Effect of toe treatments on the fatigue resistance of structural steel welds Clegg, R.E., McLeod,A.J. and Ruddell,W.

ABSTRACT Toe grinding is effective in being able to improve the fatigue resistance of structural steel welds, but it can also lead to eye injuries if not done correctly and presents a health and safety hazard. In this project, the effects of seven toe treatments on the fatigue resistance structural steel welded plates were investigated to determine whether alternative toe treatments were as effective in improving fatigue resistance as toe grinding. S-N curves were developed for each of the toe treatments by testing samples in bending at three different load levels using a replication of 70% and a load ratio R = 0.1. It was found that the best performing treatments were toe grinding, hammer peening and ultrasonic impact treatments (UIT) and there was no statistical difference between these three treatments. Thermal stress relieving did not improve the fatigue resistance of the samples significantly and the stress levels for failure were approximately 60% of the toe ground samples. Some improvement in fatigue resistance was found in samples where the weld toes were TIG dressed and where the toes were polished with a sanding disc but the results were inferior to those found for the toe ground samples and the peened samples. KEYWORDS Fatigue, toe treatment, welding, post weld heat treatment AUTHOR DETAILS Dr. Richard Clegg is Principal Consultant with Bureau Veritas Asset Management and Reliability Services in Brisbane. At the time the work was carried out for this paper he was Professor of Industrial Materials Science and Director of the Process Engineering and Light Metals Centre at Central Queensland University in Gladstone. He is also Editor-in-Chief of Engineering Failure Analysis and Case Studies in Engineering Failure Analysis. Dr.Alan McLeod is Senior Research Fellow in Engineering Failure Analysis at Central Queensland University in Gladstone. Bill Ruddell is welding supervisor for BMA Coal in Mackay.

1 INTRODUCTION It is well known that welded components have a reduced fatigue life when compared with hot rolled or forged plate. There are several reasons for this. First, the weld geometry can lead to stress concentrations at the toe of the weld. Second, the microstructure of the weld at the toe may be more susceptible to fatigue than the parent plate and often, small slag intrusions can be found at the toe of welds, even in well made welds. Third, the residual stresses in the weld may adversely affect the fatigue resistance of the weld. As a result, welds in fatigue prone situations are often treated to reduce the susceptibility of the welds to cracking. These treatments can include toe grinding, thermal stress relieving and peening. In the coal mining industry, one of the methods currently recommended for reducing fatigue susceptibility of welds is toe grinding. However, there have been some questions regarding the safety of this operation as under certain circumstances toe grinding may lead to eye injuries if the toe grinding is not carried out correctly. As a result, new and safer methods of treating the toe of the welds to improve fatigue resistance are needed. In this project, a number of methods of improving the fatigue resistance of weld toes in structural steel plates are investigated with the aim of being able to compare the effectiveness of the treatments in resisting fatigue. Kirkhope et al. [1] have surveyed a number of weld detail fatigue life improvement techniques and classified the toe treatment methods as shown in Figure 1.

Burr grinding Grinding Methods Disc grinding Waterjet eroding Weld Geometry Improvement Methods Special Welding Techniques Remelting methods Plasma Dressing Weld Pro ile !ontrol Special "lectrodes Shot peening #ammer peening Peening Methods $eedle peening Mechanical Methods %ltrasonic peening Initial overloading &verloading methods Residual Stress Method 'ocal compression Thermal Stress Relie
(i)ratory Stress Relie

TIG dressing

Stress relie methods

Spot heating Gunnert*s Method "+plosive Treatment

Figure 1 Classification of some weld improvement methods. (after [1]). The toe treatments tested in this paper are highlighted in red.

