Intensive Validation of Computer Prediction of Welding Residual Stresses in a Multi-Pass Butt Weld
Weijing He, Liwu Wei and Simon Smith
TWI Ltd, Cambridge, United Kingdom
Paper presented at 29th International Conference on Offshore Mechanics and Arctic Engineering (OMAE 2010), Shanghai, China, 6-11 June 2010.
Welding and joining technology is fundamental to offshore engineering. The construction of engineering facilities and pipelines requires the extensive use of welding and associated structural integrity assessments of safety critical or heavily loaded sections. Proof of integrity is based upon the externally applied loads and in service stresses as well as the welding residual stresses. The level and distribution of residual stresses arises from the complex thermo-mechanical history of heat flow and thermal expansion at very high temperatures during welding, so it has not been possible to make very accurate assessments of these in the same way that service stresses can be defined. Conservative assumptions are therefore made and this often means that the as-welded stresses are assumed to be of yield magnitude. The peak values of stress may well be very high, but the shrinkage of the latter passes of multi-pass welds may compress earlier passes giving rise to much lower levels of stress. There is considerable engineering interest in the utilisation of lower levels of residual stress where they exist or of the design of welds with lower residual stresses in sensitive areas such as the weld root. Currently there is no single technique that can claim to provide cost effective, accurate distributions of residual stresses in welds.
The current paper provides an important contribution to the understanding of measurement and prediction techniques. It describes an extensive set of measurements taken on a girth butt weld. The weld was made using submerged arc and was made in 18 passes. The pipe was X52 with a 32mm wall thickness and 910mm outside diameter. Temperature, strain and displacement values were measured throughout the production of the weld. The intermediate values between each pass were recorded as well as the time varying history during the production of individual passes. The final through thickness residual stress distribution was measured. Finite Element Analysis (FEA) modelling was undertaken to determine whether modelling could provide a satisfactory prediction of the final residual stresses. Intermediate results were also used to understand the behaviour of the weld and the model more clearly. The modelling used material properties measured on material from a separate specimen. The weld cross section was identified for each pass so that the heat input method could be developed to represent the actual melt pool conditions of the weld. The measured values of hoop residual stress were up to the yield stress magnitude just below the cap, but were 20% of yield in the root of the weld. The axial residual stresses were less than 50% of yield. Linear kinematic hardening provided the most accurate prediction of residual stress. The hoop stresses were predicted to an accuracy of 10% with this material model. Other hardening models were less accurate, but all models were conservative. The results provide a basis for the adoption of more accurate distributions of residual stresses in Engineering Critical Assessments (ECAs) and assessments of weld performance under fatigue and corrosive conditions.
The defect tolerance of welds is critical to the safe operation of offshore pipelines and risers. Welding defects can occur and the service conditions can give rise to crack-like corrosion or fatigue features. Codes of practice are available for the assessment of these defects. The American code API 579-1/ASME FFS-1 2007 Fitness-For-Service provides guidance for conducting Fitness-For-Service (FFS) assessments using methodologies specifically prepared for pressurized equipment. The guidelines of the Norwegian Recommended Practice DNV-RP-F108 were developed to give guidance regarding the testing and analysis for the fracture control of pipeline girth welds subjected to cyclic plastic deformation. The British code BS 7910 is more general and applies to the assessment of acceptability of flaws in metallic structures and describes the process as Engineering Critical Assessment (ECA).
FFS and ECA calculations are based upon the material toughness and stresses. The process is based upon a comparison of the crack tip loading in terms of parameters like the Crack Tip Opening Displacement (CTOD) or J with the material toughness expressed using the same parameter. The applied J or CTOD arises from the summation of the stresses arising from service conditions, such as pressure loading and thermal stresses, with the internally balanced residual stresses arising from welding. Welding residual stresses can vary between yield magnitude in tension and yield magnitude in compression in the as welded conditions. This represents a huge variation and some codes of practice recommend the only reasonable conservative assumption that the as welded residual stresses are tensile and of yield magnitude. Alternative recommendations have been generated, but these are based upon Finite Element Analysis (FEA) prediction of the stresses. FEA prediction of welding residual stresses has not been extensively validated and so there is a requirement to provide robust comparisons of the conditions that occur during welding and the final distributions of residual stresses.
