Graphical Feature Construction-Based Deep Learning Model for Fatigue Life Prediction of AM Alloys

Area of Science:

• Materials Science
• Mechanical Engineering
• Computer Science

Background:

• Fatigue failure is a critical concern for component safety under cyclic loading.
• Current machine learning (ML) models struggle with limited information from purely numeric inputs.
• Predicting fatigue life accurately is essential for reliable engineering applications.

Purpose of the Study:

• To develop a novel machine learning model for enhanced fatigue life prediction.
• To overcome the limitations of traditional ML models relying solely on numeric features.
• To improve the analysis of fatigue failure by incorporating graphical data representations.

Main Methods:

• A new ML model based on convolutional neural networks (CNNs) was developed.
• Numeric features were transformed into graphical representations using Shapley Additive Explanations and Pearson correlation coefficient analysis.
• An attention mechanism was integrated to focus on critical data regions within image-based inputs.

Main Results:

• The proposed model demonstrated superior predictive accuracy compared to conventional ML models.
• Validation using experimental data from laser powder bed fusion-fabricated metals confirmed the model's effectiveness.
• Graphical feature transformation enriched the input data, leading to better fatigue life predictions.

Conclusions:

• The novel CNN-based ML model offers a significant advancement in fatigue life prediction.
• Transforming numeric features into graphical ones enhances the understanding of fatigue failure mechanisms.
• This approach provides a more accurate and robust method for assessing the operational safety of components.