Introduction

The BRITE-EURAM Project 5907 "NDT Methods for Flaw Detection During Welding" [ Acknowledgements] is reaching its final phase. For the past 4 years a consortium of European partners has been investigating the possibilities of performing NDT between weld runs as an on-line operation. The two methods which have shown the most promise for this application are radioscopy (or real time radiography, RTR) and Time of Flight Diffraction (TOFD) ultrasonic testing. This article describes recent developments within the project aimed at online, automated weld inspection at elevated temperatures using TOFD. Firstly a description, with scan examples, of the initial high temperature experiments is given. Secondly, a description of the automatic defect detection software developed at the University of Surrey will be made and it will be shown how the software can deal with scans of very low SNR. Finally a brief outline of the final TOFD trials for the project will be given.

Table of contents

• Introduction
• High temperature TOFD scanning
• Automatic defect detection in TOFD scans
• September trials
• References
• Acknowledgements

High temperature TOFD scanning

TOFD testing has been gaining in profile over the last three or four years with much interest focussing on whether or not it can be used to replace more established NDT methods (e.g. Verkooijen (1995) [1]. So far, this debate has been made rather open ended due to the lack of internationally accepted standards for the technique. There is no doubt however of the potential use of TOFD in many applications where fast, reliable and inexpensive testing is required. The method is ideally suited to semi or fully automated data acquisition, and processing, and in addition, the scans are relatively straightforward to interpret - at least where defect detection is concerned. The method has been recognised by the BRITE 5907 consortium as having great potential for implementing ultrasonic inspection between weld runs. The problems to be encountered under such inspection conditions include constrained inspection times (so that no reheating of a workpiece is required), incomplete weld geoemetries, and elevated temperatures. The problem of fast inspection times is been addressed by using TOFD in combination with automated processing methodologies (see next section), whilst the some of the problems associated with incomplete geometries are discussed in Lawson (1996) [2]. This particular article introduces the effects of using TOFD at elevated temperatures.

A high temperature rig was designed and commissioned by Mitsui Babcock Ltd, the overall coordinators for the project. The rig was used to raise test plates to elevated temeperatures of up to 250 degrees C. Trials were carried out on thin (25mm) TIG welded plates and thicker (80mm) submerged arc welded plates containing defects of known type and location. TOFD scans were made of various plates at ambient temperture, 50 degrees C, 100 degrees C, 150 degrees C, 200 degrees C and 250 degrees C. High temperature probes were used to take the scans - however at temperatures above 50 degrees C the SNR of the scan was found to decrease sharply with the content of the scan becoming unreadable at around 150 degrees C. The probes were found to quickly degrade in performance after becoming exposed to high temperatures for more than a minute or two. Alternative very high temperature probes were sought and were found to give much better results even at sustained high temperatures. Three example scans (of the same testpiece) acquired with these probes are shown in Figure (1).

Fig 1:TOFD scans at increasing temperatures. (a) shows a scan of a submerged arc weld containing lack of inter-run penetration taken at ambient temperature. (b) shows a scan of the same weld taken at 150 degrees C. (c) the same weld scanned at 250 degrees C.

The scans shown in Figure (1) were made at fixed gain (110dB) on a Zipscan system. If the gain is increased then, as would be expected, both noise and component/defect signals are enhanced causing false alarm problems for the automated defect detection software (discussed in the following section). The SNR decrease in the three scans at progressively increasing temperatures is quite visually apparent though both defect signals and component signals can still be resolved. Problems due to decoupling of the probes from the parent plate may prove significant for automatic scanning of hot plates - for the majority of trials conducted so far the parent plate has been submerged and heated in an oil bath up to the required temperature. The oil therefore also acts as an excellent couplant though this approach is clearly not applicable to a real situation where a prototype couplant injection system will be used (see below).

Automatic defect detection in TOFD scans

The University of Surrey has been developing software techniques for the automatic interpretation of ultrasonic data, with particular emphasis on TOFD, for several years. The main elements of the TOFD interpretation software, are:
1. an area based segmentation method which will differentiate between areas of interest (i.e. component and defect signals) and areas of non-interest (background/noise). This process was originally centred on a neural network (described in Lawson and Parker (1996) [4] which has recently been replaced with a more flexible operator based on the calculation of the standard deviation in grey levels of local scan regions. This method has been found to be more tuneable when dealing with scans of varying SNR - such as those encountered when varying the temperature at the time of acquisition.

2. an intelligent methodology for classifying defect and component signals which can also seperate merged signal areas from different sources (i.e. defects merged with component echoes), and

3. a post-processing method of object analysis that can provide a description of each suspect defect region.

Note that any implementation of an automatic defect classification scheme has not been attempted as it is considered unrealistic to indentify defects from TOFD scans - even using manual interpretation. Indeed, Dijkstra, de Raad, and Bouma (1997) [3] suggest that information on defect type is limited to whether a defect has 'measurable height' (i.e. through wall thickness) or not.

The software was originally developed as a research tool using UNIX and image processing workstations. However it has always been the aim to port the software to a system more easily integrated with current PC based data acqusisition systems such as AEA's Microplus and the Rolls Royce Micropulse. To meet these aims, a graphical interface (GUI) to the TOFD processing software has recently been developed using Microsoft Visual C++ to run on a Pentium PC under either Windows 95 or Windows NT. A screenshot of the software, termed AutoTOFD, showing the labelling of automatically detected defects and component signals in a scan acquired at ambient temperature, is shown in Figure (2). Currently, the software is able to import scan files from either the AEA Microplus system or from the older Zipscan system. A series of pull down menus has been designed to allow the operator to set a list of processing parameters (often a set of default parameters are supplied) and to interactively process a scan from the segmentation stage to the defect analysis stage. Once an operator is satisfied with a particular set of parameters then the software can be instructed to run automatically to detect defects in subsequent scans. Scan parameters and testpiece descriptions can be imported from the header of the scan-file and used to perform time of flight correction calculations and to give a brief description of each defect that is located.

