Table of Contents ECNDT '98 Session: Nuclear Industry

Interpretation of Radiographs by Digital Image Processing T. Just*, W. Thale, R. Clausen, TÜV Nord, Hamburg *just@tuev-nord.de

1. Introduction

At in-service inspections in German nuclear power plants radiographic testing (RT) using X-ray and double wall technique has been applied at a larger scale after cracks had been detected at the media affected inner surface of austenitic stainless pipe welds. The pipings concerned were mainly thin-walled having wall thicknesses between 5 mm and 15 mm and inner diameters between 100 mm and 250 mm. The material is a Ti- or Nb-stabilized chromium-nickel steel. The defects are caused by intergranular stress corrosion cracking (IGSCC) /1/. The cracks start from the inner surface with nearly vertical orientation, either in ground material next to the weld or in many cases from shrinkage grooves (fig. 1). The defects which follow the heat affected zone are difficult to interprete in radiographs due to their low contrast and the presence of undercut and large density differences which occur as a result of mismatches and excess penetration.

fig. 1 Corrosion cracking in an austenitic weld

The conventional examination of radiographs in these welds revealed in many cases cracks or suspicion of cracks. Thereupon these welds were cut out and the inner surfaces were examined by penetrant testing. The areas with indications were examined by metallography. The result was that only 25% of welds cut out because of RT indications contained really cracks /2/. In several cases fine grooves in the root were found at the position of assumed crack indications. The radiographs were evaluated by us employing digital image processing and the RT indications were analyzed and compared with the results of destructive examination. For the interpretation of crack-like indications, criteria and a procedure have been developed which evaluate significant parameters (maximum contrast, halfwidth, length above a defined contrast threshold) from the digitized radiographs. These are detailed in the following chapters. In addition to weld testing the evaluation procedure can be applied to analyse linear indications in radiographs of cast e.g. hot tears, filamentary shrinkage.

2. Characteristics of the film digitizer

The scanning system has the following features /3/: CCD-film digitizer with 35 (m spatial resolution (pixel size) and a grey scale resolution of 12 bit (4096 grey values), fast digital image aquisition and enhancement software for on-line interpretation on the monitor (magni-fication of image details, contrast and brightness adjustment, image processing algorithms like filtering, measurement of indications' size, fast storing and retrieving of radiographs). Because of the large density differences which occur, the equipment used must have adequately wide dynamic ranges. With the film digitization system we have installed, it is possible to usefully digitize films up to a maximum film density of D = 4.5 with a contrast resolution better than D = 0.01 applying the maximum light level and a prolonged integration time.

3. Evaluation procedure

fig. 2 Contrast, halfwidth and noise level of a RT indication

Digitized radiographs are first qualitatively interpreted on the monitor. By optimising brightness, contrast and magnification as well as by applying image processing algorithms, defect indications are enhanced and their extent can be more easily examined and printed. In the second stage a quantitative analysis is performed using the grey-scale profile plot vertically to the indication (fig. 2). These profile plots are taken from unprocessed digitized data and at a position with the most distinct indication and the highest signal-to-noise-ratio. The severity of a defect indication is described in terms of its contrast D, of its halfwidth HWB and of its length above a defined contrast theshold (for the definition of these parameters see fig. 2). The contrast threshold is to be determined with regard to the existent noise level. The noise level depends on the applied RT technique (X-rays or Gamma-rays) and on the surface contour of the weld. Typical values lie between D_N = 0.01 (isotope radiation) and (D_N = 0.04 (RT of welds with X-rays).

In order to make evaluation of grey-scale profile plots more objective and faster, we developed an algorithm for a computer-based evaluation of crack-like indications /4/. The automatic evaluation is conducted in four steps: to suppress noise, an averaging routine is performed of a number of scan lines, for example 5 lines or 175 (m, predetermined by a parameter data set. In the second step, the search phase, the scan lines are searched for relative minima above a contrast threshold D = 0.01. To suppress signals with little or no linear coherence in the direction of the expected defect orientation (longitudinal or vertical to the weld axis), all indications are deleted in the next step whose length is less than a pre-set value. Fig. 3 shows the result of automatic evaluation with a length threshold of 10 mm. In the digitized radiograph the areas of indications which meet the pre-set criteria remain (minimum contrast D greater than 0.01 and minimum length 10 mm). The last step is to print the remaining indications with their half width and their contrast.

