mopdf.com

Section 1IntroductionIn the current environment of restricted budgets, aircraft component parts are often used beyondtheir design life. In order to avoid the large cost of replacing critical rotating parts as they reachtheir “safe-life” limits, a Retirement For Cause (RFC) program was implemented. The RFCprogram involves periodic nondestructive evaluation (NDE) inspections to assess the damage-stateof components. Components with no detectable cracks are returned to service and those withdetected cracks are discarded before they can cause an incident. This process allows parts withhigh life to be used to their full potential. The Air Force has embraced RFC and currently uses itsuccessfully to manage component life for several of their gas turbine engines.Managing the structural integrity of engine parts through the RFC program requires quantifying thecapability of the NDE system used in the inspections. The RFC uses the highly automated EddyCurrent Inspection System (ECIS) developed by Veridian. The capability of the RFC/ECIS hasbeen quantitatively evaluated by inspecting specimen sets of representative material by geometrycombinations. The ECIS response (â) at cracks of known size (a) are then analyzed to estimatethe probability of detection (POD) as a function of crack size, or POD(a). The reliably detectedcrack size, a90, defined as the crack size for which POD(a90) 0.90, is used as the one numbercharacterization of the ability of ECIS to detect cracks in the material by geometry combination.The concept of obtaining POD(a) from â versus a data is shown in Figure 1. The probability ofdetection of a random crack of some fixed size is determined from the scatter in the NDE systemresponse about the mean response for cracks of that given size [1]. Experience has shown that alinear fit of log â versus log a is often a reasonable model for the mean response over some range ofcrack sizes. In a range of crack sizes for which the linear fit is reasonable and the scatter about thefit is normally distributed and does not depend on crack size, POD(a) will be a cumulative normaldistribution function with µ and σ as given by the following equations:µ (log âdec – B0)/B1σ s/B1(1)(2)log âi B0 B1 * log ai ei,(3)whereandei the difference in response, âi, of crack i from the mean of all cracks of size a;s is the standard deviation of ei;âdec is the response decision threshold for a crack indication;B0,, B1, and s are fit parameters that are estimated from the data.The a90 value is then given by the equation:a90 exp (µ 1.282*σ).(4)1

21000POD calculated from three parameters:Intercept of line - B0Slope of line - B1Standard deviation of â deviations from line - sParameters of POD model: (ln âthr - B0 )/B1 s/B198765Ahat (counts)432100987654SIGNAL THRESHOLD3234567891023Crack Depth (mil)Figure 1. Schematic for Obtaining POD(a) from â versus a DataTo date, appropriate material/geometry combinations have been used to simulate real inspectionsfor the POD(a) characterization of RFC/ECIS capability, and the three parameters of the â versusa analysis are estimated directly from the inspection results. To avoid the necessity of producingspecimen sets for all materials by geometry combinations that will be inspected, a method forcorrelating POD results between specimen sets is being sought. Two characterization conceptsto account for the differences in geometry by material combinations have been suggested. Theseconcepts attempt to predict the unknown parameters of the log â versus log a fit for one geometryof one material, from inspections of three similar specimen sets, (1) the same geometry of asecond material, (2) flat plate data of the same material, and (3) flat plate data on the secondmaterial. Each concept attempts to predict B0, B1, and s, the unknown parameters of the log âversus log a fit, from values obtained from inspections of the other three specimen sets. Becausethese methods involve material-based adjustments to the parameters obtained from the samegeometry, these methods are also referred to as material correlation methods. The two materialcorrelation methods are referred to as the “difference” and “ratio” methods, since the parametersof the missing combination are obtained either from differences or ratios of the parameters fromthe available sets.To further demonstrate the concept of material correlation, Table 1 represents examplecomponents with five types of materials and five types of geometries. Parameters for materialand geometry combinations in the first row and first column are assumed to be known.2

Parameters for all other material and geometry combinations must be inferred from the materialcorrelation method.Table 1. Geometry – Material CombinationsGeometry12345Material1 2 3 4 5 ?A material correlation study performed by Veridian in September 1999 [2] was motivated by theneed to minimize the required number of specimen sets while providing the most accurateinspections possible. The primary proposal of the material correlation project was that the ratioof the eddy current responses from cracks of the same size in two different materials of the samegeometry should equal the ratio of the eddy current responses from cracks in specimens madefrom the same two materials having a different geometry. Because the â versus a data isconstructed in a log-log domain, the proposed materials correlation assumption equates toparameter differences.The goal of the study was to show whether or not reliability analysis results obtained from theeddy current inspection of certain specimen sets could be used to predict the reliability analysisresults for another specimen set of like material but with different geometries. For example, canthe reliability analysis parameters be determined for Ti-6246 bolt hole specimens using bolt holesin IN 718 and flat plate specimens of IN 718 and Ti-6246? To test this, the predicted probabilityof detection was compared to the actual POD using combinations of materials and geometriesthat already exist, as shown in Table 2.Table 2. Material Correlation Test MatrixGeometryMaterial10.342" Bolt HoleIN 71820.460" Bolt HoleTi-62463Flat PlateTi-62464Flat PlateIN 7183

