Neural network based toughness prediction in HAZ of low alloy steel produced by temper bead welding repair technology

Abstract

The pressure vessels of aged nuclear power plants are needed to repair or maintain, and temper bead welding (TBW) is one effective repair welding method instead of post weld heat treatment. For TBW, toughness is the key criteria to evaluate the tempering effect. A neural network based method for toughness prediction in heat affected zone (HAZ) of low alloy steel has been investigated to evaluate the tempering effect in TBW. On the basis of experimentally obtained database, the new toughness prediction system was constructed by using radial basis function neural network. With it, the toughness distribution in HAZ of TBW was calculated based on the thermal cycles numerically obtained by finite element method (FEM). The predicted toughness was in good accordance with the experimental results. It follows that our new prediction system is effective for estimating the tempering effect during TBW and hence enables us to assess the effectiveness of TBW before the actual repair welding.