A machine learning model for predicting progressive crack extension based on experimental data obtained using DCPD measurement technique

Abstract

A series of low cycle fatigue loading experiments were performed, using Inconel 718 (IN718) superalloy, to produce time history data using Direct Current Potential Drop (DCPD) approach at a range of temperatures and it was used to train a Bidirectional Long-Short Term Memory Neural Network (BiLSTM) model. The experimental investigations were conducted to understand the statistical nature of crack jumps during the fatigue loading. The BiLSTM model was trained on high sampling rate experimental data from crack initiation up through the Paris regime. For a single sample, the trained machine learning model was able to predict future crack events based on prior crack jumps in the time history. The model was able to predict the progressive crack extension at intermediate temperatures and stress intensities that lie between experimental conditions. The BiLSTM model demonstrated the potential to be used as a tool for future investigation into fundamental mechanisms such as high-temperature oxidation at the crack tips.