Developments in Ultrasonic Phased Array Inspection I

Size: px

Start display at page:

Download "Developments in Ultrasonic Phased Array Inspection I"

Harvey Hubert Bridges
5 years ago
Views:

1 Developments in Ultrasonic Phased Array Inspection I Automatic Image Correction for Flexible Ultrasonic Phased Array Inspection A.J. Hunter, B.W. Drinkwater, P.D. Wilcox Department of Mechanical Engineering, University of Bristol, United Kingdom J. Russell Rolls-Royce Nuclear, United Kingdom ABSTRACT The ultrasonic inspection of components with irregular surface geometry is challenging. Flexible arrays are available to improve the inspection of this type of component but inspection performance is adversely affected by errors in the assumed or measured position of each element within the flexible array. This can result in image aberration and reduced defect detection and characterisation performance. A novel autofocus algorithm has been developed specifically for ultrasonic nondestructive testing and evaluation (NDE). This autofocus algorithm is used for estimating and correcting errors in the assumed element positions for a flexible ultrasonic array coupled to a specimen with an unknown surface profile. Experimental results are provided using a prototype flexible array to inspect a component containing a realistic, undressed weld cap. The results demonstrate that the algorithm is capable of generating autofocused imagery that is comparable in quality to benchmark imagery generated using accurately known imaging parameters. INTRODUCTION The inspection of components with complex surface geometry is an issue within the nuclear industry. When inspecting this type of component it is not always possible to achieve 100% test coverage using conventional ultrasonic inspection techniques. In some situations this type of inspection can be addressed by mechanically removing the weld cap from the component. However, weld cap removal is an expensive and time-consuming task which can also impact the structural integrity of the component. It is therefore desirable to inspect these components with the weld cap in place. A particular inspection challenge faced by Rolls-Royce Nuclear is the inspection of welded components that contain an undressed weld cap [1]. The specific component of interest is a stainless steel section of pipe-work with a wall thickness of greater than 50mm. It is postulated that defects can occur anywhere within the austenitic weld and Heat Affected Zone (HAZ) of the pipe-work. A membrane coupled phased array device produced by Rolls-Royce in association with Imperial College, London has been successfully deployed in this type of inspection and development work on this technology is ongoing [2]. However, Rolls-Royce is also working with researchers at University of Bristol to further develop alternative approaches to the inspection of this type of component. Work completed as part of this collaborative project using a flexible contact phased array device [3] is reported in this paper. In order to apply a broad range of imaging algorithms to the phased array inspection data it is desirable to acquire full matrix capture (FMC) data. This approach allows the use of advanced array imaging techniques, such as the total focusing method (TFM/ATFM) [4, 5]. The use of these algorithms requires accurate knowledge of the imaging parameters, i.e., the relative positions of the array elements and the specimen material properties.

2 Errors in these assumed parameters can cause image corruption, including blurring, distortion, and the introduction of artefacts, as demonstrated in Figure 1. Figure 1 - Demonstration of image performance loss due to de-focusing. The top row shows the unknown deviations from an assumed planar array and the bottom row shows the corresponding TFM imagery of 5 point scatterers The work discussed in this paper will consider techniques to correct imaging parameters for array element positional errors due to irregular surface geometries. However, work is also ongoing to improve the inspection of components with anisotropic material properties, such as austenitic welds [6]. Errors in the assumed element positions can cause image corruption, including blurring, distortion, and the introduction of artefacts. This corruption adversely affects defect detection and characterisation performance and, to avoid the problem, the imaging parameters must be measured accurately. When using this type of contact array imaging parameters can be measured explicitly, e.g., using position sensors to measure the array element positions. However, the additional hardware increases the cost of inspection and does not necessarily provide the required accuracy. Alternatively, the errors in the assumed imaging parameters can be estimated from the collected data implicitly, rather than by explicit measurement. This technique is termed autofocus. In practice, crude estimates of the imaging parameters should be known a priori, e.g., the surface form is known from CAD drawings of the specimen, and autofocus will be used to fine-tune these estimates. In order to develop the flexible phased array autofocus inspection capability, the inspection of a flat plate, non-welded test-piece has been completed. Weld metal was deposited onto the surface of the flat plate to produce an undressed weld cap representative of that found in the target application. Photographs of the test-pieces and the flexible phased array used this testing is provided in Figure 2. The flexible contact phased array transducer used in this testing was produced by the Commissariat à l Energie Atomique (the French atomic energy commission) (CEA List, Saclay, France). The transducer contains 30 elements, with a centre frequency of 2MHz and element pitch of 1.4 mm. The profilometers indicated in Figure 2(b) can be used to measure the approximate position of a group of elements within the array but it is not possible to measure individual element positions. The surface geometry of the component under test abruptly changes between the parent plate and the weld cap.