Toe grinding with a burr was found to improve fatigue strength at 2 106 cycles by between 50 and 200% [1]. The effect of toe grinding is to alter the geometry at the toe of the weld in order to reduce the stress concentrating effect [2] and also to remove any small, sharp slag intrusions that

often occur at the weld toe. Zhang and Maddox [3] showed that the effect of toe grinding could be well predicted using a fracture mechanics approach and that for fillet welds, the fatigue endurance is improved by a factor of at least 4.6. Martinez and Blom [4] showed that improvements of between 60 and 110 % of fatigue life could be achieved by TIG dressing of the weld toe. Peening has also been demonstrated to improve fatigue resistance, with significant improvements reported [1]. However, some authors have indicated that the effect of peening may relax as fatigue loading is applied due to plasticity effects which lead to changes in residual stress profiles [5, 6]. Thermal stress relieving is also often used to reduce susceptibility to fatigue. Although thermal stress relieving is known to reduce residual stresses in welds, there is still a lack of specific data to enable the development of design rules. Whilst considerable evidence is present in the literature to suggest that toe treatments can improve fatigue resistance, there are few studies that directly compare the toe treatments studied here. To this end, a series of fatigue samples were prepared to study the phenomenon and S-N curves for each of the cases were developed. Seven treatments were investigated and these are outlined below. Test specimens were designed specifically to ensure that the as-welded residual stresses in the samples were retained during testing and not compromised by subsequent cutting of the samples. There was some concern that conventional cruciform specimens machined from welded sections may not retain the residual stress profiles of the original welds and so circular double plate samples were manufactured. These were designed so that the formation of fatigue cracks in the welds would not be affected by end effects in the samples resulting from specimen preparation and would not be affected by welding stop-starts if these are placed outside of the region of highest fatigue loading.
16 100 dia x 16 450 x 150 x

Strain gauge typ

86 typ Figure 2 Test piece design (schematic).

2 2.1

PROCEDURE Design of test specimens

The test specimens were designed to be tested in four point bending and the design is as shown in Figure 2 and a schematic of the loading configuration is shown in Figure 3. The rig was designed so that the distance between rollers was 80 mm. For a nominal load of 20 kN this would induce a bending moment of 800N-m in the central section of the plate. For a plate of 16 mm thick and 150 mm wide, the extreme fibre stress in the plate would be 125 MPa over the central span of the four point bend rig. Although this is only a surface stress, it is considered to be the stress than will control the fatigue resistance of the plate in the region of the toe of the welds. This is termed the Nominal Plate Stress in this report. The four point bend rig was designed with central roller supports in order to permit some self-alignment of the rig. The stresses in the plate were checked with a sample fitted with strain gauges.

500 x 150 x 50 200 80

50 dia x 150 typ 300 x 150 x 50

Figure 3 Loading configuration for the specimens.

2.2

Finite Element Modelling and Prediction of Deflection

In order to better predict the behaviour of the specimen, a model of the specimen and loading train was modelled in Comsol Multiphysics 4.3. Tetrahedral solid elements were used and the model contained 47126 elements. The model incorporated the loading fixture and the model was loaded to an equivalent total load of 20kN in a square patch on the top arm of the loading fixture. The lower arm of the four point bend rig was fixed and the rig and specimen was allowed to deform in the z direction in the model. The model of an uncracked specimen was found to deflect approximately 0.58 mm in the z direction as a result of a load of 20kN. By examining the load-deflection-frequency curves of the Instron 1273, it was estimated that the maximum frequency for testing would be limited to 5 hz and hence this was set as the target frequency for all testing. Subsequent testing of the specimens confirmed the deflection predictions. The FEA modelling also predicted that the bending stresses in the plate were approximately 125 MPa and that failure of the welded specimen would be in the area of the toe of the weld in the plate on the side of the circular doubler plate. The FEA model showed that there was a significant stress concentration in the weld toe as expected, although accurate modelling of the weld toe was not carried out. The plate stresses were confirmed using a Figure 4 Quarter model of the test rig and specimen, showing von Mises stresses. Total strain gauged specimen.
nominal load was 20kN.

Figure 5 Images of the weld toes tested.