The current paper provides some useful results from measurements made on a weld during its production and of measurements of the welding residual stresses after the completion of the weld.
The test weld was made on a 910mm diameter pipe with a wall thickness of 32mm. The weld was completed with submerged arc welding following two MIG root passes. Temperature, strain and displacements were measured during the production of the weld, and the through thickness distribution of residual stresses was measured after the completion of the weld. An axisymmetric model of the weld was prepared. The model and its results are described together with a comparison of the predictions and the measurements on the weld.
The objective of the work described in the current paper is to provide some extensive test results from a heavily instrumented test weld and to compare the measurement with a FEA of the weld.
Generation of test data
Test pipe details
The test pipe was a seam-welded, low carbon microalloyed steel with a nominal minimum yield strength of 335MPa. The chemical composition is given in Table 1. The nominal pipe diameter was 910mm and the wall thickness was 32mm. The weld was made between two sections of pipe, each 825mm in length.
Table 1 Weight percentage test pipe chemical composition (main constituents)
The welding was made into a single V bevel of 50° with a root gap of 3mm and a root face of 2mm. The welding procedure is detailed in Table 2. Measurements were made between passes so the pipe was at ambient temperature at the start of every pass.
Table 2 Welding procedure used to make the test weld
|19 - 21
|9 - 13
|4mm Oerlikon SD3
|28 - 30
|42 - 45
|4 to 18
|4mm Oerlikon SD3
|31 - 33
|40 - 45
The tensile properties of the pipe and weld metal were measured at room and elevated temperature. The respective stress versus strain curves are plotted in Figures 1 and 2. Figure 3 shows the estimated Youngs modulus from the measured stress versus strain measurements.
Fig.1. Measured stress versus strain curves for the base metal
Fig.2. Measured stress versus strain curves for the weld metal
Fig.3. Measured youngs modulus versus temperature for the base metal
Measurements on test weld
Measurements of the thermal and mechanical behaviour of the pipe during welding were made for subsequent validation of a model of the welding process. The measurements were made at locations identified by lines marked on the pipe as shown in Figure 4. The sets of measurements made are given in Table 3.
Table 3 Locations and measurements made during the production of the test weld
|Weld passes assessed
|Distances from weld centre, mm
|Line on weld (see Figure 4)
|Up to 40mm
|3, 8, 13, 18
|Up to 90mm
|3, 8, 13, 18
|Up to 90mm
|3, 8, 13, 18
|Up to 140mm
|Lines L1, L2 and L3
|Lines L4, L5 and L6
|Lines L1 and L2
|Through thickness distribution
|Through thickness distribution
Fig.4. Markings on the pipe used to identify locatons of measurements
A sister weld was made and sectioned, so that the bead shape of every pass in the weld could be defined.
Each half of the final weldment was heat treated to reduce residual stresses after machining of the weld preparation but before welding. The heat treatment was done to reduce the levels of residual stresses before welding because these would confuse the comparison with the computer model which was stress free before welding. Residual stresses were measured on one of the pipe sections using the centre hole drilling technique. The residual stresses were between -17MPa and 21MPa on the outside at the five locations of measurement, including locations on the seam weld and two locations near to the weld preparation. The residual stresses on both the outer and inner surfaces of the pipe near to the weld preparation on line L5 were all between -8MPa and 2MPa. The final residual stress measurements on the completed weld were therefore taken at this location.
Residual stresses up to 70MPa were measured near the weld preparation at line L6 on the outside and as low as -100MPa on the inside at the same location.
The pipe diameter was measured on one pipe piece before welding. It was found that the diameter varied between 912mm and 920mm.
The modelling was undertaken using 2D axisymmetric simulations. A mesh sensitivity study was carried out to ensure a sufficiently refined mesh was used for an accurate prediction of temperature distribution and welding residual stresses.
The thermal properties for both parent and weld metals used in this work were taken from materials handbook for X60 steel. The stress versus strain curves of the parent and weld metal were measured at room temperature and at elevated temperatures. The results are plotted in Figures 1 and 2. The figures also show the assumed room temperature stress versus strain curves used in the FEA model. The Abaqus linear kinematic hardening model was used. The assumed material yield strength and work hardening were temperature dependent (not shown in Figures 1, 2). The temperature dependent Youngs modulus was estimated from the measured stress versus strain curves. The estimated curve is shown in Figure 3.
