Machine Learning-Based Prediction of Mechanical Properties for Large Bearing Housing Castings

Area of Science:

• Materials Science and Engineering
• Mechanical Engineering
• Computational Materials Science

Background:

• Mechanical properties of large bearing housing castings are crucial for industrial equipment reliability and lifespan.
• Traditional prediction methods are costly, sample-limited, and lack generalization.
• Existing methods are insufficient for accurate performance prediction of large castings.

Purpose of the Study:

• To develop an efficient machine learning model for predicting mechanical properties of ZG270-500 cast steel.
• To integrate multivariate data including chemical composition and process parameters for enhanced prediction accuracy.
• To address the research gap in performance prediction for large bearing housing castings.

Main Methods:

• Collected and preprocessed real-world production data, including outlier handling, data balancing, and normalization.
• Implemented and compared four machine learning algorithms: Backpropagation Neural Network (BPNN), Support Vector Regression (SVR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost).
• Utilized SHAP (SHapley Additive exPlanations) value analysis to interpret model predictions and identify key influencing factors.

Main Results:

• The Support Vector Regression (SVR) model demonstrated superior performance under small-sample conditions, achieving a coefficient of determination (R²) between 0.85 and 0.95.
• Achieved low Root Mean Square Errors (RMSE) for key mechanical properties: yield strength (7.59 MPa), tensile strength (7.52 MPa), elongation (0.68%), reduction in area (1.47%), and impact energy (5.51 J).
• Experimental validation confirmed relative errors between predicted and measured values below 4%, and SHAP analysis identified critical process parameters and elemental compositions influencing mechanical properties.

Conclusions:

• The developed data-driven machine learning approach, particularly SVR, provides an efficient and accurate method for predicting mechanical properties of large castings.
• The model offers a theoretical foundation for optimizing manufacturing processes and improving the reliability of bearing housing castings.
• This research successfully addresses limitations of traditional methods and establishes a robust framework for casting performance prediction.