On the application of principles of artificial intelligence for eigenstrain reconstruction of volumetric residual stresses in non-uniform Inconel alloy 740H weldments

Abstract

The eigenstrain theory provides a range of fruitful concepts for advanced modelling of the behaviour of materials and components obtained using sophisticated manufacturing routes, their response to thermal and mechanical loading, and deformation under fatigue and creep conditions. In recent years the method has been shown to be able to provide predictions of residual stresses for a limited range of processing and simulated service conditions for which experimental data is available. The authors recently presented advances in the use of eigenstrain-based analysis to include accurate determination of the domain and boundaries of eigenstrain fields in the weld zone. This approach allowed effective modelling of large-scale components and the determination of volumetric distributions of residual stresses through the use of additional model coefficients that need to be determined. Due to the non-linear dependence of the prediction on these parameters, the algorithm of the decision-making process has a profound influence on the cost of the simulation, and the reliability of its output. To address this challenge, the principles of Artificial Intelligence were adopted for use in the eigenstrain contour method to develop fuzzy Finite Element Model (fFEM) for the eigenstrain the reconstruction of residual stresses in large structures. The deterministic finite element eigenstrain model uses contour measurements for reconstruction process, and the developed fFEM behaves as an artificial agent to determine the coefficients of the deterministic finite element eigenstrain model. As an example application, as-welded and post-weld heat-treated specimens of non-uniform weldments of Inconel Alloy 740H were investigated using the proposed model. The results were verified using displacement measurements and residual stress calculations of the contour method. The determination of model coefficients by artificial agent allowed effective reconstruction of volumetric residual stresses in complex shaped components using limited data without the requirement of costly and destructive multi-cut experimental procedures.