(a) Flexible Contact Phased Array Transducer Flat plate, nonwelded Test-piece (b) Profilometers Variable Coupling Weld cap Figure 2 - Photographs of the flat plate, non-welded test-piece and the

3 The conformability of the flexible contact array is limited and this leads to the variable coupling between the array and the component. The imaging performance that can be achieved with the flexible array is also therefore limited. (a) Flexible Contact Phased Array Transducer Flat plate, nonwelded Test-piece (b) Profilometers Variable Coupling Weld cap Figure 2 - Photographs of the flat plate, non-welded test-piece and the contact flexible ultrasonic phased array AUTOFOCUS ALGORITHMS Autofocus algorithms have been developed for a number of different imaging applications. These include synthetic aperture radar (SAR) and synthetic aperture sonar (SAS), where defocusing is caused by unknown deviations from an assumed trajectory and medium fluctuations [7]; groundbased astronomy, where defocusing is caused by atmospheric turbulence [8]; medical ultrasound, where defocusing is caused by unknown variations in the properties of human tissue and movement of the patient [9]; and seismic imaging, where defocusing is caused by unknown variations in the properties of the Earth s subsurface [10]. It was found that most of these algorithms were not directly applicable to the NDT problem, with the exception of the sharpness optimisation technique. The principle of sharpness opimisation is to maximise a sharpness metric with respect to the imaging parameters. The most common sharpness metric is the image contrast. The resulting algorithm maximum contrast autofocus and its application to NDT have been described previously [11]. Although it was demonstrated to provide improved image quality, it was also shown to have some practical limitations that restrict its use in NDE; these include over-sharpening, non-convergence, and slow speed of operation. This paper reviews a model-based autofocus algorithm designed specifically for NDE that addresses the limitations of the maximum contrast technique. The relevance of this algorithm to data acquired during the inspection of the target application component shown in Figure 2 is also reviewed. An advantage of the NDT autofocus application is that, generally, a great deal is known about geometric features in the specimen and the corresponding features in the imagery.

4 Furthermore, it is possible (and often straightforward) to define scattering models for typical features, e.g., point and planar reflectors. The model-based autofocus algorithm was developed based on this observation and it has been shown to address many of the short-comings of the previously proposed maximum contrast technique. The principle of model-based autofocus is simple: to minimise the errors between the echoes in the experimentally collected data and those expected from known features in the specimen, with respect to the imaging parameters. A minimum error indicates consistency between the experiment and the models and, therefore, the corresponding imaging parameters should provide a good estimate of the true values. Of course, the validity of this assertion is dependent on the accuracy of the assumed models. It is desirable to employ simple models for speed of operation. However, this must be balanced to ensure that the physics are described adequately. In principle, it is possible to estimate any unknown imaging parameters via any reasonable choice of error metric, minimisation procedure, and models for the array and feature geometries, propagation, and scattering. An overview of the concept is illustrated in Figure 3. In the case of flexible array imaging, the unknown imaging parameters are the relative positions of the array elements. In most practical situations, the imagery is not too severely distorted due the uncertainties of the element positions, such that it is possible to isolate and characterise known features in the corrupted image, e.g., the back-face reflection. By reversing the imaging algorithm, one can then isolate these features in the echo data. The arrival times of the echoes for each transmitter and receiver pair can be obtained from this isolated echo data and these values can be compared with the expected values from a known model for the feature. Least-squares minimisation of the errors between the experimental and modelled values provides the means of estimating the unknown imaging parameters. The variance of each estimated element position can be used to determine the suitability of the corresponding time traces for use in the imaging algorithm. Elements with large positional uncertainties (e.g., > λ 10 ) cannot be relied upon and, therefore, should be excluded during image formation. The exclusion of elements causes image resolution loss. However, this is preferable to the introduction of image artefacts due to element position errors. The full theory behind the model-based autofocus algorithm is provided by Hunter et al [12]. This theory is also demonstrated via implementation of the technique using the flexible contact array to inspect a component with a sinusoidal surface profile containing a series of side drilled holes. Figure 3 - Overview of the model-based autofocus approach. The dashed arrow indicates possible feedback of the estimated parameters and iteration

