|NDT.net - November 2002, Vol. 7 No.11|
This paper outlines our effort to develop an automated eddy current method assisted by a self-aligned manipulator scanning along the disk rim of Westinghouse turbine. The light-weighted manipulator was mainly designed for inspecting blade on L-2 stage disk where root cracks were most frequently found during ISI. A mock-up with changeable blades was used to assess the performance of this technique by statistical method. Experimental results showed that the POD of this technique agreed well with that originally proposed. The technique was also proved feasible in field trials.
Blades and disk rim of turbine were frequently subjected to fatigue cracking and resulted in unscheduled shutdown and even more a total failure of turbines. Hence, reliable routine inspection is crucial to operation safety and efficiency of turbines operation. Although various NDE techniques have been applied in ISI for this region for years, continue improvement of the inspection techniques is still desired due to the critical role of steam turbines and the inspection difficulties in this region. Among which, MT and manual ET have been the favorite methods.[1-8]
There is no statistical data to refer to how small the crack size can be detected by the current inspection methods. In principle, it was not difficult to detect 0.5mm crack with fluorescent MT or manual ET methods. In practical cases, since the gap between disks was quite narrow, it only allowed the inspector to observe the MT results from a distance or to stretch his arm to scan the hand-held EC probes. Hence, MT will easily miscall the tight and tiny cracks or those close to the edge of the serration. Inconsistency between MT and ET results was always found especially for cracks smaller than 1 mm. To minimize the human error, an automated EC inspection method was then developed as a supplement to the current ISI. Crack as small as 0.5mm exposed on end face was set as the target to detect with this method. And, Westinghouse type blade as shown in Fig. 1 was the inspection target of this research.
|Fig 1: Westinghouse turbine blade.|
The main goal of EC inspection automation was to develop a scanner-assisted method to inspect blade root cracking on L-2 stage of Westinghouse steam turbine. The scanner was designed to run on the extruded part of disk rim of Westinghouse steam turbine L-2 stage as its rail, and use L-1 stage disk rim as its supporting face of the scanner. Accordingly, the scanner can run along the disk in circumferential direction between adjacent disks (as shown in Fig. 2). The scanner could be stopped and locked automatically at proper position of each blade to perform following inspection by scanning probe along the tangential of blade root corners where cracks could be present as shown in Fig 3. After the inspection on each blade, the scanner would be moved to next blade and all the blades will be inspected in such a process.
|Fig 2: Automated scanner for blade crack inspection.|
|Fig 3: Different scanning routes (left) and relevant signal responses (right).|
The probe was specially designed with a set of leaf spring to produce consistent contact between probe and inspection area of the blade. The coil diameter of probe was of 3mm with working frequency of 1-2 MHz that proved sensitive enough to obtain the result not being interfered by accompanying unnecessary signal especially from corners of steeple. Practical scanning route was adjusted to have the best performance of the inspection signal during initial setup. Fig. 3 shows an example of signal on impedance plane from various scanning routes on a real 13mm root crack. Lift-off signal in horizontal direction and crack signal were almost perpendicular. When the scanning route running further from the common tangent the crack signal pattern was thin and long which could be easily differentiated between lift-off signal and noise. When the route running closer to the groove edge the crack signal was influenced by the edge effect to deform, closer the probe to the edge more obvious the influence of the edge effect until the crack signal was finally replaced by the edge and lift-off signal totally.
Performance demonstration is a crucial procedure for any NDE method to meet the inspection requirements. It was normally carried out on real components or mock-up and, sufficient number of tests is required to produce a quantitative and objective assessment.
In principle, when a NDE method or personnel is to inspect a defect of critical dimension not all the defects of the same dimension could be detected. Similarly, repeated inspections on a specific defect would not detect the defect every time. Although automated inspection could avoid most of the human errors, statistic assessment is thus required to assure the performance of the NDE method.
While assessing any NDT system and method statistically, various controllable and uncontrollable factors need to be defined. In our inspection method, the controllable factors include eddy current excitation frequency/ Gain/ Phase/ Filter, probe, scanner initial setting/ positioning, probe scanning position/ orientation/ speed, defect size/ position/ orientation, whereas main uncontrollable factors include the stability of eddy current system, probe contact/ wear, circumferential movement of scanner/ positioning/ speed, blade installation/ surface condition/ defect type and human factors.
