Figure 3A shows five individual CDLP tests with a drum diameter of 50 mm with 20 mm wide specimens of CF-epoxy partly cured. All the curves show first an increase which seems to reach asymptotically a level of approx. 0.55 N/mm. Beyond a delamination length of 60 mm the individual curves show an increasing scatter and only one curve stays at this level, while all other show a steep increase levelling off at values between 0.5 and 1 N/mm. In Figure 3B six individual CDLP tests with a drum diameter of 50 mm and 10 mm wide specimens are shown: In the first part until a delamination length of 60 mm, these curves show a similar behavior as the tests with the wider specimens, but in the second part of the tests, the toughness decreases steeply. Figure 3C shows six individual CDLP tests with the standard drum (diameter = 100 mm) and specimens width of 75 mm. In the very first part of the tests, we can recognize a sharp peak in the toughness that is likely due to an initiation effect caused by the insert film. In the determination of the propagation value however the first 5% of the data points are excluded; so these spikes are not influencing the results. In the second phase of the tests between a delamination length of 10 mm–30 mm a roughly constant value of G_c of about 0.5 N/mm is observed, and finally a steady increase to about the double of the constant propagation value, i.e., G_c = 1 N/mm follows. Figure 3D shows the Mode I DCB tests for this material using specimens with a width of 20 mm and a thickness of 3.2 mm. The curves for the Mode I DCB tests fall into two groups, two tests show a rapid increase in delamination propagation resistance leveling off in a propagation plateau of G_c around 0.6 N/mm, and three specimens with a slow increase to about the same level but not earlier than close to the end of the test. The delamination propagation values for all Mode I DCB tests are between 0.2 and 0.3 N/mm.

Figure 4A shows the CDLP test data for 50 mm diameter drum and 10 mm wide specimens. The resistance curve first drops slightly until the delamination length reaches about 20 mm. Beyond this point, the resistance remains roughly constant throughout the rest of the peel test, with a moderate scatter within each individual test and among them. The CDLP test with a drum with a diameter of 50 mm and specimen with a width of 20 mm (Figure 4B) yields a very pronounced initial G_c peak. The resistance drops dramatically within the first 5 mm of delamination propagation and then remains fairly constant throughout the rest of the test. There is little scatter within each test and among the four tests. The smallest drum (diameter = 20 mm), used with 10 mm wide specimens (Figure 4C) shows also a minor initial G_c peak that drops within 7.5 mm delamination propagation, except for two specimens: No. 4 and No. 5. This peak is followed by a smoothly decreasing resistance up to delamination length between 25 mm and 35 mm. Two tests show irregular behavior: No. 1 and No. 2. While No. 1 shows a fairly constant resistance of around 0.4 N/mm between 7.5 mm and 22.5 mm of delamination, No. 2 shows higher resistance between 10 mm and 25 mm of delamination length. The three other tests show less scatter within each individual test as well as among them. The MLP test with this material is shown in Figure 4D. All these tests show first a decreasing resistance followed by an increasing resistance. The scatter within each test is low, but significant among tests. Similar to the first batch also for the fully cured CF-epoxy laminate standard Mode I DCB tests were performed. These tests were done with 4.8 mm thick specimens (Figure 4E). The Mode I DCB tests were also performed with 3.2 mm thick specimens (graph shown in supplementary data). In both Mode I DCB tests the propagation values are dependent on the delamination length: The longer the delamination the higher the resistance is. This can be attributed to the well-known fiber-bridging effect. The scatter within individual tests as well as among tests is significant.