The model was fixed in the axial direction at one point on the inner surface of the pipe to avoid rigid body motion.
The initial temperature of the whole model was set to the ambient temperature (ie 20°C). The heat transfer coefficient and emissivity at all the exposed surfaces of the pipe were assumed to be 10W/m2K and 0.4, respectively. [Simonson 1975]
The heat input from a moving heat source was firstly converted to an equivalent value for the 2D axisymmetric model of heat transfer in which the welding process was simulated with a body heat flux applied instantaneously around the full circumference of the pipe. The heating time for each weld pass was estimated from the time for an 8mm length weld pool (about two times the weld bead width) to pass any location on the pipe. A triangular welding heat flux was adopted. The heating body heat flux rose linearly to a peak and then linearly back to zero during heating. The body flux form was chosen so that the total heat input divided by the length of the weld was equal to the welding power in kJ/mm. In addition, the modelling was repeated to ensure that the predicted melt pool shape was equal to the melt pools measured in a sister test specimen that was sectioned.
The temperature output from the thermal analysis was used in the mechanical analysis for prediction of welding residual stresses.
Results from model and test weld
A macrosection of the weld is given in Figure 5. This was taken on a section of weld that had been prepared for residual stress measurement and therefore the cap had been ground flat for strain gauging. A typical plot of the predicted distribution of temperature during the analysis of pass 11 is given in Figure 6. The plot shows the melt zone at its maximum extent (region in grey). The distributions of peak temperature around the weld from some of the passes from the test and from the model are compared in Figure 7. The correspondence is not good. The degree of disagreement is possibly because of movement of the weld bead within the weld preparation at different positions along the joint length. The pipes were mounted on a turntable with a fixed welding head, so it is possible that the weld bead wandered in the weld preparation. The correspondence between the weld bead (Figure 5) and the predicted melt pool (Figure 6) was assessed using one macrosection at one location on the weld and this was not at the location of the thermocouples (line L5, Figure 4).
Fig.5. Macrosection of test weld after grinding the cap for residual stress measurement
Fig.6. Predicted melt zone shape for pass 11 of the test weld
Fig.7. Comparison of the predicted and measured peak temperature distributions
Strains measured on the outside surface are plotted against pass number in Figure 8. The values shown were recorded when the pipe had cooled to ambient between passes so the thermal strains will be zero. The hoop strains near the weld are compressive between passes and there is a variation of axial strains from tensile near the weld to compressive further away. The model results provide the same trends but the strains are lower than the measured values. Strains measured in the pipe inside surface are shown in Figure 9.
Fig.8. Comparison of the predicted (open symbols) and measured (filled symbols) residual strains between passes on the outer surface of the pipe plotted against pass number
Fig.9. Measured residual strains between passes on the inner surface of the pipe plotted against pass number
The strain variations during passes 3, 8, 13 and 18 are given in Figure 10. The plots do not provide absolute strain, only the change in strain during each pass. The overall residual strains after each pass are given in Figures 8 and 9.
Fig.10. Measured residual strains between passes on the inner surface of the pipe plotted against pass number
Measured radial displacements taken on the weld root at line L5 during passes 3, 8, 13 and 18 are shown in Figure 11. There is a general reduction of the pipe diameter during welding. The radial displacements taken between passes at four locations around the pipe are plotted against pass number in Figure 12 and the distribution plotted against distance from the weld centreline after the completion of the whole weld is given in Figure 13. The residual strains measured after the completion the weld are plotted against distance from the weld in Figure 14.
Fig.11. Measured radial displacements (positive inward) on the pipe inside surface at the weld root taken on line L5
Fig.12. Measured radial displacements (positive outward) taken between passes when the temperature of the weld had cooled to ambient
Fig.13. Measured radial displacement distributions after the completion of the whole weld (positive radial displacement represents an increase in pipe diameter)
Fig.14. Distributions of measured residual strains measured after the completion of the weld and plotted against distance from the weld centreline
The residual stresses were measured at three locations. The full through thickness distribution of residual stresses was measured on the weld centreline using the block removal splitting and layering technique.[Leggatt 1996]
The measurements were made on the girth weld at line L5 (Figure 4). Further measurements were made using block removal alone at distances of 35mm and 60mm from the weld centreline. The block removal technique alone measures the linear part of the through thickness residual stress distribution. The linear proportion of the through thickness distribution can be divided into a membrane (the average of the linear distribution) and a bending (the difference between the membrane and the surface value) component. These were determined for both the axial and hoop stress directions and the distributions measured are plotted in Figure 15. A theoretical framework for calculations of these components of residual stress was presented by[Leggatt (1984)].
