Typically, recorded echoes from the scans are presented overlaid on a CAD drawing of the component in end, top or side elevations - respectively termed B-scan, C-scan and D-scan plots. In addition to visualisation methods, sophisticated calibration routines for range finding and attenuation may also be programmed in the scan computer. This type of system, however is relatively expensive, and is typically only used to generate comprehensive NDT data from welds in safety critical components, such as nuclear pressure vessels, where even very small flaws may prove significant. The drawback to automated scanning techniques, somewhat paradoxically, is the sheer volume of data which is produced - a scan with multiple probes will generate a series of scan plots which require involved manual interpretation. Furthermore, if a suspicious defect area is located which requires further investigation, the entire scan procedure may have to be repeated. However automatic scanning has another advantage over manual probe positioning in that it may enable 'on-line' weld scanning since high temperatures should not overly affect its use.
However there are problems associated with pulse echo testing of incomplete weld geometries. With components of peculiar geometries, even sophisticated three dimensional plots may not be sufficient to detect all defects - the characteristic 'sparseness' of a plot will be masked with large deflections from the component geometries, many of which may register over 100% FSH. These areas will serve to merge with defects and render the task of flaw detection very difficult. Flaw classification is likewise made more difficult. Fig (2) for example illustrates a similar scenario to that shown in Fig. 1 - only this time the weld is only 50% complete. Confusing signals will be prevalent due to the open faces of the unfused weld surfaces. A B-scan taken with a 60 degree probe of a 50% complete weld with a large defect. The large mass of signals from the open face of the plate and the corner of weld region serve to almost completely obscure the defect - in any case it is almost impossible to discern the actual extent of the defect area, even if it is actually detected. Other methods must therefore be sought to deal with these problems.
A schematic of an A-scan for the arrangement shown in Fig. 3 is given in Fig. 4. This shows the detection of four main signals - (1) the surface or lateral signal which travels along the surface of the component and has the shortest arrival time; (2) the top tip of the defect; (3) the bottom tip of the defect; and (4) the backwall echo, which has the longest transit time. Also present in the A-scan are reflections from mode converted signals which have a slower speed of propagation through the specimen and hence a longer transit time. Although mode converted signals are not usually examined in TOFD inspection, they can often duplicate the main body of the A-scan. If both tips of a defect can be resolved in the A-scan then the actual depth and 'through-wall' thickness of the flaw can be accurately calculated (Silk, 1977; Carter 1984). This was the main thrust behind the development of the technique, since other methods, such as standard pulse echo techniques could often locate but not size defects accurately.
TOFD has some drawbacks in that it is rare to achieve clear signals in the A-scans due to the nature of the weak diffracted signals, coupled with noise, and interference from signals derived from very small pores or non-uniformity in the weld region. However, one of the reasons why TOFD examination has attained widespread use is the associated data acquisition and presentation methodologies proposed very early in its development by Silk (1977). Each A-scan is digitised as it is collected by the receiver, to a resolution of 8 bits - the signal is not rectified and instead is mapped directly to the 256 possible grey levels. A series of A-scans, for a section of weld, is then presented in B or D-scan fashion as shown in the example scan in Fig. 5. In this example scan we can clearly see the lateral wave (at the top), and the mass of backwall echoes (at the bottom). Also present between these signals is a defect, D, around which are peculiar down curved arcs which are typical of TOFD signatures. The arcs are caused by the interaction of the defect with the wide beam spread when the probes are not exactly in line with the flaw - hence the signal becomes progressively stronger near the centre of the arc as the actual position of the defect reaches the centre of the ultrasonic beam. Since the signals associated with the presence of defects are often quite characteristic when shown in Silk's D-scan plot, the TOFD technique is also now extensively employed for defect detection as well as sizing (Silk, 1984), and has also recently been used to completely replace other NDT methods (Verkooijen, 1995)
TOFD scans are commonly undertaken with automated scanning manipulators, and the obtained scans can be compared with pulse echo scans using the same apparatus. Currently there are several commercial digital ultrasonic such systems that can acquire data in this way (including the confusingly named Micro-pulse and Microplus systems) All of these systems now employ various data enhancement algorithms such as pseudo colouring, geometric stretching, and most notably the Synthetic Aperture Focusing Technique (SAFT) which can be used to eliminate beam spread effects and thus improve defect characterisation.
TOFD testing is one of the most promising NDT techniques for on-line weld inspection since the method is virtually immune to many of the problems associated with UT testing of welds with incomplete geometries. This is because the probe separation and angle can be altered to allow for almost full weld region coverage whilst avoiding signals from either the open parent metal faces or the weld/metal apex. Furthermore, if the interpretation task can be automated then in-line process control of the welding operation is a foreseeable entity.
In addition TOFD scans are, by their nature, continuous and contain all information recorded in the individual A-scans - this serves to ease the detection of defects by a human observer since he is able to employ intuitive two dimensional reasoning to locate the extent of a defect area. If interpretation is to be automated then the manual interpretation procedure must be mimicked by computer algorithms. The use of image processing and pattern recognition technologies in automatic ultrasonic inspection are not new (Windsor, 1995) though the task has. in most cases, proved extremely difficult as is indicated by the plethora of different methods which have been described in the literature (Chen, 1994; McNab and Dunlop, 1995). The majority of such works have been applied to pulse echo testing and mostly for defect classification with relatively little work directed at defect detection and TOFD analysis. An EC funded BRITE-EURAM project NDT Methods for Flaw Detection During Welding is currently addressing a number of automatic defect detection methods - including methods for TOFD inspection.
The approach adopted for the analysis of TOFD images is to first detect 'interesting' areas of the scan - i.e. flaws and component signals. The initial problem in locating these signals however lies in the removal of background signals due to noise and small insignificant artefacts, such as air pores. General segmentation algorithms such as edge detection, or thresholding cannot cope with the varying grey intensities typical of TOFD scans. A more intelligent method, using texture analysis and a neural network has therefore been developed - see Fig. 6. The effect of processing with this method is illustrated in Fig. 7b - the raw scan shown in Fig. 7a is segmented pixel by pixel the neural network into three classes of interest - with high interest areas labelled white. Component signals (i.e. backwall and lateral waves) are then removed by rule based methods to reveal defect areas. These can be highlighted to aid manual inspection, as is shown in Fig. 7c, or be further analysed to give an indication of the defect type, size and significance. These methods have proven successful in automatically identifying most representative defect types in welds of various stages of completion and at high temperatures.
Pulse echo testing is the most commonly used ultrasonic method in weld NDT and can be mechanised to produce high quality three dimensional information on a scanned component. However, the interpretation of pulse echo scans is frequently fraught with difficulty, particularly so when peculiar geometries (such as incomplete welds) are present, or if the component is very large. Therefore, an alternative techniques, such as TOFD, are of interest for peculiar or demanding applications such as on-line weld inspection.
The BRITE-EURAM project NDT Methods for Flaw Detection During Welding * is currently addressing methods for the implementation of on-line automatic TOFD inspection. Novel pattern recognition methods have been developed to distinguish between defect and non-defect signals in the highly varying background of the scan. Once this is achieved then the extent and position of the defect may be automatically measured to give an indication of its severity. These techniques are fast, robust and capable of detecting defects in welds of various stages of completion and at high temperatures.