Uncertainty quantification in modelling welding residual stress and distortion

Abstract

Currently available welding simulation methodologies provide deterministic results for the best estimate of the input parameters, such as part geometry, processing conditions and material properties. If there is an uncertainty in any of the input parameters, then a reanalysis needs to be performed with perturbed values of each uncertain variable. However, there can be several hundred input parameters; therefore, the use of reanalysis in uncertainty quantification in welding modelling can be time consuming or computationally prohibitive, especially for three-dimensional modelling. This paper explores the application of design sensitivity analysis in quantifying uncertainty in welding residual stress and distortion computations. Analytic sensitivities are computed by direct differentiation, resulting in a very efficient computational approach. The variation of temperature, welding residual stress and distortion with respect to processing parameters is computed from a first order Taylor expansion of the model output. The approach is demonstrated in a three-dimensional model of a singe pass weld and validated by comparing sensitivity analysis results to reanalyses.

Keywords:

• Uncertainty quantification
• Residual stress
• Distortion
• Modelling