Interlaboratory Variability of Slip Coefficient Testing for Bridge Coatings

EXPERIMENTAL RESULTS

It is important to note that it was found that all labs involved in this test program followed the specification testing procedures as dictated in the RCSC specification. The only exception was that lab 3 did not use a drilled-out nut on one side of the specimen as dictated by the RCSC specification. The testing arrangement and procedures used at each lab, although slightly different, were all found to fall within the letter of the specification. Therefore, data were found to be valid for direct comparison.

The raw data from each lab and coating in terms of the slip coefficient are reported in table 2 using the existing RCSC failure definition of either peak load or load at 0.02 inch of slip. The main body of this report presents only a limited statistical analysis. The ASTM E691-13 specification dictates a variety of statistical measures that need to be reported as part of an interlaboratory variability study.⁽³⁾ Since the test matrix did not strictly follow the guidance of ASTM E691-13, those statistical measures are of limited use. Regardless, all the ASTM E691-13 statistical measures are presented in appendix B for reference, though their exact results should be interpreted carefully.

Table 2 also reports the average and coefficient of variation (COV) for sets of five samples that were tested by each lab and for each coating. Figure 7 displays all the slip data for each coating and lab. Each bar in the graph has error bars that are symmetrically plotted based on the COV value. The load versus slip plots for every specimen, from each lab, and each coating can be found in the appendix A. Two of the labs recorded data on analog X-Y plotters, and scans of the plotter paper had to be manually encoded into a spreadsheet so uniformity of the plots could be shown. It was universally interpreted by the labs that the intent of the RCSC specification was to assume that zero slip had occurred when 1.0 kip of vertical load was applied to the specimen. Therefore, some of the datasets strictly began from a point of 1.0 kip of load and zero slip displacement, and other labs continuously recorded the data and manually subtracted off the displacement at 1.0 kip when determining failure load using the 0.02-inch displacement criteria. For uniformity of the data presentation in the appendix, for the two labs that recorded all the data, their curves were shifted in the X-direction such that the curves intersected the y-axis at 1.0 kip.

Table 2. ResultsTable of slip coefficient testing considering existing RCSC failure criteria.

* The hole edge distance was out of tolerance on one plate, and the specimen could not be tested.
** The loading rod broke, and the last two specimens could not be tested within the designated 24-h period.
Note: Blank cells indicate that lab 4 did not participate in the coating series.

Figure 7. Graph. Mean slip coefficient for each coating system tested at each test lab using existing RCSC failure criterion.

From the average calculated slip coefficients for each primer tested, it is clear that different coatings have different slip coefficients. In the aggregate for organic zinc-rich primers as a class of coatings, those values appear to straddle the class B specification value of 0.5. For the five organic zinc-rich primers that were tested, coatings B and C exceeded class B slip resistance for all the labs that participated. Coating E unanimously qualified only as class A by all labs that participated. Coatings A and D were classified as class A or B depending on the testing lab. Coating A was classified as class A by lab 2 and as class B by the other three labs. Coating D was classified as class B by lab 3 and as class A by the other labs. Therefore, it is apparent that the differences inherent in the test approaches used by each lab can be important to the pass or fail result for a specific coating.

VARIATION IN RESULTS LAB TO LAB

In comparing the mean slip values as shown in figure 7, there is no discernable trend of one lab consistently producing higher or lower numbers than the other labs. However, the error bars for lab 2 were consistently much larger than the other labs. When the COVs were averaged across all the coatings tested by each lab, it was found that labs 1-4 had an average COV of 0.044, 0.188, 0.040, and 0.040, respectively. Clearly, the data indicate that variability is quite consistent between labs 1, 3, and 4 and different in lab 2.

The load versus slip displacement plots in appendix A are useful to further understand the differences between the labs. The plots from lab 2 demonstrated a very "soft" response where there was more slip displacement per given load than the other labs (e.g., the secant slope of the load versus slip curve from the initial load to the peak load is not as steep as the other labs).Lab 3 had a very "stiff" response with some plots, demonstrating almost no slip as the load increased (e.g., the secant slope of the load versus slip curve from the initial load to the peak load was nearly infinite at times). The three plots shown in appendix A for coating A1 highlight this the best, where the peak loads actually did not change much between the three participating labs. However, because of the "soft" response from lab 2, the load at 0.02-inch of slip controlled the failure criterion (per RCSC) for that lab, leading to failure loads much less than the peak load and contributing to their higher level of scatter.

