Using artificial neural networks to predict the fatigue life of carbon and low-alloy steels

Abstract

The ASME Boiler and Pressure Vessel Code contains rules for the construction of nuclear power plant components. Figures I-9.1 through I-9.6 of Appendix I to Section III of the Code specify fatigue design curves for structural materials. However, the effects of light water reactor (LWR) coolant environments are not explicitly addressed by the Code design curves. Recent test data indicate significant decreases in the fatigue lives of carbon and low-alloy steels in LWR environments when five conditions are satisfied simultaneously. When applied strain range, temperature, dissolved oxygen in the water, and sulfur content of the steel are above a minimum threshold level, and the loading strain rate is below a threshold value, environmentally assisted fatigue occurs. For this study, a data base of 1036 fatigue tests was used to train an artificial neural network (ANN). Once the optimal ANN was designed, ANN were trained and used to predict fatigue life for specified sets of loading and environmental conditions. By finding patterns and trends in the data, the ANN can find the fatigue life for any set of conditions. Artificial neural networks show great potential for predicting environmentally assisted corrosion. Their main benefits are that the fit of the data is based purely on data and not on preconceptions and that the network can interpolate effects by learning trends and patterns when data are not available.