CF is applied, with the indication from the SDH having a magnitude 5 dB above the second highest indication, which is emanating from structural noise located at a depth of 8 mm, and 8 dB above the local noise contributions. Figure 6c suffers more from shallow structural noise, with the indication due to the SDH being 5.5 dB below the highest indication in the image. Where SCF appears to perform beneficially is when the region of interest is more localised to the SDH indication, a valid approach in this scenario due to knowledge of the indication position a priori. Within a 10 mm by 10 mm region of the SDH the second highest indication value is 6 dB lower than the SDH response, with areas of indication due to structural noise being much more sparse when compared to fig. 6a.
This work has shown the benefits of coherence weighting for the inspection of coarse grained steel. The results in section 3.1.1 show how the sensitivity factor α can be varied in order to resolve indications from specular reflectors when using the CF algorithm, but care must be taken if SCF is used, as a false identification of defect position could be made, as shown in fig. 5. The effect of SCF weighting on the final cross-sectional image does differ to that of CF weighting and may be favourable depending on the inspection scenario, see fig. 6. Due to the scenario specific nature of each algorithm the authors would recommend that these weighting algorithms be implemented as an optional filter in inspection software so as to compliment conventional TFM inspections. The authors would also advise the inspection of a calibration sample with indications at known locations so that the performance of each imaging algorithm on the material in question may be known prior to the commencement of inspection.
The authors would like to thank Mark Sutcliffe at TWI Technology Centre (Wales) for his discussions on coherence weighting, as well as Neil Hollyhoke at WMG, Warwick University for the manufacture and loan of the coarse grained sample. The authors also wish to acknowledge support (including an EngD studentship to Benjamin Knight-Gregson) from the Engineering and Physical Sciences Research Council (grant number EP/G037426/1) as well as the National Skills Academy Nuclear.
- X. Edelmann, “The Practical Application of Ultrasonic Testing of Austenitic Weld Joints,” Materials Evaluation, vol. 37, no. 10, pp. 47–51, 1979.
- P. Ramuhalli, A. D. Cinson, S. L. Crawford, R. V. Harris, A. A. Diaz, and M. T. Anderson, “Evaluation of signal processing tools for improving phased array ultrasonic weld inspection,” Smart Sensor Phenomena, Technology, Networks, and Systems 2011, vol. 7982, 2011.
- M. Weston, B. Wright, D. Liaptsis, B. Knight-Gregson, and M. Rantatalo, “Reliable sub-surface ultrasonic inspection of cast manganese steel crossings,” in 6th IET Conference on Railway Condition Monitoring (RCM 2014), Sep. 2014, pp. 1–3.
- M. T. Anderson, S. E. Cumblidge, and S. R. Doctor, “Low frequency phased array techniques for crack detection in cast austenitic piping welds: A feasibility study,” Materials Evaluation, vol. 65, no. 1, pp. 55–61, Jan. 2007.
- B. W. Drinkwater and P. D. Wilcox, “Ultrasonic arrays for non-destructive evaluation: A review,” NDT & E International, vol. 39, no. 7, pp. 525–541, Oct. 2006.
- R. Mallart and M. Fink, “Adaptive focusing in scattering media through sound-speed inhomogeneities: The van Cittert Zernike approach and focusing criterion,” The Journal of the Acoustical Society of America, vol. 96, no. 6, pp. 3721–3732, Dec. 1994.
- K. Hollman, K. Rigby, and M. O’Donnell, “Coherence factor of speckle from a multi-row probe,” in 1999 IEEE Ultrasonics Symposium, 1999. Proceedings, vol. 2, 1999, pp. 1257–1260 vol.2.
- J. Camacho, M. Parrilla, and C. Fritsch, “Phase Coherence Imaging,” IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 56, no. 5, pp. 958–974, May 2009.
- J. Camacho and C. Fritsch, “Phase coherence imaging of grained materials,” IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 58, no. 5, pp. 1006–1015, May 2011.
- M. Sutcliffe, M. Weston, B. Dutton, P. Charlton, and K. Donne, “Real-time full matrix capture for ultrasonic non-destructive testing with acceleration of post-processing through graphic hardware,” Ndt & E International, vol. 51, no. 0, pp. 16–23, 2012.
- C. Holmes, B. Drinkwater, and P. Wilcox, “The post-processing of ultrasonic array data using the total focusing method,” Insight - Non-Destructive Testing and Condition Monitoring, vol. 46, no. 11, pp. 677–680, 2004.
- O. Oralkan, A. Ergun, J. Johnson, M. Karaman, U. Demirci, K. Kaviani, T. Lee, and B. Khuri-Yakub, “Capacitive micromachined ultrasonic transducers: next-generation arrays for acoustic imaging?” IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 49, no. 11, pp. 1596–1610, Nov. 2002.
- S. A. Mosey, M. Sutcliffe, M. Weston, B. Knight-Gregson, and D. Liaptsis, “Development and implementation of calibration process for Full Matrix Capture,” in Proceedings of the 53rd Annual Conference of The British Institute of Non-Destructive Testing. Manchester: BINDT, Sep. 2014.
- M. Sutcliffe, M. Weston, B. Dutton, P. Charlton, and K. Donne, “Real-time full matrix capture for ultrasonic non-destructive testing with acceleration of post-processing through graphic hardware,” NDT & E International, vol. 51, pp. 16–23, Oct. 2012.
- B. Knight-Gregson, S. A. Mosey, M. Weston, P. Scully, A. Peyton, and C. Davis, “Advanced Photoacoustic Imaging Using Full Matrix Capture Data Acquisition,” in Fourth International Symposium on Laser Ultrasonics and Advanced Sensing, Northwestern University. Norris University Center, Evanston, IL 60208, USA, Jun. 2015.
- T. Stratoudaki, M. Clark, and P. Wilcox, “Laser induced ultrasonic phased array using Full Matrix Capture Data Acquisition and Total Focusing Method,” in The 54th Annual Conference of the British Institute of Non-Destructive Testing, Telford, UK, Aug. 2015.
- S. Lonn´e, L. de Roumilly, L. Le Ber, S. Mahaut, and G. Cattiaux, “Experimental validation of CIVA ultrasonic simulations,” in International Conference on NDE in Nuclear Industry (ICNDE), 2006.
- S. Chatillon, C. Poidevin, N. Gengembre, and A. Lh´emery, “Simplified Modeling of Backscattered Noise and Attenuation Phenomena for Quantitative Performance Demonstration of UT Methods,” in AIP Conference Proceedings, vol. 657. AIP Publishing, Mar. 2003, pp. 93–100.
- S. Avramidis, M. Weston, and D. Liaptsis, “Ultrasonic modelling of coarse grained austenitic steel,” in Proceedings of the 53rd Annual Conference of The British Institute of Non-Destructive Testing, Manchester, UK, Sep. 2014.