Fig. 2: Screenshot from automatic TOFD interpretation software 'UoS AutoTOFD' Fig. 3: Defects detected by automatic software in scans shown in Fig.1

The results of processing on ambient temperture scans using the modules of the AutoTOFD software has been illustrated in previous publications (e.g. Lawson and Parker (1996) [4], Lawson (1996) [2]. Recent modifications to the segmentation stage of the processing allow for tuning of the software when confronted with very low, or very high, SNR's - such as those encountered when scanning at varying temperatures. As an example, the defect regions detected by the software in the scans shown in Figure (1) are shown in Figure (3), above. In all three cases the defect has been clearly detected with no false alarms generated.

Currently, a formal evaluation of the software for defect detection at high temperatures (and hence low SNR's) has not been made, due to frequent interchanging of different probe types (see above), and other recent parameter variations in acquiring test scans. Such an evaluation will however be possible after the completion of the final TOFD trials which will be held in mid September (see below). It is predicted, given results so far, that only slight reductions in defect detection rates will be encountered at higher temperatures. Silk (1996) also reported an increase in false alarm in his study based on the effects of noise induced in manually interpreted simulated scans. This is also likely in this case since the main cause of false alarms detected by the AutoTOFD software is due to the break up of component (backwall and lateral wave) signals which is often likely to occur at high temperatures. False alarms due to pores or general background noise, conversely, seem more likley to occur at low temperatures when diffracted signal strength is at its highest.

September trials

The software described above will be used in the final ultrasonic testing trials associated with the current BRITE-EURAM project 5907. These trials will be held at the premises of welding company Nordon & CIE in Nancy, France in mid-September and will feature on-line TOFD inspection at temperatures predicted to reach approximately 250 degrees C near to the ultrasonic probes. High temperature probes from the equipment manufacturer RTD will be employed, which will be housed in a special manipulator designed and constructed by Mitsui Babcock Ltd. The manipulator has been designed to inject couplant under the probe housing which is then subsequently removed from the surface by a pump device. This should ensure that no couplant reaches the weld region (a situation which may induce defects !). An AEA Micoplus system will be used for data acquisition which will be linked (probably via ethernet at this stage) to a Windows NT PC running the Uos AutoTOFD software. It is currently not practical to incorporate the AutoTOFD software onto the DOS based Microplus PC system.

The trials will demonstrate the techniques and methods developed during the four year study to undertake NDT between weld runs. A number of welded test plates will be prepared - one set will be of TIG welded 25mm thick steel, the other submerged arc welded 80 mm thick steel. Defects will be deliberately implanted into the testpieces by Nordon welders at various stages of weld completion. These defects will include tungsten inclusions (for TIG welds), root cracks, porosities, lack of fusion, slag inclusions (for SA welds) and lack of penetration. Ultrasonic TOFD scans will be made of the welds at 25%, 50% and 75% stages of weld completion. These scans will be both manually (by staff from Mitsui Babcock and the Institut de Soudure) and automatically (using the UoS AutoTOFD software) interpreted. After the trials the welded testpieces will be sectioned (destructively tested) which will enable a full objective assessment of the performance of both the TOFD scanning procedure and also the automatic software.

The results of the trials will be published in the project's Final Report which will be made publicly available later in the year.

References

1. Verkooijen, J., "TOFD used to replace radiography", INSIGHT, vol. 37(6), pp. 433-435, June 1995. Abstract
2. Lawson, Shaun, "Ultrasonic testing and image processing for in-progress weld inspection", UTonline Journal 'http://www.ndt.net/article/shaun/shaun.htm', April 1996.
3. Dijkstra, F.H., de Raad, J.A, and Bouma, T., "TOFD and acceptance criteria: a perfect team", INSIGHT, vol. 39(4), pp. 268-270, April 1997. Abstract
4. Lawson, S. W. and Parker, G. A., "Automatic detection of defects in industrial ultrasound images using a neural network", Proceedings of SPIE, vol 2786, 1996, pp. 37-47. Abstract
5. Silk, M.G., "Estimates of the probablity of detection of flaws in TOFD data with varying levels of noise", INSIGHT , vol. 38(1), 1996, pp 31-36. Abstract

Acknowledgements
*This work is funded by the European Commission under the BRITE-EURAM II programme (project no. 5907).
The BRITE-EURAM project NDT Methods for Flaw Detection During Welding. See also the General Information of the project

Thanks are also due to Dick Bell of AEA Technology for his help in the decoding of Microplus file headers.

Author

Shaun Lawson
Shaun Lawson has worked in the field of computer vision in automatic inspection for 6 years. During this time he has worked both in industry (for Royal Ordnance plc, investigating the automation of industrial radiographic inspection cycles) and in academia. In 1996 he was awarded a PhD for work in automatic defect detection in radioscopic and ultrasonic images. He is currently employed as a research fellow at the University of Surrey in the UK.
E-mail: s.lawson@surrey.ac.uk
WWW: http://robots.surrey.ac.uk/People/mes3sl/mes3sl.html
Mechatronic Systems and Robotics Research Group
School of Mechanical and Materials Engineering
University of Surrey
Guildford Surrey GU2 5XH (UK)

For more information see: TOFD in UTonline 09/97