Modeling and computation by artificial neural network of fracture toughness of low alloy steel to study the effect of alloy composition

Abstract

The unpredictable structure failures of carbon steel and low alloy steel leading to accidents may be caused by the propagation of a flaw or crack already present in the structure. Fracture toughness which describes the ability of a material containing a crack to resist fracture is one of the most important material properties for design applications of metallic structures. Since this material property is influenced by several parameters, namely material chemistry, heat treatment, morphology of structure, it requires millions of experiments to be conducted to understand and predict it. So, mathematical modeling is one of the solutions to find the effect of these parameters and design future alloys. Stress–intensity factor [Formula: see text] is a quantitative parameter of fracture toughness determining a maximum value of stress which may be applied to a specimen containing a crack (notch) of a certain length. An artificial neural network (ANN) model was developed using over 100 sets of data to study the effect of alloying elements on fracture toughness, [Formula: see text] for the low alloy steel. 20% of data was used for training, 60% to develop predictive model and rest of the 20% for validation. The model can predict the fracture toughness of unknown new data close to 80% accuracy which is good enough for statistical modeling. The details of program code with ANN modeling steps have been explained. Prediction of fracture toughness by the model with variation of alloy composition as well as yield stress gives interesting and important information which may help in designing alloy which will resist crack propagation in a structure and hence enhance the life of structure to fail.