2.3

Fabrication of test pieces

The specimen base plate was 16 x 150 x 450 mm AS 3678-350 plate. A 16 mm thick circular doubler plate 100 mm dia was fixed centrally to the base plate with a 16 mm lap weld so that no weld touched the edge of the base plate and stress was low where the weld was near the edge of the base plate. Seventy test pieces were fabricated by the same welding shop. The test pieces underwent seven different treatments. Images of the weld toes are shown in Figure 5. The treatments were: 1. As-welded. 2. As-welded and thermally stress relieved, ramp up and down at 50 C / hour, dwell at 590 C for 1 hour. 3. Hammer peen the base plate toe only. The steel hammer bit shall have a hemispherical tip with a diameter between 6mm and 12mm. The indentation shall be centred on the weld toe so that metal on each side (both weld metal and base metal) is deformed, resulting in a smooth surface free from obvious individual blows. The hammer should be held at 45 to the plate surface and approximately perpendicular to the direction of travel. The indentation should be approximately 0.25mm. The weld shall be checked visually with MPI prior to peening. A ball diameter of 6 mm was used in this project. 4. Ultrasonic Impact Treatment (UIT) stress relief treatment applied to the base plate toe only. 5. TIG toe dress the base plate toe only. TIG dressing consists of remelting the existing weld metal to a depth of approximately 2mm along the weld toe without the addition of filler metal. The surface to be welded shall be free from rust, slag, and mill scale. The tip of the electrode shall be kept sharp and clean. The tip shall be located horizontally 0.5mm to 1.5mm from the toe. The TIG toe dress shall observe the same requirement for stop-start location. 6. Blend Grind the full weld by sanding with a flapper disk. Blend grind to a smooth surface to comply with AS1554 Part 5. The weld profile before blend grinding should allow for meeting final weld size after blend grinding. 7. Machine a toe groove at base plate toe. Toe grinding shall be done along the centreline of the weld toe. The tool is to be a high-speed rotary burr grinder for use the tip radius shall be at least 5mm. Grinding shall be carried out to a minimum depth of 0.8 mm-1.0 mm below the plate surface. The axis of the burr shall be at approximately 45 to the main plate, the angle of the burr axis shall be a maximum 45 of th e direction of travel to ensure that the grinding marks are nearly perpendicular to the weld toe line. 2.4 Fatigue test procedure

Test pieces were placed in four-point bending in a jig where the loads were applied by hardened bearing rollers 50mm dia x 150mm long. Four rollers bore on the flat surfaces of the test piece but in the apparatus they were cradled in grooves. In addition to the four rollers which applied loads to the test piece, two more rollers at top and bottom centre equalised loads to either side of the jig. Testing was carried out in an Instron 1273 servo-hydraulic machine with Instron 8800 cycle control and monitoring software. All testing was carried out at room temperature. The test plan involved three different stress levels with each stress level tested in triplicate. The cyclic frequency was generally 5Hz, sometimes modified down to allow for equipment limitations as maintenance was required. The load range was applied with R = Pmax/Pmin = 0. The preset force range was maintained throughout the test. As cracks formed and grew, the deflection of the sample increased for the applied load and this was used to detect the cracking of the samples. Once the crosshead deflection increased by more than 0.5 mm the specimen was deemed to have failed. The presence of cracking was confirmed in a number of samples using magnetic particle inspection and one of the cracks is shown in Figure 6. At termination of the test, the number of cycles to failure, Nf, was recorded but the load-deflection data was also recorded at a number of cycle points during the tests.

Cracking was detected by measuring changes in the compliance of the specimens. As the specimens cracked, the stiffness (or compliance of the specimens) would change and so for a given load, the specimen would deflect more. In these tests, a change in the maximum deflection of approximately 0.5 mm was chosen as the point at which the test was terminated. This was chosen simply as it was readily set as a cycle stop signal and indicated the definite presence of a crack. However, some more investigation into the method of termination of the tests was carried out in order to more fully understand the relationship of compliance to crack growth and variation with stress range. Figure 6 Cracking formed at the toe of the weld. The increase in movement for the same force range occurred because the compliance of the specimen increased as a crack grew. Compliance (C) is defined as the ratio of the crosshead movement () to the load (P).

