Application of ANN concepts for prediction of crack growth and remaining life of circumferentially cracked piping components under different loading scenarios

Abstract

This paper presents the details of Artificial Neural Network (ANN) based models to predict crack depth and remaining life of circumferentially part-through cracked piping components used in the nuclear industry. Crack growth data from experimental studies on (i) straight pipes made up of SA 312 Type 304 LN stainless steel having part-through circumferential notch under combined bending and torsion and (ii) dissimilar metal pipe weld joints having circumferential through-wall crack in the weld were used. The dissimilar metal pipe weld joint is made up of SA312 Type 304 LN austenitic stainless steel and SA508 Gr. 3 Cl. 1 low alloy carbon steel and joined by Nickel-rich Inconel alloy weld. About 75 % of the mixed experimental data has been used for development of model and the remaining data for validation. For the development of ANN model, (i) back propagation technique (ii) sigmoidal transfer functions and (iii) Levenberg-Marquardt algorithm are used. The maximum percentage difference observed between the predicted and the experimental results for crack depth and remaining life was 19.59 % and 15.78 % respectively. The efficiency of the developed model was verified using several statistical parameters. The developed models are useful for the structural integrity assessment of piping components under different loading scenarios.