Grant Skennerton (1), Afshin Abedi (1), Robert G. Kelly (2), and John P.
Wikswo, Jr. (1)
(1) Department of Physics and Astronomy, Vanderbilt University, Nashville, TN 37235, USA
(2) Department of Materials Science and Engineering, University of Virginia Charlottesville, VA 22903, USA
We have demonstrated the ability of Superconducting QUantum Interference Device (SQUID) magnetometers to qualitatively monitor electrochemical activity within lap joints removed from actual aircraft. This work lays the foundation for quantitative measurements of corrosion rates within occluded structures. In this study, a SQUID magnetometer was used to examine hidden corrosion activity within lap joints removed from ageing KC-135 aircraft. During exposure, we imaged each specimen every 60 minutes. We designed the protocol to expose each lap joint specimen to a range of increasingly corrosive exposure sequences. The results obtained from this protocol show that there is a clear relationship between corrosivity of the environment and the activity generated within the lap joint. Over a range of increasing corrosive environments, the activity generated shows significant increases with each step. This technique allows us to determine non-invasively the corrosivity of the chemical environment already existing inside the lap joint relative to environments of known corrosivity.
Keywords: SQUID magnetometers, lap joints, corrosion measurement
Corrosion of aircraft aluminium is a major problem for the United States Air Force and a great deal of effort is being applied to incorporate its effects into aircraft maintenance and structural integrity decisions. Models used for such decisions require accurate values of the rate of corrosion within occluded regions, such as lap joints, in a variety of operational environments. The overall goal of this project is to ascertain the rate of hidden corrosion in lap joints specimens removed from ageing aircraft under controlled environmental conditions.
A limitation common to the existing corrosion measurement techniques is their inability to probe accurately the instantaneous rate of corrosion within lap joints or beneath thick layers of paint or specialised military coatings. This barrier has prevented incorporation of corrosion into the Aircraft Structural Integrity Program (ASIP). As a result, studies of these phenomena typically require the creation of redundant samples that are examined destructively at regular intervals throughout the duration of test. Until now, instantaneous rates of hidden corrosion, their spatial distribution within a structure and how the environment affects those rates could not be determined by existing techniques.
Unlike the variety of techniques presently utilised, such as atmospheric exposure, electrochemical techniques and NDE techniques, SQUIDs possess the ability to measure rate instantaneously, and do not need to integrate corrosion damage over long periods of time to determine corrosion rates. Preliminary work at Vanderbilt [1-4] using SQUIDs to monitor currents from either exposed or hidden corrosion demonstrated the SQUIDs' utility for the periodic, non-destructive analysis of corrosion test specimens for which the corrosion activity is not directly accessible. Their unparalleled sensitivity, spatial resolution (microns to millimetres), and DC-to-10 kHz bandwidth allows SQUIDs to map steady or time-varying currents deep within a sample without having to make contact with the sample. In addition, SQUIDs do not require visualisation of or other direct access to the metal surface that is undergoing corrosion.
We have developed instrumentation, recording, and analysis techniques that are optimised for the measurement of the spatial and temporal dependence of the magnetic fields associated with hidden corrosion in lap joints. Our two corrosion mapping systems [5,6] use SQUID gradiometers and each include a magnetic shield, a scanning stage with a custom sample-handling system, a computer control system, and custom scanning and analysis software.
Four aluminium lap joint specimens were obtained from retired KC-135 aircraft. The specimens were approximately 250 mm long, had three fastener (rivet) rows, and were three fasteners wide (plus half the fastener pitch on each side). The majority of the free-standing stringer material was cut off. The total height of each specimen at the stringer was approximately 10 mm. All exposed metal surfaces were painted with MIL-P-23377 epoxy primer and MIL-PRF-85285 polyurethane top coat. The paint was applied with a paint spray gun in the same manner the aircraft are painted. The edges were sealed with low-adhesion polysulfide sealant so hydration could only occur from the two mouths of the lap joint. Holes were drilled and countersunk in the ends of the lap joint for accommodating 4-40 flat-head nylon screws as shown in Figure 1. These holes were used to fasten the lap joints to the specimen chamber.