5 EXPERIMENTAL RESULTS The stainless steel flat plate, non-welded test-piece used in this testing contained a range of artificial defects and the results obtained from the inspection of one of these defects is discussed below. The defect of interest is a 4 x 8 mm semi-elliptical flaw, produced using electro-discharge machining (EDM). It is a surface breaking simulated lack of sidewall fusion defect in the root region and has a gape of approximately 300µm. A schematic diagram of the defect location and orientation within the test-piece is provided in Figure 4. Testing was completed with the array positioned with approximately 25%, 50%, 75%, and 100% of the elements over the weld and repeated on both sides. The array was coupled to the surface of the test-piece a water based coupling gel and sufficient pressure was applied to ensure good contact between the array and the irregular weld surface. A graphical user interface (GUI) to the autofocus software was developed to facilitate the industrial application of the model-based autofocus algorithm. The GUI (as well as the autofocus algorithm) has been implemented in Matlab and can be run from a standard laptop computer. A simple four stage process has been developed to acquire, process, image and store the inspection data, this process is outlined below, the four stages are also shown in the screen captures provided in Figure5: 1) Acquire the FMC data from the inspection: an online data display function is provided to allow coupling and data quality to be assessed prior to subsequent processing. 2) Image the inspection data: this is currently limited to the TFM imaging algorithm but additional imaging algorithms could be added as required. 3) Autofocus the TFM image: the user isolates the corrupted feature of interest in the TFM image and specifies the type of feature to be modelled (point feature, backwall etc). The inspection data is then autofocused and corrected image displayed. This step can be repeated with different isolation windows until the user is content with the quality of the resultant image. 4) Data storage: the raw data and the autofocused image are saved for future analysis. Two representative sets of autofocus results and imagery are shown in Figure 6 with: 1) 50% and 2) 100% of the elements over the weld. These results are representative of the results obtained from all of the experiments. Figure 6(a) and (b) show the unfocused images, Figure 6(c) and (d) show the autofocused images, and Figure 6(e) and (f) show the estimated array profiles. The image magnitudes are plotted in db relative to the maximum value in the autofocused image. The array profiles were estimated using the model-based autofocus algorithm, where the back wall was modelled as an infinite planar reflector, time-of-arrival estimates with correlation coefficients less than 0.95 were excluded from the autofocus procedure, and array element estimates with uncertainties greater than λ/10 were excluded from the TFM processing. Figure 4 - Schematic diagram of the defect contained within the non-welded test-piece

the raw FMC data and the final autofocused image. The unfocused images show significant distortion of the back wall reflection.

6 (a) (b) (c) (d) Figure 5 - Screen captures from the Autofocus GUI for flexible array imaging showing the inspection of five side drilled holes. (a) Acquisition of the FMC data from the inspection, (b) TFM image of the inspection data assuming plane geometry, (c) Autofocusing the TFM image to compensate for the irregular surface, (d) Storing the raw FMC data and the final autofocused image. The unfocused images show significant distortion of the back wall reflection. Application of autofocus yielded considerable improvement in image quality, with the back wall reflection appearing well-focused and attaining a 5-10 db increase in magnitude. Furthermore, the estimated array profiles matched closely the profiles observed on the specimen and were consistent between data sets. Despite successful autofocusing, no obvious reflections from the defects were observed in any of the data sets. However, shadows were observed on the back wall due to the presence of the defects preventing the energy from being reflected back towards the array. These shadows were seen consistently in all of the data sets collected away from the weld centre (i.e., where the defects had a larger apparent surface area). An example of the back wall shadow can be seen in Figure 6(c). It could be argued that some features in the imagery are reflections from the defects (e.g., in Figure 6(d)). However, they are so close to the noise floor that these observations should be considered highly speculative.

