PAPER 50 Fatigue of a welded steel attachment under combined direct stress and shear stress R. Archer, BSc, MSc i , SUMMARY A series of constant amplitude fatigue tests has been undertaken on structural steel box beams having small longitudinal attachments welded on two opposing sides. ‘The beams were loaded to give both cyclic direct stress and cyclic shear stress at the attachments. Three types of loading were used - (i) shear stress alone, (ii) direct stress and shear stress at equal frequency (both in phase and out of phase) and (iii) direct stress at twice the frequency of shear stress with various phase relationships. The fatigue results were atisfectorily correlated using both an interaction equation approach and a ‘maximum damage’ computational procedure. INTRODUCTION Many gngineering structures, e.g. railway bogies, experience fluctuating Diaxial stresses in the vicinity of fillet welded attachments. Increasingly sophisticated finite element stress analysis is enabling the designer to recognise the presence of both proportional and nonproportional biaxial stresses and he is in need of fatigue analysis methods to relate these stresses to design life. Furthermore the ability to gather increasingly more detailed experimental stress data from prototype structures must be accompanied by the ability to analyse biaxial stresses so as to predict service life. The bulk of experimental data on the fatigue life of welded joints is for Uniaxial stress. There is a limited amount of data for the biaxial stress case of bending stress and shear stress acting in phase, obtained from stiffeners welded to the webs of I-beams, e.g. (1) which includes a summary of Many other papers, and (2). The data are usually correlated using maximum Principal stress and as such are incorporated into design codes such as BS5400 Part 10 (3). ‘There is no work known to the author dealing with out of phase direct stress and shear stress (ie. 90 deg phase shift) nor with any more complex loading. he experiments reported in this paper form an initial attempt to fill this gap. Fatigue tests have been undertaken on fillet welded attachments subjected to pure shear stress, shear stress with direct stress in ise, shear stress with direct stress out of phase and shear stress with direct stress at twice the shear stress frequency. EXPERIMENTAL, peTATLS Asst_specinen test specimens were made from structural hollow section to BS4848 part 2, by spr Giuare by 10 mm wall thickness, Fig. 1. The attachments, 10 mm thick mm long were fillet welded to the sides of the box beams where the ckness had been reduced to 6 mm. The reduction in wall thickness allowed fh shear stresses to be produced without risk of bending failure by thir hi,54 Fatigue of Welded Constructions, April 1987 yielding. The box material was BS4360 grade 43C and the attachment was 434. A ‘large’ specimen was chosen so that tensile residual stresses would resul; from welding, although these have not be measured. This weld detail classified as F according to BS5400 (3). Examination of the uniarial test data from various sources suggests that the configuration may be above the mean of class P in the longitudinal directios All the specimens had three-element rosette strain gauges attached to the b beam near to each of the four weld toes lying on the beam neutral aris. Test rig The test rig, Pig 2, was designed to apply cyclic shear stress and cycli direct stress to the sides of the box. Two of the four servohydraul: actuators operated horizontally and in phase to produce four-point bending the beam and a uniform state of direct stress at the attachment. ‘The ot: two were vertical and acted 180 deg out of phase with each other to apply pure shear stress at the attachement. The actuators were all operated in 1 control from separate control units, driven either from a standard two out: phase shifting oscillator or from a purpose built oscillator which gave ti sinusoidal outputs, one output being at twice the frequency of the other at variable phase to it. The outputs from the strain gauges were recorded an ultraviolet galvanometer for the pure shear tests. Yor the tests und combined bending and shear the data were digitised on-line and stored on computer disk. Test procedure All beams were dynamically calibrated before testing for the independ relationships between loads and shear or bending stress. During the fati, test constant values of cyclic load were applied to the beams at a suitabl frequency between 1 and 5 Hz. Stresses were obtained from the strain gauges using a Young's modulus of 207,000 N/mm* and a Poisson's ratio of 0.3. Al beams were checked regularly for cracks using dye penetrant. Stress histories There was a single test under pure bending and there were ten tests under pu shear. Four beams were tested with shear stress and direct stress acting phase. That is, the turning points of shear stress coincided with the turnii points of direct stress. The shear stress ratio was -1 and direct str ratio 0, (Stress ratio = min stress/max stress in cycle). These stret histories are all described as proportional loading (beams 1 to 15 in Table 1) The first set of nonproportional loading was applied to beams 16 to 19 consisted of shear stress and direct stress (as for