Synergistic Utilisation of Informatics and Data Centric Integrity Engineering (SINDRI)

Abstract Mechanical behaviour of engineering alloys is dependent upon their microstructure which can be affected by processing and heat-treatment. In traditional mechanics the microstructural dependence is simplified by introducing laws that govern the behaviour of continuum solids and adjust their parameters according to the average properties of the material. This approach has served industrial applications well in most cases, however in high value and safety sensitive components, seemingly negligible changes to the microstructure can significantly alter the outcome of a safety assessment analysis making the current average continuum laws over-simplified, overly conservative, and potentially costly. For example, small changes in the welding parameters used in manufacturing a steam generator can affect the texture of the weld, thus influencing its hardening behaviour and residual stress profile which can in turn decide its estimated safe life [1]. For this reason, detailed experimental characterisation techniques combined with microstructural simulations have been developed in the recent decades which can underpin and predict the behaviour of materials at small length scales with precision [2]. The downfall, however, has been the transition of these computationally expensive models along with limited number of detailed characterisation to statistically representative robust models that can predict the behaviour of an industrial scale component. SIN-DRI is a project which aims to tackle this issue. It employs the recent advances in high-throughput characterisation technique such as high angular resolution electron back-scatter diffraction with state-of-the-art modelling. The high-fidelity characterisation data informs computationally expensive microstructural simulation techniques (e.g. crystal plasticity finite element), whose uncertainty is quantified using machine learning algorithms (e.g. Gaussian Process). The resulting approach is a route to build probabilistic models that can capture the variation in the mechanical behaviour of service components informed by the statistical distribution of key features within its microstructure. This paper provides a case study of this approach focusing on residual stress of a weldment. The variability of the weld residual stress as response to changes in its microstructural texture is quantified. The texture variation within a weldment has been extracted. Crystal plasticity analysis is used to connect the variation in texture to changes in the yield stress, simplified through Gaussian Process to avoid performing hundreds of crystal plasticity simulations. The variation in the yield stress is used to estimate the variability in the weld residual stress. It is argued that such changes in the residual stress as function of microstructure can be used within a probabilistic framework to assess the integrity of safety sensitive components.