Enhancing prediction of magnetic properties in additive manufacturing products through a 3D convolutional vision transformer model

With the advancement of metal additive manufacturing technology, selective laser melting (SLM) has gained significant prominence in industrial manufacturing. However, traditional methods for measuring magnetic properties need to improve efficiency and accuracy for modern manufacturing demands. This study employs a 3D convolutional vision transformer (3D-CvT) model to rapidly and accurately predict magnetic properties in products created through the SLM process. The 3D-CvT model merges the advantages of convolutional neural networks and vision transformers, enhancing the understanding of spatial and feature information. As a result, it achieves higher accuracy and efficiency in predicting magnetic properties compared to traditional machine learning methods. Utilizing heatmap technology, this model visually displays areas of the image that significantly impact prediction outcomes. These heat maps facilitate an effective understanding of how image features influence the magnetic properties of the products. Experiments were conducted on 200 specimens, with the results indicating that the 3D-CvT model’s predictions showed low mean square error (MSE), low mean absolute error (MAE), and high R-squared values compared to the actual measured values (ground truth). This indicates strong consistency between the predicted and measured magnetic properties. Additionally, a Student’s t-test was performed, and the corresponding p-values exceeded 0.05, suggesting no statistically significant difference between the predicted and actual measured values for the target population.