·Home ·Table of Contents ·Reliability and Validation 2

Reliability Assurance for Automated Eddy Current Inspection Systems Carlos Pairazaman, Customer site support manager for Veridian Engineering, Oklahoma City, Oklahoma, USA Sara Keller, Branch Chief, Processing Engineering, United States Air Force, Tinker Air Force Base, Oklahoma, USA Ward D. Rummel is an independent consultant for D&W Enterprises, LTD., Littleton, Colorado, USA. Contact

Abstract

Implementation of fully automated eddy current inspection systems required validation and process control to assure initial and continuing reliability in operations. This paper summarizes the principles, procedures and controls used to establish the system operations (detection) capabilities for initial applications; procedures for set-up and implementation of additional inspection modules; and procedures for periodic assessment and validation as a part of operating system process control.

The protocol for establishing and maintaining system operational integrity involves

1. system performance CAPABILITIES validation using representative flawed specimens to establish a probability of detection (POD) for detection of flaws at a given level;
2. system validation of additional units (REPRODUCIBILITY) by scanning representative "master gage" flawed or notched parts to verify equivalent output responses to those obtained during the initial CAPABILITIES validation (POD); and
3. periodic system revalidation (REPEATABILITY) by scanning representative "master gage" flawed or notched parts to verify equivalent output responses to those obtained during the initial CAPABILITIES validation (POD).

The methods and procedures for establishing and validating system design CAPABILITIES; for assessing REPRODUCIBILITY of additional system modules; and for revalidation of operating modules (REPEATABILITY) are described and discussed.

INTRODUCTION:

The structural integrity of critical engineering components is often dependent on the capability and reliability of the inspection procedures used in acceptance and in life-cycle management. Inspection reliability is often expressed in terms of the probability of detection (POD). Although the probability of detection (POD) is an end to end measure of the performance of an inspection process at a specific time, it reflects the capability of a procedure and can only be viewed as a process reliability measure if the inspection process is under control. Inspection process control is in turn, widely viewed as an issue of human skills and skills variables. It is commonly assumed that automation of a process produces a reliable process. Indeed variance in the performance of an automated process is a product of the variances of critical process parameters that are inherent to the process.
Establishing a POD for an automated system provides a baseline characteristic capability for that system design and system configuration. It is unnecessary to repeat POD demonstrations for subsequent systems produced to the same design and configuration. Subsequent confidence is verified by determining that the system process variables are within the allowable variances for the system design. Indeed, greater confidence can be established by control of system variables than can be gained by a single point demonstration of capability by the POD method. This paper discusses acceptance of automated eddy current inspection systems that are used to inspect a wide range of gas turbine engine components and component configurations.

DISCUSSION:

The performance capability of any nondestructive inspection procedure is dependent on the signal generated from the target artifact (flaw - hereafter referred to as a "SIGNAL") and the non-relevant signals generated by material, geometry, surface texture, etc. (hereafter referred to as "NOISE". A typical response from experimental measurements from a single flaw is shown in FIGURE 1.

Repetitive response from multiple flaws of equal size results in broadening of the response distribution. This broadening is the result of flaw to flaw variations as well as measurement variations and are accounted for by using multiple flaws in the generation of a typical POD curve. The spread between the upper limit of the noise and the lower limit (signal and noise) of the flaw response enables repetitive detection and discrimination / identification of flaws of that size without false calls (Type II errors). The practical threshold detection and discrimination limit is at that small flaw size at which the signal and noise responses converge without overlap and detection / discrimination can be attained. It is wise to maintain a signal / noise margin (safety factor) in practical applications to allow for unanticipated variations in the NDE procedure.

Fig 1: Repetitive response from a single flaw

Fig 2: Repetitive response from flaws of equal size

Small sample sizes assume that the flaws selected are representative of the population of flaws to be detected and that flaw to flaw variance in application is bounded by the flaws selected for assessment. Various thrusts have been directed to modeling the flaw to flaw variance and have been successful for simple flaw configurations. In many applications, this variance is accounted for by including a margin (safety factor) in detection requirements and by follow-up data collection and analysis of signal responses from service hardware. For complex configurations, larger margins may be used to address difficulties in validating margin assumptions.

