Machine Learning-Based Fatigue Life Prediction for E36 Steel Welded Joints.

Area of Science:

• Materials Science
• Mechanical Engineering
• Computational Science

Background:

• E36 steel is crucial for shipbuilding and offshore structures but suffers from fatigue failure in welded joints.
• Weld toe and heat-affected zone cracks significantly reduce the fatigue life of E36 steel components.
• Existing fatigue life prediction methods for E36 steel welds are complex, inefficient, and lack precision.

Purpose of the Study:

• To develop an efficient machine learning (ML) framework for predicting the fatigue life of E36 steel welded joints.
• To address data scarcity in fatigue testing through data augmentation techniques.
• To compare the performance of various ML models against traditional prediction formulas.

Main Methods:

• A dataset of 23 fatigue test data points for E36 welded joints was established.
• Data augmentation using the Z-parameter model and SMOTE was employed to address data scarcity.
• Machine learning models, including artificial neural networks and random forest, were trained and evaluated on augmented and original datasets.

Main Results:

• Machine learning models significantly outperformed the traditional Z-parameter formula (R² = 0.643, MAPE = 16.15%).
• The artificial neural network achieved the highest accuracy (R² = 0.972, MAPE = 4.45%).
• The random forest model demonstrated robust consistency across validation and testing sets (R² ≈ 0.89).

Conclusions:

• The proposed ML framework provides a more accurate and efficient method for fatigue life prediction in E36 steel welded joints.
• Artificial neural networks and random forest models show significant promise for enhancing fatigue life assessment in structural components.
• This approach can improve the reliability and safety of structures fabricated with E36 steel.