The observers tried to discover why labs 2 and 3 produced peak loads that were similar but had vastly different slip displacement responses. It is a nuance, but careful inspection of figure 3 through figure 6 show that labs 2 and 3 only used one displacement transducer to monitor the slip displacement. Since the RCSC specification requires that a spherical platen be used to load the specimens, there is a chance that the middle plate of the specimen can rotate about the loading rod during the test. Measurement errors from only one displacement transducer would either be additive or subtractive depending on the rotation of the middle plate. This explains the soft and stiff responses of lab 2 and 3. The displacements measured by lab 2 also contained the fictitious displacement of the spherical platen rotating towards the transducer, whereas it rotated away for lab 3. In fact, this is why some of the plots from lab 2 tended to hook backwards as if slip decreased as load increased (see figure 19, figure 21, figure 22, and figure 32 in appendix A). It also explains some of the peculiar results attained by lab 2, in particular the abnormal responses for specimens B2-3, D1-2, and D2-4. To highlight the notion, the load versus slip data recorded from each LVDT used by lab 4 for specimen C2-1 is shown in figure 8, which shows that the response from each of the two displacement measurements were quite different. In terms of the RCSC failure criterion for this example, if only LVDT 2 was used, the peak load controlled failure, and the failure load was 60.9 kip. If LVDT 1 was used, the 0.02-inch slip criterion controlled failure, and the failure load was 46.9 kip. Therefore, the coating was classified as class A using only LVDT 1 and class B using only LVDT 2. This highlights the need for two displacement transducers to be averaged together to be used in lieu of one. Otherwise, the use of one transducer must be referenced to rigid (non-rotating) points on the loading system.

To further show the effect of variability using one displacement measuring device, all data were reanalyzed ignoring the 0.02-inch slip failure criterion, and the slip coefficients were calculated using just the peak loads. The slip coefficient data based on just the peak load response are shown in table 3 and figure 9. When looking at just the peak load data, for many of the coatings where lab 2 deviated considering the 0.02-inch criteria, the average was closer to the other three labs as well a reduction in scatter. This further indicates that the measurement technique of slip displacement was the major factor in variability between labs.

Figure 8. Graph. Load versus slip displacement response of two LVDTs from lab 4 for specimen C2-1.

Table 3 . Results of slip coefficient testing considering just peak load failure criteria.

* The hole edge distance was out-of-tolerance on one plate, and the specimen could not be tested.
** The loading rod broke, and the last two specimens could not be tested within the designated 24-h period.
Note: Blank cells indicate that lab 4 did not participate in the coating series.

Figure 9. Graph. Mean slip coefficient for each coating system tested at each test lab only using peak load response.

VARIATION DUE TO PAINT THICKNESS

The RCSC procedure requires reporting the coating DFTs. However, for this study, that information could lose the anonymity of the five specific coatings. Therefore, table 4 shows the average deviation of the DFTs from the target thickness considering all the test plates used for each series. A negative deviation represents a DFT thinner than the target and vice versa for a positive deviation. The control of the application relative to target application was within 1 mil for all but one of the sets of panels. The table also reports the difference between the average DFTs of the +1- and +2-mil targets (i.e., the difference should ideally be 1 mil). For coatings A-E, the differences were actually 0.9, 1.7, 0.7, 2.7, and 0.2 mil, respectively.

The far right column in table 4 reports the difference in the average slip coefficient between the +1- and +2-mil specimens using lab 1 data only. The data from labs 2 and 3 were not used because of the disparity noted previously with the slip measurement. Data from lab 4 were not used because the lab did not test all coatings. However, based on lab 1 data, thickness variations ranging from 0.2 to 2.7 mil caused no appreciable change in slip coefficient. The testing of manufacturer recommended thickness specifications of +1 versus +2 mil (as required in the RCSC specification) for each of the coating systems did not seem to be the determining factor in whether a coating qualified as class A or B.

Table 4 . Deviations from manufacturer’s recommended DFT.