Figure 1: Schematic diagram of a typical lap joint specimen.
We designed environmental chambers that allowed us to expose lap joints to a variety of environments while we scanned them under the SQUID. After attachment of the samples and baseline SQUID scanning, we introduced the various gaseous or liquid environments to the chamber. We did not remove the lap joints from the chamber or from the magnetic shield until the test was completed.
To determine the chloride-dependence of corrosion within the lap joint, and to examine experiment repeatability and cycling effects, we designed a protocol that exposed each lap joint specimen to a range of increasingly corrosive exposure sequences. These sequences are shown in Table 1. Each sequence, or "step," was six days in duration. We applied three passes of each step to each specimen before progressing to the next step in the protocol. At the beginning of each pass, we baked the specimen at 50�C in a vacuum oven for three days to dehydrate the sample and halt any ongoing corrosion. Prior to scanning, we degaussed the specimen in order to provide a background level that was repeatable over multiple passes. To quantify the background magnetic field of the lap joints, we scanned samples in dry nitrogen gas for 24 hours. We then exposed the specimen to a corrosive environment for 48 hours. We intended that the corrodent used in each subsequent step was more corrosive than the one used in the previous step. During step 1, we exposed specimens to humid air. During step 2, distilled water was used to provide a measure of the corrosivity of the material already present in the lap joint. Steps 3 and 4 used 0.01 M Cl- and 0.1 M Cl- respectively. We only completed one pass of step 4.
Table 1: Exposure sequence of lap joint protocol III. We repeated each step twice.
|1||Vacuum bake (3 days), degauss, dry N2 (1 day), 98% RH (2 days)|
|2||Vacuum bake (3 days), degauss, dry N2 (1 day), distilled water (2 days)|
|3||Vacuum bake (3 days), degauss, dry N2 (1 day), 0.01M NaCl (2 days)|
|4||Vacuum bake (3 days), degauss, dry N2 (1 day), 0.1M NaCl (2 days)|
We define several useful magnetic field metrics in a previous paper . Briefly, the background magnetic field image, BG, is defined as the spatial average of K dry scan images (raster scans) at a fixed height above the dry (non-corroding) sample.
Here Bz represents the average magnetic flux density measured over the area of the z-axis pickup-coil of the magnetometer, xi and yj represent the position of the (i,j)th pixel with respect to the scanning co-ordinates, K represents the number of dry scans to be averaged, tk denotes the starting time of the kth scan and is measured with respect to the beginning of the first scan of the experiment, and Dt represents the time it takes to finish a single raster scan and is a constant for all the scans taken in one experiment. The notation in this and the following equations contains Dt in order to remind the reader that we acquired each raster image in a time interval [tk , t+Dt].
After recording a stable background, we introduced the corrosive solution to the sample. The resulting change in magnetic field with respect to BG we refer to as Bcorr and define (2).
where k represents the kth magnetic field image. The Magnetic Activity (MA) of the sample, recorded in the kth scan of the experiment, is defined as the absolute value of (2), and represents the absolute change in magnetic field due to corrosion with respect to the non-corroding (dry) background image.
Furthermore, we define the Spatially Integrated Magnetic Activity (SIMA) as
where we take the summations over all the pixels in the kth image, and Dx and Dy represent the distance between two consecutive pixels in the corresponding scan direction. SIMA(tk,Dt) is a number proportional to the net magnetic activity of the sample as recorded during the kth scan, and should be proportional to the net corrosion activity of the sample during this time-interval (i.e., [tk , tk+Dt]). The graph of SIMA(t,Dt) versus time represents the time-course of corrosion activity of the sample.
