Machine-Learning-Enhanced Trial-and-Error for Efficient Optimization of Rubber Composites

The traditional trial-and-error approach, although effective, is inefficient for optimizing rubber composites. The latest developments in machine learning (ML)-assisted methodologies are also not suitable for predicting and optimizing rubber composite properties. This is due to the dependency of the properties on processing conditions, which prevents the alignment of data collected from different sources. In this work, a novel workflow called the ML-enhanced trial-and-error approach is proposed. This approach integrates orthogonal experimental design with symbolic regression (SR) to effectively extract empirical principles. This combination enables the optimization process to retain the characteristics of the traditional trial-and-error approach while significantly improving efficiency and capability. Using rubber composites as the model system, the ML-enhanced trial-and-error approach effectively extracts empirical principles encapsulated by high-frequency terms in the SR-derived mathematical formulas, offering clear guidance for material property optimization. An online platform has been developed that allows for no-code usage of the proposed methodology, designed to seamlessly integrate into the existing experimental optimization process.