Prediction of mechanical properties of bi-disperse magnetorheological fluids and study on their temperature dependence

Bi-disperse magnetorheological (MR) fluids have shown superior rheological properties in numerous studies. Unfortunately, there are few investigations on the constitutive model development with temperature-related factors, which will hinder its applications in engineering. In this study, bi-disperse MR fluids of different sizes were successfully prepared, and compared the prediction of rheological properties by four machine learning algorithms under different conditions in detail. The results suggest that the Category Boosting (CatBoost) algorithm performs best in predicting rheological properties, attaining a testing set R² of 0.998, root mean square error of 0.388, and mean absolute error of 0.280. In addition, the changes in shear stress of bi-disperse MR fluids at different temperatures were analyzed. At room temperature, bi-disperse MR fluids have better rheological properties, the shear stress of MRFS was 47.34 kPa at the highest magnetic field (426 mT). As the temperature continues to rise, bi-disperse MR fluids become more susceptible, simultaneously, the shear stress reduction amplitude under higher magnetic fields is stronger than low. It is worth mentioning that, in this study, we also analyzed the effect of different temperatures on MR fluids by the Shapley Additive Explanations (SHAP) method for the first time. The analysis results by using the SHAP method are consistent with the systematic analysis, which provides a theoretical basis for the development of constitutive models and the preparation of high-performance MR fluids.