Table of Contents Session: Aerospace

Determination of the POD for High Resolution Eddy Current Inspection on Turbine Disks W. D. Feist*, G.-R. Tillack *Corresponding Author Contact: MTU-München, Abt.: TWP, 80976 München, German ; wolf-dieter.feist@muc.mtu.dasa.de Tel: +89-14892527

TABLE OF CONTENTS

Abstract Introduction Investigative Procedure Test Results and Evaluation Reliability of Inspection Conclusion References

Abstract

Turbine disks for aero engines are highly-stressed components. The integrity of these parts must be proven by nondestructive inspection especially in critical zones such as the hub bore. The disks must be tested by the eddy current method after production as well as during overhaul inspection.
An investigation was carried out on the hub bore of a previously spin tested disk with natural fatigue cracks to find a technique for high resolution eddy current inspection and the method to calculate the probability of detection function (POD vs. a) for this inspection.
A special eddy current probe was chosen to improve the signal-to-noise ratio. The instrument parameters were optimized on a calibration block with artificial half penny shaped slots of different sizes. A replica technique was used to measure the true size of the natural fatigue cracks. The POD was calculated using the â-versus-a method.

Introduction

Fig 1:Turbine Rotor Disk

At overhaul, turbine disks are inspected to make sure they are free from incipient cracks jeopardizing their safe operation for the duration of another inspection interval. The optimum inspection method for the purpose is imaging eddy current inspection. To ensure maximum operational safety, the engineer must know which crack sizes the inspection method will safely detect and how safe detection can be maintained with confidence. Cyclic spin tests on production disks, which not only produce real cracks but also permit their growth to be tracked, prove an excellent approach to the problem. They provide an ideal opportunity to determine and optimize detectability limits and the reproducibility of inspection. A high-pressure turbine disk (Fig. 1), which had previously undergone a cyclic spin test under a certification program, served to witness the capacity of the high-resolution imaging eddy current inspection method and adapt the method to routine disk inspection requirements.

2. Investigative Procedure

The hub bore of a high-pressure turbine disk that had previously undergone a 46,600 cycle spin test was inspected by the eddy current method. For the purpose, an extremely sensitive probe was used that consisted of one exciter and two differential-connected receiver coils. The calibration block used had half-penny-shaped slots (Table 1) and the same inside diameter as the hub bore of the disk. It was used to calibrate the inspection procedure prior to each inspection and so permit direct size comparison between artificial and natural defects, as well as precise size estimation.

Slot No.	Surface length µm	Depth µm
2	170	70
4	210	120
6	330	160
8	420	190
10	550	270
12	670	270

Table 1: Dimensions of slots in calibration block for hub bore (d = 42 mm, material: U700-PM, slot orientation: axial/radial, slot shape: half-penny-shaped)

3. Test Results and Evaluation

Fig 2: C-scan presentation of eddy current inspection, unfolded section. Hub surface above, calibration block below Fig 3: Selective enlargement from Fig. 2, indication No. 4, typical crack indication pattern

Fig. 2 shows the test result in the shape of an unfolded hub bore section, with slot Nos. 2 - 12 of the calibration block shown underneath for direct comparison. To verify the eddy current test output, a foil replica was made of the hub bore for optical evaluation under a microscope. Agreement between the two methods was noted for 12 indications. With indications Nos. 13 and 14 (Fig. 2), no crack was detected in the replica. Whereas optically, two cracks 118 and 150 µm in surface length were detected which even in repeated eddy current inspection failed to show up in the C-scan presentations.

For size evaluation based on eddy current signals, the indications of the C-scan presentations of hub bore and calibration block were compared. This may be helpful when in straight amplitude evaluation, the probe prematurely reaches a saturation point above which size conclusions are prevented. Compared with the imaging presentation of indications, however, amplitude and local dimension enter into evaluation to provide greater dynamics of the test data.

Juxtaposition of size evaluations from replica and eddy current signal, respectively, shows very good agreement (Table 2). It appears that a surface length of 200 µm is the limit for safe detection with either method. In this range (200 µm), two each indications were obtained from either method that escaped detection by the respective other.

The good, i.e. distinct detection and assessment of the cracks is made possible by the area-type of inspection data presentation. If assessment would resort to threshold evaluation only, there would be either many false indications, e.g. in the 10mm (vertical) range (Fig. 2), or the smaller indications would not be detected. The cracks each produce a characteristic pattern consisting of at least one peak and one valley. Even with small indications this pattern seems to be continued, although diagonally offset. Larger indications are a multiple of the pattern. This becomes readily apparent from the enlarged view of indication No. 4 (Fig. 3).

