FlightGlobal.com
Home
Premium
Archive
Video
Images
Forum
Atlas
Blogs
Jobs
Shop
RSS
Email Newsletters
You are in:
Home
Aviation History
1955
1955 - 0348.PDF
348 FLIGHT AIRFRAME FATIGUE . at low altitude than at high altitude and so such differencesrequired careful exploration. Finally, the average time spent at low altitudes as a proportion of the total flight time must beassessed extremely accurately when establishing safe lives. The second main ingredient of the life estimate was the fatiguelife of the component as established by laboratory tests. Generally it was not practicable to make tests under the realistic spectrum ofloads experienced in operation, and one was usually limited to testing say half-a-dozen nominally identical components undera combination of a steady load (corresponding to level flight con- ditions) and a fluctuating load (corresponding to regular up anddown gusts of about 10 ft/sec magnitude). These tests gave an average number of cycles to failure and it was then necessary toconvert the results to an equivalent number of flying hours. This conversion entailed the assumptions that the cumulative damagerule held good and that the S-n curve followed a certain standard shape (unless sufficient tests had been made to provide data atvarious stress levels). In the diagram (Fig. 2) a typical load spectrum correspondingto 10,000 hours' flying was shown together with a typical S-n curve for a spar component. The loads were divided into "packets"—for example, 10,000 applications of load occurred in the bracket from 10 to 14 per cent of ultimate stress. At a stress level of 12per cent the number of cycles to failure was 1.1 X 105. According to the cumulative damage rule, if stresses s,, s2, s3,etc., were applied n,, n 2, n3 times respectively, and if a stress suchas s ( if applied alone would cause failure at N, cycles, then failureunder the mixture of loads occurred when 1 Therefore, the percentage contribution to the damage in terms of the stresses in the 10 to 14 per cent of ultimate bracket was 104 • i ' 10° X i-i x 105 = 9Percent The right-hand curve in Fig. 2 showed the percentage damagedone at various stress levels and it was seen that, if the cumulative damage rule were true, most damage was done over a relativelynarrow band of stresses. This band lay between 4 and 14 per cent of ultimate, the most damaging level occurring at about 8 percent and, to produce this loading, a 10 ft/sec gust was typically the most damaging. Grave doubts had been expressed concerning the use of thecumulative damage rule, and when loads were applied in certain sequences it had been shown to be incorrect. However, the gustload spectrum, although random, formed a reasonably repetitive pattern if taken over long periods.Concerning the second assumption, the variation of shape over the important part of the S-n curve was often fairly small so that,having determined the stress/cycle relationship at a particular stress level, it was reasonable to draw a standard shape of curvethrough this point. It was possible, however, for errors to arise in this.The final step in determining a safe life was to take account of tile variation of life of nominally identical structures. Testson joints indicated that the life of the weakest component of a large batch might be as little as one-third of the average life, andthe one with the longest life might be three times the average. The figures varied from case to case but a ratio of 9 to 1, or10 to 1, could be considered typical. At first sight this was a surprisingly large scatter. However, if scatter was consideredas the variability of the stress to give a particular life, rather than o 20- is- le- s' 20 TYPICAL I LOAD SPECTRUM1 FOR 10.000 HOURS FLYING TYPICAL ,-SPAR JOINT S-n CURVE IOJ I04 CYCLES 10' lO7 0 5 10 PERCENTAGE DAMAGE Fig. 2. An example of the use of the cumulative damage rule for estimating safe life. The right-hand curve shows the percentage damage done at various stress levels. the more usual concept of variability of life at a particular stress,the ratios quoted did not seem quite so surprising. The consequences of scatter were serious. Usual present-daypractice was to test about six components, from these results to obtain a logarithmic mean, and then to divide by a factor of aboutthree to obtain a safe life (exclusive of any factor to allow for ignorance of the external loading). The safe life thus obtainedwas then presumed to be consistent with odds of about 1,000 to 1 against failure occurring if an aircraft flew up to its safe life. Underlying this calculation were two main assumptions: (1) thatthe mean life obtained from about six results was a good estimate and (2) that the scatter pattern was the same for all components.Neither of these assumptions