When a small sample size is accepted as being representative of the population, repetitive measurements can be made on the selected flaws to establish the signal / measurement variance for each flaw size sampled. FIGURE 2 illustrates the broadening of response due to multiple measurements from flaws of equal size. This method produces a data set that can be used to establish a discrimination threshold and for plotting a POD curve. Although the number of measurements can provide a high measurement confidence level based on the small sample set, the measurements are not fully independent and thus less rigorous than that obtained by independent measurements on independent cracks. The POD curve generated is a measurement curve and is representative of most important elements of the characterization task, but may not fully describe a capability if the response from service flaws of equal size varies significantly from the selected small sample set.

The second part of NDE procedure optimization is in setting the acceptance level for the signals provided. For large flaws, the signal and noise are well separated and the threshold decision level can be easily set to provide clear discrimination as shown schematically in FIGURE 3.

Fig 3: Signal and noise separation for large flaws which provide clear signal discrimination

Fig 4: Theresholdlevel set too high with resultant misses.

If the threshold decision level is set too high, flaws will be missed. This condition may be imposed when signal and noise separation would otherwise allow clear discrimination as illustrated in FIGURE 4. This is a condition often experienced when the threshold signal level is set on a slot in a "calibration" specimen and consideration is not given for the reduced response of a crack of a size that is equal to that of the slot.

The limit of capability of an NDE procedure is reached when the signal and noise separation approaches zero. When signal and noise overlap, both misses and false calls will result as illustrated in FIGURE 5. In this case, if the threshold level is set to assure detection, the number of false calls will increase. A level of false calls can be tolerated if a secondary procedure (usually NDE) is applied to resolve false calls and provide the required discrimination. CAUTION: Applying the same NDE procedure cannot resolve false calls since the same signal and noise conditions are application. Likewise, an NDE procedure with lower discrimination capabilities does not provide resolution. This error has been frequently observed in the use of visual inspection to resolve eddy current findings.

Fig 5: Overlap of signal and noise results in both misses and false calls.

When an optimized signal, noise and discrimination (acceptance) level are established, the capability of the inspection process can be characterized in the form of a Probability of Detection (POD) measurement as shown in FIGURE 6. For an automated system, performance RELIABILITY, at the capabilities level established, is dependent on REPRODUCING the conditions which reproduce the same signal level response to the same size flaw (REPRODUCIBILITY) and on REPEATING the same distribution of signal and noise responses to the same flaw sizes that were used in establishing CAPABILITY (POD). The RELIABILITY of an inspection procedure may thus be viewed in terms of REPRODUCIBILITY and REPEATABILITY.

Fig 6: Typical probability of detection(POD) curve

REPRODUCIBILITY in an inspection process is primarily dependent on the "calibration" of the inspection system. Most classical measurements are made with the aid of a reference "calibration" artifact or "standard". Calibration "standard" artifacts are "measured" by reference to a master standard that is traditionally retained as a as a national resource. Control is approached by international agreements to provide a common basis for exchange in commerce. It was therefore logical that a reference slot has evolved as a calibration artifact for most NDE measurements and physical measurement of slot size may be traceable to a national "master standard". Slots are economical to produce with available technology and are commonly specified in establishing and applying NDE procedures. Traceability of reference calibration artifacts (slots) are assumed when they are used in validated NDE procedures. Users are cautioned that apparent consistency in slot size and configuration (mechanical / geometric consistency) may not provide a consistent NDE response and NDE response verification is required before artifact acceptance. Unfortunately, a single slot is often used for reference and set-up and linearity of response of the NDE procedure is assumed. This practice partially satisfies the REPRODUCIBILITY requirement, but is incomplete in providing confidence in end to end system response. A system (instrument, cables, probes and accessory commodities) "calibration" which includes not only a baseline set point, but a response linearity verification (characteristic response hereafter referred to as "linear") is necessary to provide high confidence in the REPRODUCIBILITY system response / output.