Since SIMA(tk,Dt) is proportional to the net corrosion current (and hence the mass-loss rate) in the kth image, then the area under the curve SIMA(t,Dt) versus time must be proportional to the total mass-loss of the sample during the corrosion experiment. For an experiment consisting of K raster scans, we define this quantity as the Temporally Summed SIMA (TSSIMA):
In the case where SIMA is proportional to mass-loss rate, TSSIMA should be proportional to the total mass-loss of the sample during the experiment.
Finally, to determine the cumulative magnetic activity at each pixel for the course of the experiment, we define the Temporally Summed Magnetic Activity (TSMA) at point (xi,yj) as
where T is the length of time that the sample was exposed to corrosive solution. Therefore, TSMA(x,y) is a two-dimensional image, wherein each pixel is the time-weighted cumulative measure of the magnetic activity of that pixel throughout the experiment, and hence will spatially identify the regions with differing net magnetic activity.
Before we computed final quantitative metrics, we processed the data to minimise the influence of magnetic contamination on the calculated corrosion activity. We observed localised regions of high magnetic activity within the scan's area, and the time-course of magnetic activity in these regions was often different from the time-course of activity in the rest of the corroding sample. The spatio-temporal property of this activity is consistent with behaviour we have observed for the corrosion of localised ferro-magnetic contaminants. As these localised regions were not necessarily within the area of the specimen, it cannot be assumed that these signals are associated with localised distribution of corrosion current sources. Because of this, we removed the magnetic contribution of these localised particles from that of the corroding samples in order to have proper measurements of the corrosion-related changes in magnetic activity.
We used a post-processing technique based on temporal-correlation to distinguish and then remove the contaminated regions from the data. Once we identified a pixel as exhibiting time-dependent magnetic contamination, we correlated the time-course of that pixel's magnetic activity with the time-course of all other pixels in the image. The correlation coefficient, S, of the contaminant pixel (A) with another pixel (B) is:
All pixels whose cross-correlation with contamination exceeded a predetermined threshold were eliminated from the calculation of quantitative metrics. Using this user-controlled threshold limit, we generated a mask and applied this to the data. We zeroed the magnitude of the magnetic activity of pixels with correlation coefficients greater than or equal to the threshold limit. We did not change the magnitude of activity of pixels with correlation coefficients less than the threshold limit. Though some loss of data is unavoidable, the post-processing technique has proved to be a viable technique for studying magnetically contaminated samples. In the future, we expect to improve the method used for identifying and eliminating contaminated areas from the data.
This procedure is demonstrated in Figure 2. Figure 2(a) is a TSMA image obtained from lap joint SM11-SP1. We identify region A as one typical of contamination, while we identify region B as one typical of the lap joint's true activity. Figure 2(b) shows the magnitude of the dipolar contamination source along section X-X. The dipole is clearly much greater in magnitude than other areas along section X-X. Figure 2(c) shows the time-course of the magnetic activity of region A. Figure 2(d) shows the time-course of the magnetic activity of region B. Comparison of Figure 2(c) and 2(d) shows the marked difference between the contaminated and uncontaminated reigions, both in magnitude and nature of the magnetic activity. Figure 2(e) shows the mask generated by an 80% cross-correlation of the data with region A. Once the mask is applied, Figure 2(f) shows the magnitude of the dipolar contamination source along section X-X after masking. Comparison of Figure 2(b) with 2(f) demonstrates the success of this masking technique in removing the signal of a contaminant source from the data.
Figure 2: (a) The TSMA image from lap-joint SM11-SP1. We identified region A as contamination. We identified region B as a region typical of the lap joint's activity. (b) The magnitude of the dipolar contamination source along section X-X. (c) The time-course of magnetic activity in region A. (d) The time-course of activity in region B. (e) The mask generated by an 80% correlation of the data with region A. (f) The effect of masking on the region A along section X-X.
