|About this Abstract
||MS&T21: Materials Science & Technology
||Late News Poster Session
||LSTM Model to Predict Low-cycle Fatigue in IN718
||Jacob Keesler-Evans, Ansan Pokharel, Terence Musho
|On-Site Speaker (Planned)
Understanding low-cycle fatigue is critical to designing new alloys that can operate in extreme environments. The following study uses time history data collected from a creep-fatigue experiment at a range of temperatures to train a Long-Short Term Memory Neural Network (LSTM) to predict crack dynamics. This study focuses on a polycrystalline nickel-based superalloy, known as Inconel 718. The LSTM model is trained on data taken from a DCPD measurement at a high sampling rate. The model is trained on data from crack initiation up through stage two crack propagation. The LSTM model was able to reproduce creep-fatigue data for a continuous range of temperatures and stress intensities. The LSTM model provides a tool for bettering understanding crack initiation and propagation in the absence of experimental data. Moreover, the model can also provide better damage prediction provided loading and thermal history of parts.