Data-Driven Prediction of Dwell Debit in Titanium Alloys: Decoupling Creep-Fatigue Interactions via Machine Learning

Abstract

A data-driven machine learning (ML) framework has been developed to predict dwell debit in titanium alloys processed through both conventional and additive manufacturing routes. The dataset combines newly generated experimental data with literature data and includes feature descriptors related to alloy composition, heat treatment condition, mechanical loading parameters for the fatigue life and creep rupture time. Principal component analysis further demonstrates that composition-processing-loading interactions dominate the variance of the dataset. Different Machine learning algorithms, including Linear Regression, Support Vector Regression, Random Forest, Gradient Boosting, Extreme Gradient Boosting, and Adaptive Boosting, has been evaluated. Among these models, the AdaBoost algorithm provided the highest test accuracy for dwell debit prediction. SHapley Additive exPlanations (SHAP) has been employed to quantify the influence of each feature on the output variable. The analysis reveals that defect size, Nb concentration, annealing temperature, and dwell time are the most influential parameters governing dwell fatigue behaviour. The results indicate that alloy the composition and processing parameters implicitly capture microstructural effects, enabling reliable prediction of dwell fatigue behaviour without explicit microstructural descriptors. The proposed framework provides a computationally efficient approach for accelerated screening and optimization of titanium alloys for improved dwell fatigue resistance.