Our statistical analysis of the data involved a test of the significance on the mean . In all cases, we hypothesised that environmental changes under either protocol would produce no significant change in magnetic activity, i.e., the ratio of the activity before the transition to the activity after the transition would be unity, or 1 (m0 = 1). Thus our null hypothesis can be written: H0: m = m0, i.e., the population mean (m) was expected to be equal to m0. Each protocol was analysed independently. The selected significance level was 1%, or a = 0.01. In testing the significance of the mean, we reject the null hypothesis if the probability (P) of the observed difference between m and m0 is less than a (0.01), which is to say that m is not compatible with m0. In such a case, variation in the sample population is not able to explain the difference between the null hypothesis value and the sample values. However, should P exceed the significance level of 0.01, we expected that there is no significant difference between m and m0.
We examined four KC-135 lap joints in Protocol III (Sample 11 Specimens 1 through 4). SIMA versus time curves obtained from each are shown in Figures 3 to 6 respectively. In each step, upon introduction of the corrodent at 24 hours, the magnetic activity increased. For step 1 (exposure to 98% RH air), the new level reached is constant over the 48 hour period when compared to the levels reached in steps 2 to 4 (exposure to solutions). We observed transients in the level of activity that did not coincide with changes in the corrosive. We observed this behaviour in specimens exposed under earlier protocols (unpublished).
Figure 3: SIMA versus time curves obtained from KC-135 SM11-SP1.
Figure 4: SIMA versus time curves obtained from KC-135 SM11-SP2.
Figure 5: SIMA versus time curves obtained from KC-135 SM11-SP3.
Figure 6: SIMA versus time curves obtained from KC-135 SM11-SP4.
For each lap joint, we calculated the average SIMA for the 24-hour period that we held the specimen in dry N2. We also calculated the average SIMA for the 48-hour period that we exposed the specimen to corrodent. Figure 7 shows the ratio of the average SIMA of the corrosive environment to the average SIMA obtained during exposure to dry nitrogen averaged over all lap joints for each step. We hypothesised that the corrodent would have no significant effect on the magnetic activity within the lap joint, i.e., that there would be no increase (H0: ratio = 1). We tested our null hypothesis using the populations of SIMA ratios obtained in each step. Table 2 shows that introduction of corrodent produces a statistically significant increase in activity in the lap joints (P<0.01). As shown the level of activity achieved during corrosion in each step increases relative to the earlier steps. During step 2, the level of SIMA attained during exposure to the corrodent was 2.6 times greater than that attained in step 1, step 3 2.1 times greater than step 2, and step 4 1.4 times greater than step 3.
Figure 7: Average increases in activity upon humidification/immersion of KC-135 SM11 lap joint specimens. The figure shows the inter-step increase in SIMA.Table 2: Average increases in activity upon humidification or immersion of KC-135 SM11 lap joint specimens, and the statistical significance calculated for this increase. H0: ratio = 1, a = 1%.
|SIMAHUM||SIMADW||SIMA0.01 M Cl-||SIMA0.1 M Cl-|
Table 3 shows the average TSSIMA obtained during each step for each lap joint. Figure 8 shows an upward trend in activity over the four passes. These data are consistent with the relationship seen in the SIMA data, discussed above. Figure 9 shows the average TSSIMA obtained from each step. This figure shows a clear upward trend in TSSIMA as the environment becomes more corrosive. However, the increase in activity between steps 3 and 4 is not as pronounced. We tested the significance the increase in TSSIMA seen between steps. We hypothesised that there would be no significant increase in TSSIMA, i.e., the ratio would be unity (H0: ratio = 1, a = 1%). Table 4 presents the ratios of TSSIMA from step to step and the significance of this increase. As can be seen, the increase in activity from step 1 (98% RH) to step 2 (distilled water) is significant, as is the increase from step 2 to step 3 (0.01 M Cl-). The significance levels for these ratios are <0.0001 and 0.0044 respectively, which are both below our significance level of 0.01. The ratio of TSSIMA between step 4 and step 3 is only 1.1, or an increase of 10%. As expected, this increase was calculated to not be significant (P = 0.32). However, as we took only four data points in step 4, we cannot assume this last observation to be definitive.Table 3: TSSIMA data from KC-135 lap joint specimens averaged over each step. The table also shows the average TSSIMA for all four lap joints for each step. (TSSIMA � sTSSIMA (mT�mm2�hr))
|Specimen||Step 1||Step 2||Step 3||Step 4|
|SM11-SP1||8 � 4||23 � 13||37 � 5||51|
|SM11-SP2||11 � 2||28 � 10||39 � 4||63|
|SM11-SP3||9 � 2||34 � 15||52 � 19||35|
|SM11-SP4||6 � 1||29 � 14||42 � 3||35|
|All||8 � 3||28 � 14||43 � 12||46 � 12|
Figure 8: TSSIMA data of KC-135 lap joint specimens averaged over three passes.