The POD has been found properly represented by POD(a) function. As shown in Fig. 4, the most common used function was cumulative log normal distribution function or log odds function and the parameters could be calculated by Maximum likelihood method. Such a so-called â analysis method could allow fewer samples, says 30 samples is reasonable. Of course more the sample number higher the confidence level and more the assessment precision.
|Fig 4: POD curve.|
|Fig 5: Example correlation between measured value â and defect size a.|
Basically, â analysis method uses a measured value â to correlate with real defect size quantitatively as in Fig. 5. Such a relation requires that:
In â is the relative measured value of defect,
In (a) is the relative value of defect,
d is the random error and is a normal distribution,
b0 + b1In(a) is the regression of average of each distribution.
To meet the number of defects required for experiment, we used notches as artificial cracks on blades for evaluation. Fig. 6 showed defect signal of notch and real crack were recognizable. The only differences between them were phase and amplitude. The signal characteristic as shown suggested that the notch can be used to simulate crack. To test the mechanical performance of the scanning system, 9 pieces of cracked blades were used on a mock up at each experiment as shown on Fig. 7. Scanner was installed on the mock up to perform simulated scanning, each blade could be replaced easily according to the requirement of random sampling. The cracks on the blades were simulated by EDM notch with length between 0.3-2.5mm.
|Probe type UNIWEST RPDS-12 differential|
|Parameter Settings||Test sample Real blade with EDM notch|
|Frequency 2.0 MHz|
|Bandwidth HF||Scanning root area of blade|
|Preamplifier 6 dB|
|Gain Y/X 17dB / 17dB|
|Phase 119 deg|
|Dot Position Y / X 0 / 0|
|Filter LP 80Hz / HP 0.5Hz|
|Fig 6: Signal response of 2mm×0.3mm (length×width) notch on sample 1.|
|Fig 7: Mock-up with 9 blades interchangeable.|
The flowchart of POD analysis was shown in Fig. 8. Relative dimensions (Di) of a certain notch could be defined with reference to a specific sample with notch 2.5mm in length as in Eqa (4). The signal amplitude(Ai) of a certain notch on impedance plane (Fig. 9) was normalized by the maximum defect amplitude(A2.5) of the reference notch.
|Fig 8: Flow chart of POD determination.|
|Fig 9: To determine defect size from reference signal.|
Two charts of distribution of various defect detections were shown in Fig. 10, then to transfer the individual signal by natural log. The range and variance of defect length were first shown in table 1. Theoretically, all the distributions should be normal which could be verified by the Bartlett test by confirming that the variance equivalence of various test data was satisfied where the significance a level was assumed as 0.01. Then critical value of Bartlett distribution (bk(a , n1, n2, n3, , nk)) was then found as
The mixed estimate of sample variance was
b was found as
Since b > bk, we may assume the variance of all different defect size were equal which suggested that all distribution were normal.
Distribution of In â vs. In(a) was found as shown in Fig. 11. By regression, it was found that intercept (b0)= - 0.0299 and slope (b1)= 0.8784. The freedom (n) for 6 group of different defect size was 5. To assume confidence level = 95% and set a = 0.025, it was found that
Furthermore, to assume
It was found that
The POD curve could be drawn by finding POD(a) of every defect size as shown in Fig. 12.
|Measured Length (mm)||
|Table 1: Experimental data of signal response method.|
Fig 10: Measured value distribution of various defects.
|Fig 11: Correlation between measured value â and defect size a.||Fig 12: POD curve of automated EC inspection.|
The signal response analysis was used to establish the practical reliability analysis method for this automated eddy current inspection system. The probability of detection curve was shown as a unit step function which indicated the quality of the system was good enough. When defect length was greater than 0.5mm the POD was almost 100%, in another word the defect greater than 0.5mm in length could be detected reliably, which was compliant with the originally proposed target. However, the result was obtained from a mock up in a well-controlled experimental environment then further test should be carried out in the field. Through the result this method could be verified whether or not it could be an alternative of the current inspection method.
The authors gratefully acknowledge all their colleagues in Taipower Company in supporting this program.
|© NDT.net - email@example.com|||Top||