The third data set (Figure 5) shows data for the unidirectional CF-PAEK laminate. The first observation from the data presented in Figure 5 is that the measured G_c values for the unidirectional carbon fiber thermoplastic laminate are clearly higher than those for the thermoset laminates (shown in Figures 3, 4). The difference roughly amounts to a factor between three and four. The resistance curves from the two peel tests, e.g. with the climbing drum with a diameter of 50 mm (Figure 5A) and with the mandrel with a diameter of 20 mm (Figure 5B) show again an initial G_c peak that drops within the first 5–8 mm delamination length. After that, both tests yield roughly constant values for the next 30 mm of delamination length, and then the values clearly increase. The values for the MLP tests rise more slowly than those for the CDLP tests. The CDLP tests shows at least one instability each, essentially an instantaneous drop in G_c. No clear indication of instability is observed for the specimens tested in the MLP test. The scatter in both peel tests is low within a test and moderate among them, with one exception: No. 4 of the MLP tests seems to have a lower resistance throughout the whole test. An explanation of this outlier could not be found. For comparison also Mode I DCB tests with the same material have been done (see Figure 5C). After a clearly increasing part the resistance curve flatten out and tend to decrease the longer the delamination is. All specimens in the Mode I DCB test yield a delamination propagation plateau between about 1.5 and 1.9 N/mm. The scatter within each test is small and among tests moderate. Possible explanations for the increasing part at the beginning of the tests are the same as those for the epoxy laminates, e.g., development of fiber bridging (even though not observed visually) or migration of the delamination. Further investigations will look into this in more detail.

Figure 6 shows a comparison between the CDLP tests using a drum with a diameter of 50 mm and specimen with a width of 10 mm (Figure 6A) and the MLP tests with a mandrel with a diameter of 20 mm and specimens with a width of 10 mm (Figure 6B) for the CF-PEAK cross-ply laminate. Both data sets show several cases of unstable delamination propagation, essentially in all specimens. This is reflected in the fracture surfaces that, upon visual inspection, show alternating ranges of “dark” and “bright” zones, see Figure 7. This is a clear difference compared with the unidirectional laminates. If initial peaks in G_c are presented, they are of the same order of magnitude as the overall scatter in the curves. The data from the CDLP tests show a slight decreasing trend with increasing delamination length. Within the scatter, the data from the MLP tests look fairly constant over the whole range of delamination length. However, the specimens seem to fall into two groups with higher and lower G_c values, respectively, differing by about 0.5 N/mm.

The fifteen series of data are treated as independent groups in a statistical analysis with the goal to perform a preliminary evaluation of various effects. First, the normality test according to Shapiro and Wilk (Shapiro and Wilk, 1965) is done, which showed that all series are normally distributed, except for series 1 and 7. Second, pairs of series are tested with the Welch’s t-test in order to identify differences between series. The Welch’s t-test (also known as Welch’s unequal variances t-test, (Welch, 1947), is applied, because it assumes only normal distribution of the data but not equal variance. The significance of material differences, differences in the curing process, specimen width effect and drum size effect are statistically evaluated. The sensitivity of the three test methods is also evaluated using Welch’s t-tests. Some limitations had to be accepted in this preliminary study because not all data sets were of equal size and the test matrix was incomplete: The variability of the data as well as repeatability of the three methods (CDLP, MLP, and DCB) cannot be quantified. Nevertheless conclusions concerning the scatter in data gained by the different methods can be evaluated. A critical p-level of 0.05 was chosen for the judgment of significance of an effect. Comparing the results from the fifteen test series in Table 2 can provide preliminary information on the various aspects that are important for assessing the feasibility of a CDLP test as well as for the comparison with the other test methods for the determination of delamination resistance of thin fiber-reinforced laminates. The summary of the results are numerically given in Tables 2–4 graphically shown in Figure 8.

Figure 8 shows a clear difference between the test results gained with laminates with the thermoset epoxy matrix (Figure 8A) and laminates with the thermoplastic PAEK-matrix (Figure 8B). This statistically significant difference has been found with all test methods (see Table 3) and has been expected, of course.