C=

(1)

The compliance of the specimen, and 3.40E-05 hence the progress of a crack, could be monitored historically with good accuracy after testing using stored Specimen data 3.30E-05 load-deflection data. Figure 7 shows a Uncracked typical compliance vs number of cycles curve. As can be seen the compliance of the specimen was stable in the initial 3.20E-05 stages of the tests, but increased towards the end of the test, indicating the presence of cracking. In some 3.10E-05 tests, up to 1/3 of the life of the test was taken up with crack growth. The relationship between compliance 3.00E-05 vs crack length is difficult to determine 0 50000 100000 150000 200000 analytically, as crack shapes are Cycles complex and not known prior to the analysis. However, an estimate of the Figure 7 Typical compliance vs number of cycles curve. crack length vs compliance curves for the specimens can be made using finite element 3.0E-06 analysis. This was done by inserting cracks that were segments of a circle into the previously 2.5E-06 developed FEA model (see 2.2, above). Although this does not accurately represent the 2.0E-06 cracks developed in the samples, the results are able to provide an order of magnitude estimate of 1.5E-06 the effect of crack growth on the compliance of 1.0E-06 the specimen. Modelling was carried out using Comsol 5.0E-07 Multiphysics 4.3 using the previously developed model. A straight crack was inserted at the toe 0.0E+00 of the weld by using a 200 mm radius circular 0 5 10 cylinder placed at a distance of 200 mm from the Crack Depth, a (mm) surface of the specimen. The height of the
Change in compliance (mm/N)

Specimen Compliance (mm/N)

Figure 8 Relationship between change in compliance and crack depth for the FEA model.

cylinder was 0.2 mm and this was considered

thin enough to simulate the thickness of a crack. Crack length was increased by progressively increasing the radius of the cylinder. As the crack depth, a, increased, the surface crack length also increased. This relationship is purely a result of the modelling technique and so is only an approximation to the real crack geometries. In the FEA modelling of the cracked component, the crack was grown from a = 0 to 8 mm at 1 mm increments. The compliance was determined by determining the ratio of the deflection of the top arm at the point of loading to the load. As this was a quarter model and was symmetrical about the centreline of the sample, this compliance would have reflected the growth of two cracks one at each end of the welded patch. However in most cases, only one side of the circular welded doubler plate cracked. As a first approximation, the change in compliance was halved to represent cracking only in one area. The resultant compliance curve is shown in Figure 8. Based on this modelling, an increase in compliance of 5e-7 mm/N represents a crack depth of 3.3 mm and a surface crack length, c = 36 mm or a full crack length of approximately 70 mm. More accurate modelling of the crack surface will provide a more accurate representation of the relationship between crack size and compliance, however for the purposes of this report, the modelling is considered to be accurate enough to indicate that a change in compliance of 5e-7 mm/N would result in a crack of measureable size. In this work, the End of Test criterion was that the specimen deflection increased by 0.5 mm. An alternate method would be to determine the number of cycles until the compliance increased to 5e7. The advantage of the second method would be that in each case, the test would be terminated once the crack had grown to a specified size in this case and equivalent FE model crack of approximately 70 mm on the surface and 3 mm deep and the results should be more consistent than the current approach. In order to compare the two methods, the compliance method was applied to Series 4 and Nf was defined as the number of cycles taken to produce a change in compliance of 5e-7 mm/N. The results of this comparison are shown in Figure 9. As can be seen, the compliance based method of determining the end of life provided slightly shorter lives than for the deflection based method. However, the difference between the two curves was not great. As a result, whilst the compliance method may be a more complete method of determining the end of life of the sample, it is not considered to provide sufficiently improved data to warrant the reanalysis of the data. Therefore, the displacement method of determining end of life is considered to be adequate for the purposes of this paper.
450 440 Nominal Plate Stress Range (MPa) 430 420 410 400 390 380 370 360 350 10000 100000 Nf (cycles) 1000000 Compliance based Deflection based

Figure 9 Comparison of compliance based deflection based method of assessing end of life (Series 4).