stood up well to examination. Concerning the first assumption, it was clear that, with thesmall numbers of test pieces (drawn from a population with a wide scatter), the chances that the experimental mean wouldclosely approximate to the true mean were not high. As regards the second, the prospect of obtaining an accurate idea of thescatter from only a few results was even more unlikely. It was for this latter reason that it was preferred at present to use a standardscatter, rather than to use a particular value based on an inadequate sample. Statistical methods had been used by Kennedy to determinethe additional allowance which should be made when informa- tion was limited to a few specimens. His calculations had beenused to obtain the example illustrated in the diagram (Fig. 3). Assuming the results of n tests, these indicated an average liferepresented by 1.0, and a scatter such that a life of 0.32 of average occurred once in 1,000 occasions. It was possible that the truemean could be lower or higher, and the scatter greater or less than the experimental values, and so there was no certainty thatthe safe life would be 0.32. Using statistical methods we could say with 90 per cent confidence (i.e., with 9 chances out of tenof being right) the safe life indicated by the experimental results would be no lower than the curve shown in Fig. 3. Thus, ifthere were but six results from which a safe life of 0.32 were deduced, there was in fact a 1 in 10 chance that the life wouldbe only 0.14. This very discouraging conclusion implied among other thingsthat one could feel more confidence than was justified when a, standardized degree of scatter was assumed, even when the experi-mental results supported the use of a standard value. It also meant that if the component indicated low scatter it would beunwise to accept this indication unless many results were available. Turning to the subject of wing tests, Mr. Tye pointed out thatalthough the test of a complete wing was of value, its limitations must be appreciated. Such tests, he said, had shown failures inspar and skin at unexpectedly low lives and in positions not suspected of being critical. Another benefit of such tests wasthat they indicated the type of failure, e.g., whether cracks were of a safe "seeable-before-catastrophe" kind er not.The limitation of wing testing was bound up with the problem of scatter as it was normally practicable to test one wing only,and one could not assume that the test wing was an average one. If the usual order of scatter were assumed then the life of thetest wing could be taken as not more than twice the average with 97^ per cent confidence. Hence to establish a safe life from asingle test result meant dividing by a factor of six—excluding any factors on the gust spectrum. In practice it would probably take a year or 18 months to carryout a test to establish a safe life of 30,000 hours. As well as additional tests on local areas where minor failures were encoun-tered, the main test would need to proceed to a number of cycles greatly in access of the desired operational life. Another difficultywas that minor cracks, if sufficiently numerous, might make the test unrepresentative and, in the lecturer's opinion, it was unlikelythat a wing test could be relied on to provide undisputed evidence that every weak point had been found. While such a test was auseful adjunct to the search for such weak points, it should be considered whether the same effort employed in strain-gaugeexploration and a greater volume of detailed testing would pay better dividends than a test on a complete wing. Concerning the testing of pressure cabins one of the principleloading actions, namely the internal pressure, was known pre- cisely, but the local external air pressures (which might be appre-ciable in high speed aircraft) somewhat complicated the matter. Difficulties of detailed testing had lead to tests on partial cabinstructures and more recently to tests on complete cabins with wings attached. In both cases the cabin was tested under waterloading; partly for safety's sake and partly to limit the destruction so that the origin of failures could be found and the cabin repairedfor further testing. These advantages of under-water testing were to be weighedagainst the draw-backs. Firstly, it was important to know whether cracks were of the safe or "explosive" variety, but there was nocertainty at the present that a safe crack in an under-water test might not be of the catastrophic variety if it occurred in air.Secondly, general testing evidence of aluminium alloys in water and in air suggested that, even in relatively pure water, corrosion
Sign up to
Flight Digital Magazine
Flight Print Magazine
Airline Business Magazine
E-newsletters
RSS
Events