There are small, but significant differences between the results of the tests with specimens produced in the first and the second batch of CF-epoxy. These differences in resistance are attributed to differences in processing: As noted above, the specimens from the first batch turned out to be not fully cured (only partly cross-linked to about 85% as shown by DSC, see supplementary data for details) compared with the fully cured (>95% cross-linking) specimens of the second and third batch. The higher cross-linking state reduces the delamination resistance, since the epoxy matrix becomes more brittle and less compliant compared to the less cross-linked. A SEM spectroscopy study showed also a difference in the cracking mechanism: While in the fully cured material the fracture is within the matrix the fracture is at the interface between the fibers and the matrix in the partly cured material, see details in supplementary data This demonstrates the sensitivity of both, the CDLP and the MLP test method to discriminate slight processing differences in nominally identical laminates (p = 5.15E-03 and 6.37E-04, see Table 3). Interestingly, this difference could not be found by the Mode I DCB test: The statistical test did not show a significant difference in the mean propagation value (p = 3.27E-01, see Table 3). The quality of the determination of the fracture toughness was however not very high for the partly cured specimens. As can be seen in Figure 3 the delamination resistance does not show a clear plateau but can increase (Figures 3A, C, D) or decrease (Figure 3B). While an increasing behavior in the Mode I DCB test (Figure 3D) might be due to fiber bridging, for CDLP tests a fiber bridging effect was not observed. It might be that an inhomogeneous distribution of material properties has affected the resistance measurement. In series 6 the softer wires might have influenced the resistance measurement negatively, although any compliance effect of the climbing drum device should not significantly affect the resistance measurement: The peel force is determined as the difference between the peel experiment and a base-line measurement done for each specimen.

Does the width of the specimen influence the test result in the CDLP test? The Welch’s t-test with series 1 against series 3 does yield a clear answer: The width of the specimen does not influence the results from the CDLP test (p = 5.34E-01, see Table 3). However, because the scatter in tests with the most slender specimens (10 mm width) is doubled compared to test with 20 mm wide specimens (see Table 2), 20 mm wide specimens are preferred.

It was important to compare the CDLP test with the MLP test using the same diameter, giving the same peel radius assuming full conformity. However it was difficult to measure a stable value with the drum of this diameter (series 2). The reason was the following: The drum diameter is limited by a requirement of a minimum delamination length necessary to measure a stable value in the test: Geometrical restrictions limit the maximum delamination length to about 2/3 of the circumference of the drum: After 2/3 of the first round of the drum a collision of the wound layer of the specimen with the unpeeled part of the specimen is inevitable. Therefore the maximum possible peeling length is roughly 2 times the diameter (circumference = Diameter۰π), which gives 40 mm for the drum of 20 mm in diameter. In addition most of our CDLP tests show a peak in the data for initial delamination lengths between about 5 and 10 mm that is attributed to an effect related to the insert film. Therefore, the available delamination length for the measure of the propagation resistance was only 35–40 mm. Within this range a stable value of G_c was not reached. In order to measure a stable G_c, a minimum delamination length of about 80 mm is provisionally recommended. Therefore, a minimum drum diameter of 40 mm should be used. Hence, the evaluation of the CDLP test method was continued with the drum with a diameter of 50 mm. There is also likely a lower limit set by the conformance of the specimens around the climbing drum or the mandrel. Specimen conformance is depending on peel arm flexibility and drum diameter. The diameter limit may, to some extent, depend also on the material to be tested. For the epoxy and the PAEK-material investigated here, it is concluded, that a diameter of 50 mm is sufficient. The Welch’s t-test gave no indication of a drum size effect (p = 4.53E-01, see Table 3). The Welch’s t-test for results gained with a drum of diameter 50 mm versus 100 mm was not clear. The p-value was only 7.07E-02, see Table 3. A judgment of the existence of a difference between the two series based on a confidence interval of 95% would results in a negative answer. A difference is statistically found to be not significant. Therefore an effect of the drum diameter cannot be excluded if the diameter is increased beyond 50 mm. With respect to the best choice for the drum size, a diameter of 50 mm is a good compromise between the requirement of minimum delamination length, conformity and amount of material needed to perform the tests.