Reliability of Inspection

Fig 4:Linear correlation between â and a for the hub bore Fig 5: POD curve and confidence interval for the hub bore __POD --- 95% confidence bound       a90/95 = 348.6 µm

NDT techniques are increasingly judged by the reliability with which they detect certain types and sizes of defect.
If the probability of detection (POD) versus defect size is used to describe the reliability of a NDT system, differentiation is made between two different approaches:

• The test output is evaluated for testimony of the presence of a defect (experimental detection probabilities)
• An alternative analysis method was developed which relates defect size a to corresponding signal size â.
Both methods are detailed in reference [1]. The present investigation bases on the latter of the two methods.
This method assumes that system response â and defect size a are correlated. It is only when this parameter â exceeds a given decision threshold â_dec that the signal is interpreted as coming from a void in the material. The variable â accordingly contains the information whether and with what probability a defect can be detected. Introduced additionally is a recording signal threshold â_th below which the signal is obliterated by noise to make detection impossible. The scope of application of NDT systems is generally limited at the upper end by their saturation behavior, a phenomenon being considered by introducing a saturation threshold â_sat.

To determine the POD curve in accordance with the above evaluation method, use was made of the sets of data from Table 2, columns 2 and 4 (defect size a and associated signal size â). No suitable algorithm being available for automatic evaluation of indications, the defect size estimated after comparison between the imaging eddy current indication of the real crack and that of an equivalent artificial flaw was used. The recording signal threshold was determined at 180 µm and the saturation threshold at 670 µm. The data in Fig. 4 show that in the range used, a linear relationship exists between log a and log â. The results of the investigation are shown in Fig. 5. The solid line represents POD versus defect size, the dashed line the associated 95% confidence bound. Inspection systems are as a rule rated using the value a_90/95, which gives the defect size which with 90% probability is found at a confidence level of 95%. In the present case, it amounts to a_90/95 = 348 µm, whereas the POD curve is already up against the 100% boundary. In a further development of the method [2], defect size can be correlated directly to the eddy current amplitude of the artificial defect. To determine the recording signal threshold, the test data distribution was in this case determined at a mean 1660 and a maximum 1755 corresponding to â_th in an unflawed portion of the calibration block [3]. Fig. 6 represents the linearity between log a and log â and shows that there are no saturation effects in the range used. The results of the POD analysis are shown in Fig. 7, which puts a_90/95 at 239 µm, whereas the POD curve for cracks in excess of 200 µm already hits the 100% boundary.

Indication No.	Surface length (µm) Evaluation of replica	Eddy current comparison test slot No.	Surface length estimated from eddy current signal
1	514	10	550
2	399	6-8	375
3	758	6-10	485
4	450	8	420
5	559	10	550
6	213	2-4	190
7	2,949	often greater, no comparison possible	>670
8	357	4	210
9	1,169	>12	>670
10	374	6	330
11	367	6	330
12	202	4	210
13	no crack detected	2	170
14	no crack detected	2	170
15	150	no indication	-----
16	118	no indication	-----

Table 2: Comparison of indication assessments,
replica vs. eddy current signal

Fig 6:Linear correlation between â and a for the calibration block Fig 7: POD curve for the calibration block and confidence interval __ POD       95% confidence bound --- a90/95 = 238.9 µm

When comparing the results from the two methods (Figs. 5 and 7), you will note that in the second case (Fig. 7) the POD curve is distinctly steeper, so that 100% POD is reached at a crack length of a mere 200 µm. This seems to contrast with the observations made in the juxtaposition of the test results from foil replicas and C-scan presentations, respectively, on the hub bore (see 3. Test Results and Evaluation). The difference is chiefly attributed to the fact that for determining the scatter, use was made of the test data distribution in the unflawed portion of the calibration block, whereas the first evaluation, where the signal amplitude â was determined by optical comparison of the C-scan presentations, refers to the disk hub, which apart from the cracks also contains disturbances that were not identified as cracks.

Conclusion

The EC-method here presented provides a high probability of detecting LCF cracks to an approximate surface length of 350 µm max. in an already stressed but still smooth-surface bore. In the inspection process, the quality of the probe, the exact operator-independent manipulation of the probe and the area-type image of the test result all play an important role. For quantitative assessments, the method additionally requires the use of a calibration block corresponding to the part to be inspected. Without the imaging presentation of the test result, the detection sensitivity is no doubt appreciably lower. To what extent the quality of inspection suffers from the effects that earlier engine service leaves on a disk (e.g. surface contamination), can be determined only through comparison with a service-exposed disk. Imaging presentation here again is sure to give substantial improvement over straight threshold evaluation. Future eddy current inspection systems, then, must be designed accordingly if they are to satisfy high detection capacity and confidence requirements.

Reliability (POD versus a) analysis shows a distinct difference between the two POD curves. The curve generated by evaluation of the real cracks in the disk hub (Fig. 5) appears to apply more readily to conditions on service-exposed components. Whereas Fig. 7 more probably reflects conditions encountered on virgin parts.

Further work should focus on the development of a suitable algorithm enabling, perhaps by pattern recognition, quantitative, operator-independent evaluation of indications and discrimination of disturbances to be achieved.

References

1. A. P. Behrens, NDE Reliability Data Analysis, Metals Handbook, Vol. 17, pp. 689-701, ASM International, Metals Park, Ohio
2. O. Köser, G. Mook, W. D. Feist, L. Steinhauser, Determination of a POD versus Crack Size based on the ROC Method for Eddy Current Inspection, European Conference on Non-Destructive Testing, 1998, under preparation
3. G. Mook, O. Köser, private communications

|Top|