7 (a) (b) (c) (d) (e) (f) Figure 6 - Autofocus results with (a,c,e) approx. 50% of the array elements over the weld cap on the left-hand-side and (b,d,f) 100% of the elements over the weld cap: (a,b) unfocused TFM images; (c,d) autofocused TFM images using a plane model for the back wall (feature isolation windows are indicated by the dashed lines); and (e,f) estimated array element positions the crosses denote elements with positional uncertainty greater than λ/10. CONCLUSION The model-based autofocus algorithm has been designed specifically for NDT. The model-based algorithm can provide an estimate of the errors in the estimated imaging parameters, i.e., the estimated array element positions. The element position error estimates can be used to quantify the confidence in the autofocus result and to exclude elements with uncertain positions during the image formation process.

8 The results from experimental testing demonstrate that the flexible array model-based autofocus algorithm can be used on industrially relevant components. The algorithm was shown to estimate the array profiles successfully and to generate well-focused imagery. Although reflections from the defect were not observed, shadows on the back wall were observed, indicating the presence of a defect. Future analysis of these data could use the estimated array profiles to investigate other imaging methods such as the ATFM [2] which could be used to favour certain angles of reflection to enhance the defect response. ACKNOWLEDGEMENTS This work is supported by Rolls-Royce Nuclear and the UK Ministry of Defence. REFERENCES 1) J. Russell, R. Long, P. Cawley, N. Habgood, Inspection of Components with Irregular Surfaces Using a Conformable Ultrasonic Phased Array In: Thompson, D.O. & Chimenti, D.E. (eds.) Review of Progress in QNDE, American Institute of Physics: Melville, New York, 2008, 28, pp ) J. Russell, R. Long, P. Cawley, Development of a Twin Crystal Membrane Coupled Conformable Phased Array for the Inspection of Austenitic Welds, Proc. 8 th Intl. Conf, on NDE in relation to Structural Integrity for Nuclear and Pressurised Components, Berlin, In press. 3) O. Roy, S. Mahaut, O. Casula, Control of the ultrasonic beam transmitted through an irregular profile using a smart flexible transducer: modelling and application, Ultrasonics, p ) C. Holmes, B.W. Drinkwater, and P.D. Wilcox, Post-Processing of the Full Matrix of Ultrasonic Transmit-Receive Array Data for Non-Evaluation. NDT&E Int., : p ) Long, R., Russell, J., Cawley, P., Habgood, N. Ultrasonic Phased Array Inspection of Flaws on Weld Fusion Faces using Full Matrix Capture. In: Thompson, D.O. & Chimenti, D.E. (eds.) Review of Progress in QNDE, American Institute of Physics: Melville, New York, 2008, 28, pp ) A. J. Hunter, B. W. Drinkwater, and P. D. Wilcox, A Study into the Effects of an Austenitic Weld on Ultrasonic Array Imaging Performance, In: Thompson, D.O. & Chimenti, D.E. (eds.) Review of Progress in QNDE, American Institute of Physics: Melville, New York, 2010, In press. 7) J. R. Fienup, Phase Error Correction by Shear Averaging, in Signal Recovery and Synthesis III. North Falmouth MA, USA: Optical Society of America, 1989, pp ) K. T. Knox and B. J. Thompson, Recovery of images from atmospherically degraded shortexposure photographs, Astrophys. J., 1974 vol.193, pp. 9) S. W. Flax and M. O Donnell, Phase-aberration correction using signals from point reflectors and diffuse scatterers: Basic principles, IEEE Trans. Ultrason., Ferroelectr., Freq. Control, 1988, vol. 35, no. 6, pp ) J. A. Hileman, P. Embree, and J. C. Pflueger, Automated static corrections, Geophysical Prospecting, 1968, vol. 16, pp ) A.J. Hunter, B.W. Drinkwater, and P.D. Wilcox, Autofocusing Ultrasonic Imagery for Non- Destructive Testing and Evaluation of Specimens with Complicated Geometries. NDT&E Int., (2): p ) A.J. Hunter, B.W. Drinkwater, and P.D. Wilcox, Least-Squares Estimation of Imaging Parameters for an Ultrasonic Array using Known Geometric Image Features. IEEE Trans. Ultrason., Ferroelect., and Freq. Contr., 2010 (submitted for review)

Developments in Ultrasonic Phased Array Inspection III

Developments in Ultrasonic Phased Array Inspection III Improved Phased Array Mode Conversion Inspections Using Variable Split Aperture Processing R. ong, P. Cawley, Imperial College, United Kingdom J.