The difference in delamination propagation resistance between the two types of fiber-reinforced laminates (epoxy versus PAEK matrix) was clearly found (all p < 1E-07 for series 7 vs. 1, 10 vs. 9 and 15 vs.12, see Table 3). The comparison between the two laminate lay-ups, e.g., UD vs. CP of the PAEK matrix laminates give indications of a difference in both peel tests (p < 1E-03 for series 8 vs. 7 and 11 vs. 10, see Table 3). Due to limited sample size of series 7 (only 2) and 11 (only 3) such a difference cannot be claimed to be clearly significant for both peel tests. Such an effect could be found in Mode I DCB tests of another thermoset carbon fiber epoxy laminates with unidirectional and cross-ply lay-ups reported and discussed in (Brunner and Blackman, 2003; Brunner et al., 2003). Clear differences in delamination propagation resistance, i.e., the plateau values between the two lay-ups were shown. Therefore, a lay-up effect can be expected in both materials and all three test methods. More statistically better established testing is necessary to obtain proof of this effect in both materials. A very interesting case is the comparison of the partly cured first batch of the epoxy laminate (series 4, 5, and 14) with the fully cured second batch (series 1, 3, and 13): This case shows that the CDLP test is sensitive enough to detect such small process-induced variations in material morphology in contrast to the Mode I DCB tests which cannot detect this difference! This has shown the Welch’s t-test of series 4 vs. 3 giving a p-value of 6.37E-04 for specimen with b = 20 mm and the test with specimens with b = 10 mm, series 5 vs. 1, giving a p-value of 5.15E-03! The corresponding t-test for the Mode I DCB test gave a p-value of only 1.26E-01. This is an important finding for process quality checking were the identification of process variations or irregularities influencing mechanical properties of the material or parts fabricated have to be detected with sufficiently high accuracy. For future studies it is important to base them on balanced samples such that the statistical evaluation will give clear evidence of such effects.

Are the G_c propagation values gained with the CDLP test method comparable to the values gained with the MLP or the standard Mode I DCB tests? A graphical overview is given in Figure 8. This question could not be answered unequivocally; it is necessary to differentiate between the different materials, processing conditions and test methods: A clear difference was statistically found between G_c determined with the CDLP test and the MLP test for two materials: Epoxy-fully cured and PAEK-UD or CP layup (Epoxy partly cured was not tested with the MLP test). This could be shown by comparing series 2 vs. 9, 7 vs. 10 and 8 vs. 11: All p-values are less than 5E-03. No such difference could be found between the CDLP test and the Mode I DCB test for the epoxy-material in both conditions (series 12 and13 vs. 3, 14 vs. 4): The p-values are 4.42E-01 and 3.86E-01 respectively. For the PAEK-UD-material such a difference however does exist: The Welch’s t-test of series 15 versus 7 gave a p-value of 2.23E-03 (PAEK-CP was not tested in a Mode I DCB test). One reason why a difference with the Mode I DCB test could not be found is the larger scatter in this test. Because the sample size of series 2, 7, and 11 affect these test, all these conclusions are preliminary and should be confirmed with complete analysis based on a better data set. The CDLP test might also be preferable because the force measurement is easier to perform compared to the MLP test: The forces are in the order of 20–25 N compared to 4–7 N with the MLP test, which is five times larger. Nevertheless, from these first findings we can conclude that the three methods presented here should not be treated as fully interchangeable.

The last aspect which can be addressed is the Mode-mixity. How much the delamination is loaded in Mode I or Mode II in a peel test (CDLP or MLP) is still not answered. The mode mixity can be estimated to be close to 100% Mode I by comparing G_c for pure Mode I of 0.399 N/mm (average of series 12 and 13 with the average of the CDLP and MLP test results (series 1 to 3 and series 9) of 0.386 N/mm. This is an indication that laminate peel tests give a Mode I dominated resistance value.

A first step was made with this study to explore the applicability of the CDLP test. The results look promising: The two types of laminates, one with a brittle thermoset and one with a tougher matrix, were successfully tested. The latter in two different lay-ups (UD and CP). Beside the need for a better data base to confirm the preliminary results, there are several possible directions for continuing the exploration of the climbing drum test method: One is testing of glass-fiber reinforced epoxy laminates, another extending the tests to highly toughened carbon fiber laminates. Testing woven fiber reinforcement instead of unidirectional or cross-ply is another direction. With respect to three-dimensional reinforced laminates, e.g., stitched or with z-pins, the applicability of peel tests CDLP or MLP is judged questionable.