Performance Prediction and Process Optimization of Aging-Resistant Rubber-Modified Asphalt via Enhanced BP Neural...

Area of Science:

• Materials Science
• Civil Engineering
• Computational Intelligence

Background:

• Predicting anti-aging rubber asphalt performance is challenging due to complex parameter-property relationships.
• Bio-inspired algorithms offer potential for improving predictive accuracy in material science.

Purpose of the Study:

• To develop and evaluate hybrid models integrating bio-inspired algorithms with neural networks for multi-target prediction of anti-aging rubber asphalt.
• To identify key preparation parameters influencing asphalt performance and optimize them for enhanced stability and resistance.

Main Methods:

• Integration of Crested Porcupine Optimizer (CPO) and Dung Beetle Optimizer (DBO) with Backpropagation (BP) neural networks to create CPO-BP and DBO-BP models.
• Application of Shapley Additive Explanations (SHAP) for identifying influential factors and feature interactions.
• Optimization using Non-dominated Sorting Genetic Algorithm II (NSGA-II) and microstructural characterization via Atomic Force Microscopy (AFM).

Main Results:

• The CPO-BP model demonstrated superior predictive accuracy compared to standard BP and DBO-BP models.
• Mixing temperature was identified as the most influential factor, affecting rutting resistance, ductility, and aging resistance.
• Synergistic effects between mixing temperature and shear time, and coupling effects between rubber content and temperature were observed.

Conclusions:

• Hybrid bio-inspired neural network models provide a robust framework for predicting anti-aging rubber asphalt performance.
• Data-driven prediction and multi-objective optimization are crucial for the rational design of high-performance materials.
• Understanding parameter interactions is key to optimizing asphalt properties for diverse performance requirements.