beams 12 to 15) with a deg phase difference between them so that the turning points of direct stre: coincided with zero shear stress. The second set of nonproportional loading was applied to beams 20 to 27. direct stress frequency was exactly twice that of the shear stress. Foul different relationships between direct stress and shear stress were used shown by the time histories, Fig 3. These stress histories produced a wid range of principal stress variations. Test results All the test results are presented in Table 1. Failure is defined as a cra ca Session ilfa): Archer 20 wu long on the surface growing from the toe of the weld. Most beams failed from only one of the two attachments, only beams 5, 10, 12 and 24 failed from oth attachments. One beam had a failure where the direct stress component was compressive (12a). Four beams did not produce a fatigue crack by 10 million cycles. For the 1:2 frequency tests the life is expressed in cycles of direct stress. DISCUSSION OF RESULTS Lssrrre Because of the small number of results it is not possible to produce a fatigue life curve for each stress combination. However there are enough pure shear results, Fig 4, to enable a regression line to be calculated (ignoring run-otits) which has a slope of -2.92, rounded to -3. All the results (except run-outs) are plotted on a single diagram, Fig 6, as ‘normalised’ stresses. Normalisation For each result the actual shear stress (+) is divided by that shear stress which would be required, acting alone, to give the actual number of cycles of shear stress at failure (ry). Two alternative fatigue curves have been used for normalisation in Fig 6: the ‘best fit' regression line and the BS5400 class F mean line. The direct stress (¢ ) components are treated similarly using the number of eycles of direct stress at failure (on). (This automatically accounts for the 1:2 frequency ratio results). The ‘best fit' normalisation in Fig 6 assumes that the regression line passes through the single pure direct stress result with a slope of -3. A striking feature of Fig 6 is that the results from all the combined stress tests are closely grouped together. Shear stress-direct stress interaction relationships A number of relationships between direct stress and shear stress vere examined and three are plotted in Fig 6. The most obvious one for proportional loading is based on the theory that failure occurs at a particular value of maximum principal stress range. This implies that the fatigue strength under shear stress is the same as under direct stress. However the present data do not confirm this and so a Generalised interaction equation is proposed: fou + (rf )e =1 Q] where ¢ /cy and + / ty are the normalised stresses. ‘This equation Fepresents maximum principal stress only if the same fatigue curve is used to formalise both shear and direct stress, ie. oy = T , and if the stresses ‘rein phase. This curve lies below the data using both normalisations (Fig 6) The second equation is in the form of an ellipse quadrant: (e/ oy)? +(e sry 222 (2] This is a good fit to the date using the bel ‘best fit! Ow the data when using class F. normalisation but liesa7 Session Illa): Archer 66 Fatigue of Welded Constructions, April 1987 is shown in Fig 8. The loading is shown in Fig 3, Type ‘me Pature of beet he Desks of Sirect stress coineiding with ‘the turning e er hear stress and has produced a smooth crack. Beam 22, load type 2, points ord beam 26, load Type 4, Fig 10 had a most complex crack pattern. ES SE sinter load Type 3 had similar complex cracks patterns. the © The third equation is derived by assuming that fatigue damage accum simultaneously and independently from the shear and direct stresses, summation of these damages for a fatigue curve slope of -3 leads to: (oley 34+ (t/ ty )3 21 This is a good fit to the data using class F normalisation but is above rack patterns were fairly consistent for each load type. of the data using ‘best fit’ normalisation. CONCLUSIONS SORCLUSTONS he results from fatigue tests on welded attachments under combined cyclic shear stress and direct stress (voth proportional and nonproportional loading) fave been examined using two procedures. The first relates the normalised shear stress and direct stress by interaction equations and the second ets fatigue life by computing the plane of maximum fatigue damage. The yrimental data obtained support both these procedures for predicting fatigue life under combined shear stress and direct stress. There is clearly scope for investigating other interaction equations normalisations using the data plotted on the interaction diagram. Shear and bending tests results A direct comparison may be made between the in phase and out of phase frequency results. At the higher stress level’ there is no difference be the results from the in phase beams, 12 and 13, and the out of phase beams, and 17. However at the lower stress level the out of phase beams 18 and did not fail at stress levels which caused the in phase beams 14 and 15 fail. This suggests that out of phase loading may be less damaging than phase loading around the fatigue limit. A computational procedure for nonproportional loading The ASME pressure vessel code (4) recommends examining each plane in a sti field