Researchers at the University of Dayton Research Institute (UDRI) reviewed, analyzed, andevaluated the study to predict POD through the material correlation study performed by Veridian.This report will present a discussion of requirements for performing materials correlation inSection 2, examination of the data acquisition methods of the study in Section 3, and anevaluation of the POD inference methods using the data collected in the study in Section 4.Finally, general recommendations for optimizing the material correlation study will be providedin Section 5, and Section 6 summarizes the study and its results.4

Section 2Requirement for Performing Material Correlation StudiesMany variables must be controlled to acquire eddy current responses from different sets of testspecimens if they are to be useful for material correlation studies. Some of the known variablesthat can affect the amplitude of the eddy current responses from a crack are listed in Table 3.Table 3. Eddy Current Inspection nherent Coil SensitivityControllableüRelative/Differential SensitivityüCoil sizeüLift-offüGainüPhase AngleüFrequencyüFiltersüConductiveüCrack OrientationCrack OpeningAspect Ratio DepthResidual StressGrain SizeInspectionüStep SizeCalibration VariationGeometryScan SpeedüMaterial NoiseüRandom NoiseüLift-offüGeometric VariablesüAn extensive amount of work is required to examine each variable and determine the extent ofhow and why it affects the eddy current response data acquired from cracks in reliability specimens.It would be difficult, if not impossible, to control all of the variables. If they are not controlled,the effect of not controlling them must be understood and predicted. This is the crux of theproblem. Since many of these variables necessitate different material/geometry combinations, they5

cannot be strictly controlled. Therefore, they must be understood and their behavior predicted. Thequestion that remains unanswered is: How well can this be done?Phase angle calibration is a good example of how difficult it is to manage a variable during eddycurrent inspection. Differential eddy current probes are commonly used for RFC-type inspectionsto achieve the signal-to-noise ratios needed to detect small cracks. The differential nature of theprobes greatly affects the phase angle setting in the eddy current instrument. All responses from theprobe result from the difference between the signals coming from the two coils. Thus, the phaseangle setting is as much the result of the imbalance of the two coils as it is the interaction of theeddy currents with the electrical properties of the material. The responses from each coil aredependent on the material, the crack, and the relationship between each coil and the metal in itsproximity. It is difficult to determine how each differential probe responds to the uniquecombination of its own imbalance, the specimen, geometry type, and the crack. Acquiring eddycurrent responses from differential probes from three specimen sets, and then using the data topredict the eddy current response for cracks in a fourth specimen set, requires a thoroughunderstanding of the probe’s interaction with the fourth specimen’s material and geometry, as wellas the cracks.Essentially, the problem in this example is twofold:1. Whether the unknown calibration phase angle could be predicted for a material/geometrycombination based on the known phase angles of three other material/geometry combinations2. Given that unknown calibration phase angles can be predicted and the phase angle can be setaccordingly, whether the crack response (phase and amplitude) could be predicted for thismaterial/geometry combination from the crack response on different cracks found in the otherthree material/geometry combinations.These two problems must be resolved in light of the other variables, such as lift-off. Thequestion deals with modeling the phase response and then the crack response behavior at thatphase angle setting.Another issue that complicates any effort to do materials correlation is the application of thematerials correlation results to real engine part inspections. The conditions under which thematerials correlation tests are conducted must be reproduced in the inspection of real engineparts. Such inspection conditions include coil type, inspection algorithm, calibration (phase andgain calibration), scan rates, filter settings, inspection parameter values, and others. Often,however, the particulars of a real engine part inspection will constrain these variables. This meansthat the materials correlation tests must be designed with the inspection in mind.An example of this would be bolt hole inspection of a material for which there is no PODspecimen set. These inspections are conducted with noncontact bolt hole probes scanning at avery high rate. Many bolt hole inspections are conducted at 1500 RPM. A bolt hole diameter of0.5 inches gives a surface velocity of about 40 inches per second. Performing a materialscorrelation using flat plate and bolt hole specimens of the appropriate materials would requireattempting to collect flat plate data using proper controls. However, it would be virtuallyimpossible to match surface velocity on the flat plate specimens because the RFC linear axeshave a maximum velocity of 300 inches per minute, or 5 inches per second. The flat plate6

