Low-Cycle Fatigue Life Prediction of Titanium-Based Intermetallic Alloys Using Machine Learning and Finite Element Analysis.

Abstract

This study explores the low-cycle fatigue characteristics of three structural components fabricated from Ti₂AlNb-based alloys utilizing Seeger's fatigue life theory and an improved Lemaitre damage evolution model. The validity and accuracy of the simulations based on these theoretical methods are verified by experimental fatigue life tests conducted at high temperatures. Additionally, the potential of employing long short-term memory (LSTM), extreme learning machine (ELM), and partial least squares (PLS) algorithms to predict the high-temperature, low-cycle fatigue life of Ti₂AlNb alloy components is examined. Comparative analyses of the training effectiveness and practical applicability of these machine learning approaches are conducted, demonstrating that ELM exhibits superior predictive capability. This investigation thus provides a practical and efficient predictive methodology for assessing the low-cycle fatigue life of structural components composed of Ti₂AlNb-based alloys.