Experimental study and machine learning model to predict formability of magnesium alloy sheet.

Area of Science:

• Materials Science
• Mechanical Engineering
• Computational Materials Science

Background:

• Magnesium AZ31-H24 alloy sheets offer a lightweight, moderately strong material for automotive and aerospace applications.
• Assessing material formability through experimental stretch forming tests is crucial for these industries.

Purpose of the Study:

• To explore and develop machine learning models for predicting the formability of magnesium alloy sheets.
• To address the gap in recent literature regarding machine learning applications in magnesium alloy formability prediction.

Main Methods:

• Conducted Nakazima tests on rectangular magnesium AZ31-H24 alloy samples at varying temperatures (24°C, 250°C, 350°C) and strain rates (0.01 mm/s, 0.001 mm/s) using a hemispherical punch.
• Developed and compared three machine learning models: Random Forest, Extreme Gradient Boosting, and Multiple Linear Regression to predict material formability.

Main Results:

• The Random Forest model demonstrated a high prediction accuracy of 96% for magnesium alloy formability.
• The study successfully established predictive models for material formability based on experimental data.

Conclusions:

• Machine learning significantly minimizes the requirement for physical experiments in material formability studies.
• The developed models offer a computationally efficient approach to understanding and predicting the formability of magnesium alloys.