specimens could be mounted in a test fixture clamped to the turntable, which is routinely done,and the turntable rotated. The current flat plate fixture holds specimens at a radius of about 7inches. At the maximum turntable speed of 18 revolutions per minute, the surface velocitywould be only 14 inches per second. Since it is apparent that the surface velocity cannot bematched, the only alternative is to assume that the slower surface velocity can be accounted forby scaling all parameters that are scan-speed dependent, such as filter settings and inspectionparameters.Another variable that would need to be controlled in this example is lift-off. Since bolt holeprobes are noncontact, the flat plate specimens would have to be scanned with the same lift-off,which poses implementation challenges for a surface probe. Another issue is the type of notch usedto calibrate the probe. Bolt hole probes are calibrated using corner notches, and surface probesare calibrated using surface notches. The surface probes on the corner notches used for bolt holecalibration would have to be calibrated to remove this source of variation. Since this isimpossible, an alternative would be to use the bolt hole probe to perform the flat plate specimeninspection. Obviously, this would be difficult if not impossible, especially in light of the need tokeep lift-off uniform; however, it has the advantage of removing the coil as a variable.Another complication in this example is the need to control the inspection routine. The bolt holeinspection routine, including inspection parameters, step sizes, and sampling rates, would have tobe used on the flat plate inspection. If any of these steps cannot be controlled, then the results ofthe tests and models would depend upon the validity of the assumptions used to ignore thedifferences.It may be impractical, if not impossible, to control the variables. In Veridian’s attempt to test aproposed model for material correlation, an application was selected that would allow many ofthe variables to be controlled. Their tests and controls will be discussed in more detail in thenext section. While Veridian used flat plate and bolt hole specimens, they tested the materialscorrelation model using a scallop setup and inspection method. They had to design a specialprobe that could fit into a bolt hole and also ride on a surface specimen. They used a specialphase and gain calibration routine. In effect, Veridian created a hybrid test condition to avoid theproblems that would be encountered in an attempt to apply materials correlation to a realinspection. This is not a criticism of Veridian’s methods, but a concession to the difficulty ofcontrolling the variables.From this discussion, it is clear that each inspection application forces a unique set of testingconditions on the material correlation test setup. The uniqueness of the test conditions almostassures its inapplicability of results to other applications. This narrow applicability results in therequirement of three sets of data for each and every possible material/geometry combination forwhich there are no specimen sets. Materials correlation trades the cost of one set of newspecimens for the cost of three POD runs of data on existing specimens, plus the engineering costof designing and creating the scan plans and possibly other hardware (such as probes) to carry outthese runs. This assumes that the required test conditions can be met.7

One final complicating factor is that not all geometries have a corresponding POD specimen setin any material. This case indicates the existence of what might be called a geometry correlation;that is, POD is determined on an existing set of specimens of one geometry and it is assumed thatthese results apply to another geometry. For instance, a scallop set of specimens might be used torepresent an anti-rotation tab fillet radius. Issues of probe, coil, lift-off, and curvature furthercomplicate the assumptions made for materials correlation.8

Section 3Examination of Data AcquisitionAs mentioned in the previous section, Veridian created a hybrid test condition, which wasessential for testing materials correlation. Veridian understood this, as shown by the followingquote from their report: “A primary concern in this project was to attempt to minimize anyvariation in the multitude of parameters in data collection, including, among others, gain, phase,lift-off, coil orientation, filtering, scan speed and signal processing”. As mentioned earlier,Veridian designed and built a special probe that would allow them to scan both the bolt hole andflat plate specimens. The probe could be configured in two different ways to furtheraccommodate this constraint. Veridian also created special software to allow the routine toperform inspections on both specimen sets. Necessity dictates differences in setup for thesedifferent inspections, but as noted by Veridian, “Critical functions were performed by commonsubroutines.” Inspection setup was chosen to be consistent with the scallop inspection.To perform gain calibration, Veridian used the master set of reference standards to avoidadditional variation due to notch parameter uncertainties. They performed phase calibration onthe IN 100 reference block across the long notches. The notch signal was then rotated into thevertical component of the impedance plane. For Ti-6246 materials, an additional phase calibrationwas performed. Gain calibration for both materials was performed on the IN 100 block after phasecalibration on the IN 100 notch. For Ti-6246 specimens, the phase was rotated after gaincalibration to be consistent with the phase determined on the Ti-6246 notch. This strategy wasused to reduce variation caused by gain calibration – Veridian stated, “This method of calibrationwas to provide consistent gain setup across all data sets while compensating for phase rotation indissimilar metals” [2]. The propriety of such a protocol is arguable, but their attempt furtheremphasizes the difficulty in performing a good material correlation.Veridian describes their method of selecting which cracks to use on their test. The need for thisselection process arose out of the design of the specimen set. Since the specimen set wasdesigned to accommodate fluorescent penetrant inspections, there were multiple cracks on anygiven specimen, which could be in different orientations. Veridian eliminated those cracks thatwere inappropriate for this test.Veridian used a D20 diffe

cause (rfc)/engine structural integrity program (ensip) nondestructive evaluation (nde) system development volume 3 – material correlation study alan p. berens wally hoppe david a. stubbs