1. indication search     2. noise suppression    3. density profile fig. 3 Computer based evaluation of digitized RT indications

The printed out indications are evaluated with regard to their pattern and the parameters (maximum contrast, width and length) derived from the grey scale values. They are classified as

• form indications like root grooves, misalignment
• indications of defects like pores, inclusions or lack of fusion which definitely originate from manufacture
• indications of defects which may have developed or grown during plant operation like cracks or wall thickness degradation.
In the case of defect indications found at in-service inspections the available radiographs of preceding inspections have to be analysed in the same manner and indications have to be compared with respect to an increase during plant operation.

4. Criteria for discrimination of crack and form indications at weld inspections

fig. 4 Contrast distribution of crack and non-crack indications fig. 5 Max. contrast and halfwidth of crack and non-crack indications fig. 6 Max. contrast vs. wall thickness of crack and non-crack indications fig. 7 Comparison of interpretation results between conventional film viewing and digitized radiographs applying the proposed criteria

The before mentioned overestimation by the conventional examination employing a film illuminator i.e. the interpretation of form indications in many cases as cracks has several reasons: after some cracks undoubtedly were present in pipe welds, the films were examined more sensitively by the radiographers. The evaluation is performed on the basis of pattern recognition without a well defined registration threshold. Most of the crack indications are of very low contrast (see fig. 4). The width and the contrast of RT indications are assessed qualitatively only.

The values of these can be quantified in digital radiographs very easily. By comparing RT indications with the results of the destructive examination, the following criteria have been found by us to discriminate the relevant indications e.g. cracks and lack of root fusion against non-relevant indications as superficial shrinkage grooves:

• Indications with high contrast D ( 0.08 are generally interpreted correctly by conventional examination. RT indications from cracks which were not detected or from indications which were misinterpreted as cracks possessed a contrast lower than that. Indications with a contrast below D = 0.04 (approximately the noise level) could generally not be interpreted correctly.
• The half width of RT indications of corrosion cracks is between 100 (m and 300 (m (fig. 5). Low contrast indications with a half width greater than 300 (m refer to smooth undercut.
• RT indications in radiographs of lower film density are underestimated compared with those indications in radiographs of higher film density which have the same contrast. Since the indication contrast D is proportional to optical density the actual contrast at a given film density D is converted and normalised to the reference value D = 3.0 by multiplying with a factor 3.0/D in order to have comparable conditions.
• At the same contrast level RT indications are overestimated in the case of lower wall thickness. This is the case for RT indications caused by fine shrinkage grooves with a depth of about 0.1 mm. For taking into account this parameter, a contrast threshold is introduced which is dependant on the wall thickness to characterize the relevant RT indications (fig. 6). This curve takes into account the contrast increase of a defect (slit) of the same depth with lower wall thickness due to less scattered radiation. Diffuse indications having a lower contrast than the threshold curve are to be classified as form indications.

The interpretation results from conventional film viewing and from digitization applying the discussed criteria are shown in fig. 7. Even with digital image processing only those indications can be interpreted as defects which differ from the noise level. Therefore, the radiographic testing parameters have to be optimised when generating a radiograph in order to reach a flaw detection sensitivity as high as possible. With the proposed evaluation criteria a more reliable interpretation of defect indications can be achieved with the help of digital image processing of radiographs.

5. References

1. M. Erve et al. Rißbildung in Rohrleitungen aus stabilisierten austenitischen Stählen von SWR-Anlagen - Schadensbefunde und Schadensursache - 20. MPA-Seminar, Stuttgart, 6.-7.10.1994, Conference proceedings vol. II, 29
2. G. Csapo, T. Just, H. Eggers, E. Hein, R. Nimtz Flaw Detectability of NDT in Thin-walled Austenitic Pipe Welds 6th Europ. Conf. on NDT, Nice, 24.-28.10.1994, Conf.proceed. p.995-999
3. Du Pont NDT Scan Manager User Handbook, June 1993
4. W. Thale, R. Clausen, T. Just Digital Image Processing for the Interpretation of Radiographs used for Condition Monitoring Insight vol 38 no. 9, September 1996, p.632-635

|Top|