Figure 9: Average TSSIMA data of each step of Protocol III. Average inter-step increases in TSSIMA of KC-135 SM11 lap joint specimens, and their probability (H0: ratio = 1, a = 1).
We designed a protocol to determine the effects of moisture and chloride on lap joints relative to the initial condition of a dry, inactive joint. We exposed lap joint specimens to what we anticipated was a range of increasingly corrosive exposure sequences: humid air, distilled water, 0.01 M Cl-, and 0.1 M Cl-. In between each exposure, we effectively "reset" the specimens by baking them in a vacuum and degaussing before the subsequent exposure. We show that the SQUID has sufficient sensitivity to detect corrosion activity within the joint upon exposure to only humid air. By then immersing the lap joints in distilled water, we have observed the corrosivity of the chemical environment existing inside the lap joint. We show that distilled water generates significantly more activity than humid air.
Upon exposure to 0.1 M Cl-, we show chloride to generate significantly more activity than the natural conditions with the lap joint, as brought out by the distilled water. However, there was no statistically significant difference between the activity generated in 0.01 M Cl- and that generated in 0.1 M Cl-. Hence we conclude that SQUIDs are capable of making quantitative measurements of the relative rates of corrosion of aircraft lap joints under differing conditions of humidity and chloride.
We have yet to obtain the calibration of mass-loss to magnetic-signal for lap joint specimens. Although we have established a quantitative calibration between magnetic signal and corrosion rate for other geometries, its use for lap joints is currently under study. Hence, we present a magnetic measurement that can be used to compose the corrosion rate under different circumstances, but it is not yet possible to use these data for quantitative assessment of absolute corrosion rates. In the future we will be conducting experiments to determine the mass-loss to magnetic field calibration for specimens of lap joint geometry.
Although magnetic contamination had been previously thought to be a limiting factor in the use of SQUIDs to study corrosion activity in aircraft samples, we show that the problems of magnetic contamination can in general be overcome by selection of samples without ferromagnetic fasteners, modest care in sample preparation, thorough degaussing of the samples at the beginning of each step in the exposure protocol, subtraction of time-independent background fields, and masking of the regions of the magnetic images that show a time course consistent with the corrosive activity of ferromagnetic contaminants. While work will continue to refine this technique, we demonstrate clearly that high-quality corrosion data can be obtained from samples removed from aging aircraft.
This work was funded in part through Air Force contract F09603-97-C-0050, "Corrosion/Fatigue Effects on Structural Integrity," issued through the Air Force Corrosion Office (AFCO) at Robins Air Force Base in Warner Robins, GA, NCI Information Systems, Inc., and Air Force Research Laboratory (AFRL). We thank Garth Cooke and Jim Suzel (NCI), Deborah Peeler (AFRL), and Richard Kinzie (AFCO) for their comments and suggestions regarding the experimental design and these measurements. We thank Steve Kaldon (NCI) for provision of the lap joint samples.