Characterising residual stresses in repair weld pipes using machine learning

Welding is a widely used joining technique for safety-critical components in nuclear power plants, particularly for pressure vessels and piping assemblies. Weld repair is often performed either during fabrication to mitigate manufacturing defects or during service to restore components to their original design life. However, the welding process involves highly localised thermal cycles, which can generate significant tensile residual stresses in the weld region. When combined with operational loading, these residual stresses may lead to unexpected failure, such as cracking, at loads well below design limits. Consequently, for safety critical components it is important to accurately characterise residual stresses generated by the repair welds for structural integrity assessments. Recently, machine learning techniques have emerged and have shown potential for enhancing weld residual stress characterisation using the data from measurements and FEA predictions. However, its application to predicting residual stresses in repair-welded pipe has been minimal. In this work, the feasibility of using machine learning to achieve accurate, spatially resolved, full field prediction of residual stress in repair welds for safety critical nuclear components is investigated.