# Polyester Resin–Quartz Composites in the Age of Artificial Intelligence and Digital Twins: Current Advances, Future Perspectives and an Application Example

## TL;DR

This paper reviews how AI and digital twins can improve the production of polyester resin-quartz composites by enabling data-driven manufacturing.

## Contribution

The paper introduces a practical application of AI models for predicting peak time in quartz–UPR composites and integrates digital twin concepts with materials science.

## Key findings

- Gradient Boosting models outperform Random Forest in predicting peak time with stratified bootstrap augmentation.
- Curing processes are governed by three physical dimensions: reactivity, thermal environment, and resin state.
- AI and digital twins enhance prediction of rheology and quality in thermoset polymer processes.

## Abstract

Unsaturated polyester resin (UPR)–quartz composites have become increasingly important in structural, sanitary, and architectural applications. However, their manufacturing processes still rely heavily on empirical knowledge. This review compiles recent developments in materials science, curing kinetics, and digital manufacturing, outlining a pathway toward data-driven, adaptive production of quartz-filled thermosets. The chemical and physical fundamentals of UPR polymerization are summarized, including the influence of initiator systems, filler characteristics, and thermal management on network formation. Challenges associated with highly filled formulations—such as viscosity control, dispersion, shrinkage, and exothermic peak prediction—are discussed in detail. Recent advances in digital twins (DTs) and artificial intelligence (AI) are reviewed, demonstrating how physics-based simulations, machine learning models, and hybrid mechanistic–data-driven approaches improve the prediction of rheology, curing behavior, and quality outcomes in thermoset polymer processes. A practical application example demonstrates the prediction of peak time in quartz–UPR composites using Random Forest and Gradient Boosting ensemble models. Two prediction scenarios are evaluated: Scenario A with gel time by Leave-One-Out cross-validation, and Scenario B without gel time, representing post-mixing and pre-process prediction contexts, respectively. Stratified bootstrap augmentation improves Gradient Boosting in both scenarios. Principal component analysis confirms that the curing process is governed by three independent physical dimensions: curing reactivity, thermal environment and resin thermal state.