Gaussian Process for Machine Learning-Based Fatigue Life Prediction Model under Multiaxial Stress-Strain Conditions

Area of Science:

• Materials Science
• Mechanical Engineering
• Computational Science

Background:

• Predicting material fatigue life under multiaxial stress-strain conditions is crucial for engineering safety.
• Existing parametric fatigue models often struggle with diverse material and loading scenarios.
• Developing robust fatigue prediction models remains a significant challenge.

Purpose of the Study:

• To develop a novel machine learning-based method for fatigue life prediction under multiaxial stress-strain conditions.
• To incorporate physics-based stress and strain invariants into a Gaussian process regression model.
• To enhance the accuracy and applicability of fatigue life predictions for engineering materials.

Main Methods:

• Utilized machine learning, specifically Gaussian process regression, to construct a fatigue prediction model.
• Employed physics-based stress and strain invariants as input features to capture fatigue failure mechanisms.
• Validated the model using experimental data for CuZn37 brass under various cyclic loadings, including non-proportional paths.

Main Results:

• The developed machine learning model successfully predicted fatigue life under complex multiaxial stress-strain conditions.
• The model demonstrated superior performance compared to established parametric fatigue models.
• Physics-based invariants effectively represented fatigue failure mechanisms within the Gaussian process framework.

Conclusions:

• The proposed machine learning approach offers a more accurate and adaptable solution for fatigue life prediction.
• Gaussian process regression with physics-informed inputs provides a powerful tool for materials fatigue analysis.
• This method overcomes limitations of traditional models in handling complex loading conditions and material behaviors.