Bending Strength Prediction of the Cu-Sn Alloy Through a Visual Quantization Model Integrated With Microstructure Characterization and Machine Learning

The multimodal properties of the grinding wheel matrix significantly impact grinding performance, while research on the interactions among these properties remains notably limited. To investigate the latent relationship between the microstructure and the bending strength of the bronze matrix, a visual quantization model based on the microstructure of the Cu-Sn alloy samples was established. The proposed model integrated a image segmentation network module, a quantitative characterization module, and a multivariate prediction module. The enhancement of the segmentation network is based on the synergistic combination of full-scale feature fusion with attention mechanism. The quantitative characterization parameters of metallographic microstructure features are proposed, and the most prominent intercorrelation between these parameters is studied from multiple dimensions. The results show that the modified image segmentation network exhibits superior performance compared to Unet, as evidenced by a 3% increase in Mean Intersection over Union (MIOU). The optimized output strategy ( $\eta Dd$ -PSO-SVR) can contribute to the model’s prediction accuracy of material bending strength (MSE = 23.558,R $^{2}=0.934$ ). Finally, this work shows that the microscopic information demonstrates great adaptability for the machine learning models in predicting bending strength.