Modern electronic instruments are produced with attention to linear response, but periodic validation of the response linearity is required to maintain confidence in instrument / system output. It is often assumed that classical electronic "calibration" performed by an independent "calibration laboratory" assures linearity of response. Since the electronic instrument constitutes only a part of the NDE system, an end to end system linearity response verification is an essential part of a validated NDE procedure. A change in eddy current probes or cables can significantly change the response linearity. When an NDE procedure is intended for application at different locations or with multiple instruments, a "master gage" "calibration" specimen and sensor probe (transducer) are required to establish and document the output variability (or conformance) due to "calibration" practices. It is necessary to establish output variability bounds and to assure performance remains within those bounds for each inspection procedure. This is an essential requirement for both manual and automated eddy current systems.

REPEATABILITY is assumed to be constant for a fully automated inspection system. That assumption can be supported for scanning and measurement processes which have no variance. Automation of a variable process will continue to produce a variable output. For instrumented methods, the inherent REPEATABIITY of measurements is assumed to be constant due to the ability of electronic devices to produce the same decision for the same signal levels (alarms, recording devices etc.). A variation in part surface finish, a variation in scanning speed, a variation in a scan raster or a variation in part geometry will be reflected in variations in REPEATABILITY. For automated scanners or robots, parts wear may result in an increased variance in the scan path and hence a decrease in REPEATABILITY. It is important to note that REPEATABILITY is a multi-dimensional characteristic and depends on the precision of probe location, orientation and on consistence of probe, cable and instrument response.

APPLICATION:

The "Retirement for Cause" (RFC) system is a precision automatic eddy current inspection system that is used to assure structural integrity and thereby provide life extension of gas turbine (jet) engine components. Reliable detection of small flaws in complex geometries are required. Each geometric feature requires eddy current probe design / selection; wave form analysis; setting thresholds in accordance with acceptance criteria; and programming the robot motion control. The CAPABILITY of each inspection sequence is validated by the full POD method using reference test specimens and slots in engine components. This method provides validation of both the detection capability of the system and the scanning coverage.
Once the CAPABILITY has been validated, software and master-gage (slotted) component / specimen are characterized, configuration control for each "scan plan" / procedure is established and fixed. Installation of the procedure on other machines requires only validation of the REPRODUCIBILITY of outputs during system "calibration" and verification of the REPEATABILITY of output responses to the same flaws or artifacts on a subset of test specimens and the master-gage slotted component(s).
The same data can be used as indicators that some element of the automated scanning system requires repair, rework or replacement. Since the background (NOISE) data from the slotted artifact specimen should remain constant, a variance or trend in offset of the "NOISE" can also be used as an indicator for system adjustment and/or repair. Conversely, an offset of "NOISE" data from production components can either indicate a need for system adjustment or repair, or a change in the usage or age of the production components.

SUMMARY:

Modern demands for incorporating inspection as a part of the life-cycle management of component structural integrity requires quantification of inspection capabilities. Capability characterization of automated inspection processes is necessary as a part of inspection procedure validation. Probability of detection (POD) is the classical method of characterizing inspection CAPABILITY an inspection procedure and is considered to be an essential tool in nondestructive inspection (NDI) engineering. Once the capability of a procedure has been established, additional systems may be qualified by sub-set measurements to assure that the same values are produced in response to known artifacts (for example, cracks or slots). Such measurement provide confidence in the REPRODUCIBILTY of an additional system. In like manner, the automated scanning procedure should provide the equal responses to known artifacts (cracks and slots) at critical locations in the "master gage" component. Such measurements provide confidence in the REPEATABILTY of an additional system. Together, the REPRODUCIBILTY and REPEATABILTY provide confidence in the RELIABILITY of additional systems operating at the validated system CAPABILITY level. The same measurements provide confidence in continuing system RELIABILITY and in restoration of a system after repair, rework or replacement of significant components.
The logic and methods discussed have been applied successfully to a precision automated eddy current inspection system for gas turbine engine components (RFC) during the past 15 years and provides the basis for assessing and validating system improvements and upgrades as new inspection requirements and challenges are met.

© AIPnD , created by NDT.net

|Home|    |Top|