OUTLIER ANALYSIS

During the study, there was some discussion of the proper treatment of outliers in the datasets. Although there were several replicates which appeared to have a high variance from the mean, an analysis of the dataset did not point to any single data point as a true statistical outlier. That is, no data point showed an excursion of two standard deviations from the mean for the set. This result is in part a result of the fact that the mean for each dataset was only generated from a set of five replicates. It also indicates that for those datasets where an apparent outlier existed, there was generally a larger standard deviation for the dataset.

However, regardless of statistical significance, it is clear that within some of the datasets, there were replicates which provided results that conflict with the other four replicates. Given the care to maintain consistency in preparation, application, and cure of the panels used in this testing, there is no apparent physical reason (related to the paint) for extremely low slip results for a single replicate. Therefore, it must be assumed that there is an occasional result which deviates from the group due to slight but important differences in testing protocol (loading rate, alignment, equipment function, etc.). For this reason, it is worth considering instituting a protocol which allows some measure of judgment on the part of the test agent to either retest extra, duplicate panels, or dismiss a single data point when calculating slip coefficient. Given the fact that there were only five replicates in this test protocol, the use of a sixth panel as a replacement for a questionable data point may be a prudent testing option to consider. Of course, this substitution should require a substantive justification from the testing agent. Out of the 175 total replicates tested by the 4 labs in this program, a total of 3 (lab 1 C2-4 and lab 3 E2-4 and E2-5) were not reported due to testing difficulties. But there were other points that were reported that may have been considered questionable (primarily low) by testing agents if that option had been available. The inclusion of a sixth sample would be a small incremental (possibly negligible) cost and would allow reporting of five results with more confidence. Specifically, this approach could potentially be applied in cases such as lab 1 D2-2 and D1-3 and lab 2 D2-2, C1-1, and A1-3 which all show exaggerated low load profiles compared to the other replicates in their respective sets.

OTHER IDENTIFIED CONCERNS

The operation of the test in quasi-load control (labs 2 and 3) was perfectly acceptable under the specification; however, this mode of control creates some practical issues that make the test somewhat more difficult and may introduce some of the errors seen in the results. The use of load control requires the operator to focus his or her primary attention on the load control mechanism to capture the end result of the test while still attempting to "catch" the end point and quickly unload the test apparatus prior to full dramatic slip of the coating. This phenomenon is clearly a focus point of the test operators in both of these labs, and it requires a majority of their attention while also focusing on maintaining a consistent lateral compressive load and visually capturing the highest slip load on their readout. Without digital data recording and a fail-safe device on the very high slip load, this is a large responsibility to ask of a single operator.

Additionally, the periodic crashing of the slip load onto the rod has real consequences. Three of the four labs involved in this study had rods break during this test program. The specific causes of these breaks involved various causes, from improper heat treatment of one rod to improper material selection of another rod. Breaks were also due to repeated (unintended) impact loading of the rod with the slip load. Other rods did not break but were replaced due to significant bending. Bent rods are another source of error in the alignment and test results. The protocols at the labs should be addressed to either eliminate the possibility of load crash through the use of a combined load and displacement control (as in labs 1 and 4) or through the modification of the test rigs to minimize the effect of the impact of the load through an attenuating device.

Both labs 1 and 4 reported difficulties in attaining proper alignment after the clamping load had been applied for some specimens. For various reasons, the outer plates would sit flat on the lower plate under no load, but application of the 49-kip clamping load would cause one of them to uplift at times. It was not a general alignment issue with the load frame as it only happened for a small number of specimens. Lab 1 investigated the problem further and found in many cases the outer plates are never fully in contact with the lower platen. After clamping, a piece of paper could be placed under the plates, sometimes unperceivable to the eye. However, under vertical load, the paper could no longer freely move under the specimen, though it was observed that the piece of paper could freely be placed between the horizontal jack and the saddle. It was determined that if one outer plate had more gap under it than the other, the vertical load would force both into contact with the platen, though that would also rotate the horizontal jack off its support. When this happened, there was a characteristic load versus slip curve that had two plateaus. Examples of this can be found in the appendix A, specifically with the following:

• Lab 1 specimens D1-3 and D2-2.
• Lab 2 specimens A1-3, B2-5, and C1-1.
• Lab 4 specimens C1-2 and C2-1.

The double plateau load/slip curves have fictitious slip displacement because the reported displacement is actually shakedown of the bearing surfaces into the loading platens. This led to a soft response curve where failure would likely be controlled by the 0.02-inch slip criterion.