and evaluating the stress at the extremes of the load cycle. general situation the procedure involves calculating the resolved value direct stress along every plane for every instant in the time history. each plane the resolved time history is examined and turning points of st extracted. A 'Rainflow’ cycle count and Miner's rule damage summation is to give the calculated fatigue life based on the stress along that parti plane. This gives a whole faimily of life predictions, a different 1 prediction for each plane. The plane which gives the shortest calcul fatigue life is the plane of ‘maximum damage’ and is assumed to determine fatigue life of the weld. This procedure has been applied to the data beams 16 to 27, the nonproportional cases, using a class F mean line fat: curve to give calculated values of fatigue life, Fig 5. Note that for proportional loading cases, beams 1 to 15, this ‘maximum damage’ me’ degenerates to the maximum principal stress range method. There is quite reasonable agreement between actual and predicted lives using this netl although the use of the class F line is probably a little too conservative shown by the pure shear results. he interaction equation procedure would be useful in the design stage when the shear and direct stresses can be calculated independently. The maximum fatigue damage plane computational procedure would be useful in two situations. Firstly when the full complex stress history can be ealculated in design or secondly when stresses are measured from a component in service undergoing complex loading. The small number of tests reported make these conclusions somewhat tentative. It is clear that further work is required. (a) at other ratios of direct to shear stress (b) under variable amplitude loading ACKNOWLEDGEMENT The author wishes to thank the Director, Research, British Railways Board for Permission to publish this paper. REFERENCES i. Fisher J W, Albrecht P A, Yen BT, Klingerman D J and McNamee B M: ‘Fatigue strength of steel beams with welded stiffeners and attachments’. WCHRP Report 147, Transporation Research Board, Washington D.c., 1974. CRACI iC Bi LV: SRACEENG BEHAV TOUR 2. Moyar G J and Garg V K: ‘Biaxial fatigue design procedure applied to welded pate structures’. In proc conf ‘6th International Conference on Under proportional loading the cracks grew beyond the 20 mm failure point Fracture’ (Ic-6).. New Delhi, India. Pergamon, 1984, pp 1969-1877 conventional manner. The cracks were approximately normal to the direction maximum principal stress and propagated smoothly to the full width of beam. Under nonproportional loading there was no constant principal st: direction and the long crack behaviour was complex. 5. 835400: part 10:1980 ‘Steel, Concrete and Composite Bridges, Part 10, Code of Practice for Fatigue’. British Standards Institution, 1980. Figure 7 shows the failure of beam 16, (shear and direct out of phase and ra Sask Boiler and Pressure Vessel Code, Section III, Nuclear Plant equal frequency). The crack front has a very jagged appearance, possi i68 6 Fatigue of Welded Constructions, April 1987 Session Ifa}: Archer Table 1 Fatigue test results 50 re Beam Loading Shear stres: Direct stregs Cycles (x10) range, N/am? _ range, N/am’ to 20 mm crack ew 180 Pure direct stress See 200 1 195 0.453 Pure shear stress 2 203 3.56 1. Test specimen boxbeam, dimensions in mm 3 135 0.766 4 109 2.98 5a 130 0.794 50 1 0.864 6 iil 0.685 q 108 1.08 Vertical Actuators 3 120 10.0 # (Shear Stress) 9 ce 10.0 * 10a 150 0.600 100 161 0,662 i 186 0.347 Shear & direct in phase Simply Supported 12a 1s 122 0.460 12b 124 17 0.693, 15 1oL 102 0.783 14 al Ei 1.247 15 9u 94 0.793 Shear direct out of phase 16 123 126 0.404 at 105 102 0.944 Lateral Actuators 18 6 83 10,0 * Simply (Direct Stress} Over supports 19 91 92 10.0 # Sialerriad 1:2 Sheer & direct Type 1 ** 20 101 102 1.068 aL 138 206 0.704 2. Loading arrangement :2 Shear & direct Type 2 ** 22 108 108 1.24 wo Type 2 23 142 107 1.078 % 1:2 Shear & direct ee : Type 3 ** 34 Vali 24a 138 ul 0.860 ze: Type 1 24 113 100 1.136 i 25 115 109 0.600 1:2 Shear & direct Type 4 * 26 136 105 0.564 at 102 102 0.866 Time Shear Stress * not failed ** count cycles of direct stress “«* for beans 16-19, refers to nunbers-of cycles of shear or direct stress 3. Stress variation with time tor 1:2 frequency ratio (which were the same), not their sum.70 Fatigue of Welded Constructions, April 1987 Session Il(a): n : Archer e F = i gis 8 ™ ma vy 200 hs = $%+ 25 @ = = 2s 2 3 e ao. " s a> CU ss oy 5 — 2¢cze i 3 a4 “ie @ 100 Seite = 3 4 + ‘ a aes *y FS & x Pure shear stress ~ Regression on shear 3 #2 3 3 €28 ae ace Pure direct stress ~ — Sigh 528 2 ole Pils mos. 5 gifs: 1.83 & | ¢ 3 \ : ieee a i xH+<10 tao = | <0 05 1 2 5 10 E Life, cycles «108 4. Shear stress and direct stress results 5 4 on Normalised direct stress Predicted life, cycles « 106 Pure shear stress x © Pure direct stress + Shear and direct in phase 4 ° 6. Shear stress and direct stress interaction diagram Shear and direct out of phase 1-2 Shear and direct N, Tied M2 ssaujs sDays pas! owsoN S 10 psa 05 1 Actual life, cycles x 106 5. Life prediction by maximum damage’ using class F mean72 Fatigue of Welded Constructions, April 1887 9.Beam 22 10.Beam 26 Crack Patterns