Integrating semi-analytical modeling and machine learning for dynamic behavior prediction of short fiber reinforced viscoelastic microbeams

Abstract

In this study, an innovative approach combining semi-analytical modeling with machine learning-based prediction methods is proposed to analyze the dynamic behavior of short fiber-reinforced viscoelastic microbeams. Developed based on the modified couple stress theory, the model examined the vibration characteristics of the system by considering microstructure effects and the damping mechanism. The analysis results show that the viscous damping coefficient is the most influential parameter in the frequency behavior of the system, while the scale parameter increases the frequencies by increasing microstructural stiffness. To evaluate the complex and nonlinear interactions of different parameters, a data set consisting of 5000 samples was created using uniform random sampling, and various machine learning algorithms were compared. The results showed that linear models were insufficient, while the optimized Artificial Neural Network (ANN) model provided the highest prediction performance with R 2  = 0.999 and MAPE = 3.46%. SHAP analysis validated the model’s internal consistency and explained the dominant effects of the damping and scaling parameters. In this context, the study presents a hybrid modeling strategy that can predict the complex dynamic responses of micromechanical systems with high accuracy.