Machine Learning Method for Fatigue Strength Prediction of Nickel-Based Superalloy with Various Influencing Factors

Area of Science:

• Materials Science
• Mechanical Engineering
• Computational Science

Background:

• Accurate fatigue performance prediction is crucial for component safety and reliability.
• Complex influencing factors and incomplete understanding of fatigue mechanisms hinder accurate prediction.
• Machine learning offers potential for modeling complex, multi-variable relationships in fatigue behavior.

Purpose of the Study:

• To evaluate machine learning models for predicting fatigue strength of GH4169 superalloy.
• To investigate the impact of training data composition on predictive accuracy.
• To explore machine learning as a tool for fatigue performance prediction under varying conditions.

Main Methods:

• Utilized gradient boosting regression tree, long short-term memory, and polynomial regression models.
• Applied machine learning to predict fatigue strength of GH4169 under different temperatures, stress ratios, and fatigue lives.
• Investigated the influence of training/testing set composition on model performance.

Main Results:

• Machine learning models demonstrate significant potential for fatigue strength prediction using limited data.
• Predictive accuracy is closely correlated with the composition and quantity of the training dataset.
• Models struggle to predict anomalous data points absent from the training set.

Conclusions:

• Machine learning provides a promising approach for predicting fatigue performance with complex influencing factors.
• Sufficient and representative training data is essential for enhancing machine learning model predictive capabilities.
• Further research with more abundant data is needed to improve fatigue strength prediction accuracy.