2.5

Statistical analysis

The test results were fitted to a log-log regression line of the type y=mx+c. The independent variable was force range, log ,P, and the dependent variable was cycles to failure, log Nf. This resulted in the power law relationship: Nf=A,Pm, derived from log Nf = log A + m log ,P This regression analysis was performed by Microsoft Excel array function LINEST which also gave standard deviation of log Nf. Values of standard deviation were calculated for some test series from the formula given by Lee and Taylor [7], and the same value was obtained. 450

Nominal Plate Stress Range (MPa)

400

350 Series 1 Series 2 Series 3 250 Series 4 Series 5 200 Series 6 Series 7 150 10000 100000 Nf (cycles) 1000000

300

Figure 10 Summary of experimental data - mean curves.

3 3.1

RESULTS Fatigue test results

The fatigue test force range ,P, Nominal Plate Stress Range, and cycles to failure Nf are given numerically in Table 1 which also shows the standard deviation. The results are summarised in Figure 10 and Figure 11. The results fell into three groups. The poorest performers were Series 1 and Series 2, the as welded (1) and the thermally stress relieved (2) test pieces, which overlapped each other within statistical significance. The next group of two conditions with similar, improved performance were Series 5 and Series 6; TIG toe dressing (5) and polishing with a flapper sanding disc (6). However these treatments had wide distributions and the upper halves of the distributions overlapped the top series.

Table 1 Test results for all test pieces with statistical analysis
Nomin al Plate Stress Range (MPa)

Number of cycles to failure, Nf Series 3 9 tests Series 4 9 tests Series 6 9 tests

Force range P, kN

Series 1 9 tests

Series 2 8 tests (X)

Series 5 6 tests (X)

Series 7 8 tests (X)

30 36 36 36 40 40 40 46 46 46 52 52 52 58 58 58 64 64 64 70 70 70 Log(A) A m sd(1) sd(2) sd(3) s(200k)

187.5 225 225 225 250 250 250 287.5 287.5 287.5

2045285 515532 695213 494356 194178 206290 274863 121290 218721 153263 252749 238217 156089 138692 105209 2000000 (*)
1021165

325 325 325 362.5 362.5 362.5 400 400 400

57120 64751 86368

575528 337203 235613 212710 175576 502704 218640 132122 223456 328314 136200 328314 (X) 273101 143719 128353 116045 361225 97786 132034 58206 505360 1000000 278605 394715 227432 132681 334161 143771

947459 873886

169637 217967 166915

166659 182843 207674 53006 71403 75515 18.45

2.8110
18

437.5 437.5 437.5 13.94

8.6910
13

117465 84372 104040 12.14

1.3910
12

18.99
9.8410
18

11.56
3.6510
11

14.81
6.4410
14

16.75
5.641016

-5.2604 0.1758 -0.0334 0.0741 43.9

-4.2430 0.1025 -0.0242 0.0541 41.0

-7.5895 0.0579 -0.0076 0.0174 63.7

-7.3569 0.1637 -0.0223 0.0500 61.2

-3.5647 0.1069 -0.0300 0.0667 57.1

-5.4436 0.2393 -0.0440 0.0963 55.8

-6.3720 0.1749 -0.0275 0.0613 62.7

sd(4) 3.25 2.22 1.11 3.06 3.81 5.37 3.84 Notes (*) Unbroken not used in statistical analysis (X) Some tests were ruled out because of equipment malfuntion or test piece peculiaritiies m A Coefficient in the finite fatigue life equation N=As m m Exponent in the finite fatigue life equation N=As sd(1) Standard deviation on N as specified in Lee and Taylor for log-log regression lines sd(2) Standard deviation on S (log-log) sd(3) Standard deviation as a decimal fraction of mean stress range. s(200k) Mean stress range to cause failure at 200,000 cycles, kN sd(4) Absolute value of standard deviation, kN

10

Comparison of treatments at a life of 200,000 cycles

0.400 1 As welded 2 PWHT 3 Hammer peen 4 UIT Peen 5 TIG toe dress 6 Polish 7 Underflush with burr

0.350

0.300

0.250 Probablility

0.200

0.150

0.100

0.050

0.000 150 200 250 300 350 400 Nominal plate stress range, MPa 450 500

Figure 11 Comparison of all series at 200,000 cycles, showing statistical spread.

