Influence of Hardening Model on Weld Residual Stress Distribution

This study is the third stage of a project sponsored by the Swedish Radiation Safety Authority (SSM) to improve the weld residual stress modelling procedures currently used in Sweden. The aim of this study was to determine which material hardening model gave the best agreement with experimentally measured weld residual stress distributions. Two girth weld geometries were considered: 19mm and 65mm thick girth welds with Rin/t ratios of 10.5 and 2.8, respectively. The FE solver ABAQUS Standard v6.5 was used for analysis. As a preliminary step some improvements were made to the welding simulation procedure used in part one of the project. First, monotonic stress strain curves and a mixed isotropic/kinematic hardening model were sourced from the literature for 316 stainless steel. Second, more detailed information was obtained regarding the geometry and welding sequence for the Case 1 weld (compared with phase 1 of this project). Following the preliminary step, welding simulations were conducted using isotropic, kinematic and mixed hardening models. The isotropic hardening model gave the best overall agreement with experimental measurements; it is therefore recommended for future use in welding simulations. The mixed hardening model gave good agreement for predictions of the hoop stress but tended to under estimate the magnitude of the axial stress. It must be noted that two different sources of data were used for the isotropic and mixed models in this study and this may have contributed to the discrepancy in predictions. When defining a mixed hardening model it is difficult to delineate the relative contributions of isotropic and kinematic hardening and for the model used it may be that a greater isotropic hardening component should have been specified. The kinematic hardening model consistently underestimated the magnitude of both the axial and hoop stress and is not recommended for use. Two sensitivity studies were also conducted. In the first the effect of using a bilinear model with a saturation strain of 0.01 was evaluated. This model was not capable of accurately predicting the weld residual stress field and its use is not recommended. In the second sensitivity study, the effect of defining a viscoplastic model was evaluated. For predictions of axial and hoop stress, the viscoplastic model was found to give slightly lower stress peaks, although the effect was sufficiently small that the increased complexity and computational cost of defining such a model probably outweighs the benefits. It was noted that a similar effect could be obtained by lowering the annealing temperature.