Fig.15. Measured distribution of the membrane and bending components of welding residual stresses
The full through thickness measured and predicted distributions of the welding residual stresses on the weld centreline are given in Figure 16. The through thickness distributions of both the axial and hoop residual stresses were predicted with good accuracy. The predicted distributions of these stresses are in good correspondence with the measured distributions. The highest tensile values predicted by the model were higher than the measured values. The hoop stress was over-predicted by 8% and the axial stress was over predicted by 30%.
Fig.16. Comparision of the measured and predicted distributions of the welding residual stresses
The error in the predicted peak axial residual stress is large. This difference, however, is unlikely to be significant if the predicted stresses are used for the Engineering Critical Assessment (ECA) using codes like[BS7910 (2005)] or [API 579 (2000)]. This is because the full stress distribution is needed to determine the loading on defects in the weld. It is clear that the predicted distribution is close to the measured distribution so the defect tolerance determined from the measured and predicted distributions of residual stress will be similar.
The measurements show that the pipe behaviour was not axisymmetric during welding. The radial displacements on line L1 are larger than the same measurements made on lines L2 and L3 (Figures 12 and 13). This behaviour may be due to the variation in diameter of one of the pipes before welding (see text above). It may also be related to the non-axisymmetric nature of the weld start and stop positions which were located near line L1.
The model produced a good prediction of the residual stress distribution in the weld even though other comparisons of the model with the test measurements were not good. The comparison of peak temperature was fair, except very close to the weld where the temperature gradients are likely to be very large. The large temperature gradients mean that slight errors in the position of the melt pool in the model and the test could produce significant differences between the results. The predicted evolution of measured strains near the weld (Figure 8) also diverges from the measured history. For hoop strains the major difference occurred during pass 4 and the predicted change in strain per pass is closer to the measured values after that pass. The absolute values of the measured axial strain are much larger than the predictions. Here, the change in strain per pass after pass 10 are similar for both the measurements and the predictions. A more complete comparison of the modelling results with the complete set of measurements is needed. A further three-dimensional model should be run to determine the effect of pipe ovality and the start and stop locations.
Despite the differences in intermediate results the overall outcome of the work is positive. The axisymmetric model has successfully predicted the distribution of residual stresses in the girth butt weld. It is therefore clear that this sort of modelling approach provides the possibility of significant advances in pipeline and riser related ECA studies. BS7910 recommends the use of yield magnitude residual stresses for defect tolerance calculations. The current approach has allowed this assumption to be relaxed for defects in the hoop-radial plane. The standard approach would requires residual stresses of 360MPa whereas the use of a modelling method has allowed the possible assessment using a peak stress of 300MPa and a predominantly compressive stress distribution near the weld root. This result would have a significant effect upon the predicted tolerable defects and is therefore very attractive for future ECA activities.
A comprehensive set of measurements taken during the production of a test weld has been presented and the results have been compared with an FEA based model of the weld. It was found that the predicted distributions of residual stress are conservative (that is they give larger values than the measured distribution) and that these predicted stresses provide considerable benefit for ECA calculations in terms of reducing the required level of conservatism in the assumed levels of residual stresses.
The high temperature tensile testing was done at the University of Nottingham under the supervision of Professor T Hyde.
API 2000: API 579: 'Fitness for service', First edition, American Petroleum Institute.
BS 7910:2005 (Incorporating Amendment No.1): 'Guide to methods for assessing the acceptability of cracks in metallic structures'. British Standards.
Leggatt R H (1984): 'Residual stress at girth welds in pipes', In conference 'Welding in energy related projects', Canada 429-440.
Leggatt R H and Friedman L M (1996): 'Residual weldment stresses in controlled deposition repairs to 11/4Cr-1/2Mo and 21/4-1Mo steels', In conference ASME Pressure Vessels and Piping, Montreal, PVP-Vol 327 127-146.
Simonson J R (1975): 'Engineering Heat Transfer', MacMillan Press Ltd.