The best performance was achieved by undercutting with a tungsten carbide burr and peening by two different techniques (Series 3, 4 and 7). One treatment used a conventional air tool with a 6 mm dia ball (3) and the other employed Ultrasonic Impact Treatment (UIT)(4). There was no statistically significant difference between Series 3, 4 and 7. The difference in the seven test series can be seen clearly in Figure 10, which shows the mean life values for each series as calculated from the test results. The statistical scatter found in the test specimens is shown in the probability distribution curve shown in Figure 11. 4 DISCUSSION The stresses predicted using calculations and FE analysis predicted well the stress values measured using strain gauges, validating the experimental approach taken. The level of replication in the test program allowed a reasonable statistical analysis of the data. The results demonstrated that toe grinding provided significant benefit in improving fatigue resistance when compared with the as-received welds and that the two peening methods provided a similar level of improvement. This was interesting in that toe grinding and peening target two different issues in the weld. A major effect of toe grinding is to remove the toe intrusions and improve the stress concentration effect at the toe, whereas peening is predominantly aimed at changing the residual stress profile at the toe of the weld, putting the area into compression at the surface. However, the effects on fatigue resistance are similar. Also, an unexpected result was that post weld heat treatment did not improve the fatigue life over that of the as welded plates. The PWHT performance was similar to that of the as-welded plates, but the mean is lower and the distribution is tighter. It may be that the fatigue resistance of the welds was controlled by toe intrusions and whilst the peening would have put the toe intrusions in compression, the PWHT would not have affected the toe intrusions. This has to be confirmed by sectioning of the welds. The TIG toe dressing was comparable with the flapper disk-polished finish, its distribution of results lying entirely within the range of results of the more superficial treatment. These treatments also

11

overlapped upwards into the mechanical treatments but the width of their distributions would imply a poor low-life tail which would not allow the upper end of the range to translate into reliable performance. The mechanical treatments improved the fatigue life further as might be expected: In order of peak distribution, the best was hammer peening followed by UIT and burr cutting underflush. However the distributions indicate that these processes cannot be separated at the 50% level as measured by area under the curves. 5 CONCLUSION

The results fell into three groups. The poorest performers were Series 1 and Series 2, the as welded and the thermally stress relieved test pieces. There was no statistical difference between these two series. The next group of two conditions with similar, improved performance were Series 5 and Series 6; TIG toe dressing and polishing with a flapper sanding disc. However these treatments overlapped the top series too. The best performance was achieved by undercutting with a tungsten carbide burr and peening by two different techniques (Series 3, Series 4 and Series 7). One treatment used a conventional air tool with a 6 mm dia ball and the other employed ultrasonic Impact Treatment (UIT). The top three treatments overlapped and no statistical difference could be seen between these treatments. 6 REFERENCES

ACKNOWLEDGEMENTS:
The authors wish to acknowledge the support of BMA Coal, who sponsored the work and supplied the samples.

REFERENCES:
1 2 Kirkhope K.J., Bell R, Caron L Basu R I and Ma K-T 1999 Weld detail fatigue life improvement techniques. Part 1: review. Marine Structures 12 447-474. Ferreira J A M and Branco C A M 1989 Influence of the radius of curvature at the weld toe in the fatigue strength of fillet welded joints. International Journal of Fatigue 11 (1) 29-36. Zhang Y-H and Maddox S J 2009 Fatigue life prediction for toe ground welded joints. International Journal of Fatigue 31 1124-1136. Martinez L L and Blom A F 1997 Influence of life improvement techniques on different steel grades under fatigue loading. Fatigue Design of Components: European Structural Integrity Society 22, 175-191. Zhuang W Z and Halford G R 2001 Investigation of residual stress relaxation under cyclic load. International Journal of Fatigue 23 S31-S37. James M N, Hughes D J, Chen Z, Lombard H, Hattingh D G, Asquith A, Yates J R and Webster P J 2007 Residual stresses and fatigue performance. Engineering Failure Analysis 14 (2) 384-395. Lee Y-L and Taylor D 2005 Stress-based fatigue analysis and design. In Lee, Pan, Hathaway and Barkey (eds) Fatigue Testing and Analysis, Elsevier, Amsteram, 103111.

3 4

5 6

12