A machine learning platform for development of low flammability polymers

2025-03-31

AI Summary

This study addresses the dual challenges of low predictive accuracy and scarcity of experimental data for key polymer fire-safety parameters—flammability index (FI), peak heat release rate (PHRR), time to ignition (TTI), total smoke release (TSR), and fire growth rate (FIGRA). We propose a novel machine learning framework integrating synthetic data augmentation with multi-source molecular descriptors. Specifically, we leverage the Synthetic Data Vault to generate high-fidelity synthetic polymer datasets, combine custom polymer topological descriptors with RDKit-based chemical features, and implement an ensemble multi-model regression architecture. The resulting POLYCOMPRED module is deployed on the MatVerse cloud platform as an open-source, interactive web tool, enabling high-accuracy fire-safety prediction under low-data conditions (mean R² > 0.92). To our knowledge, this work establishes the first end-to-end, machine learning–driven paradigm for inverse design of low-flammability polymers.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Multi-modal Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

Flammability index (FI) and cone calorimetry outcomes, such as maximum heat release rate, time to ignition, total smoke release, and fire growth rate, are critical factors in evaluating the fire safety of polymers. However, predicting these properties is challenging due to the complexity of material behavior under heat exposure. In this work, we investigate the use of machine learning (ML) techniques to predict these flammability metrics. We generated synthetic polymers using Synthetic Data Vault to augment the experimental dataset. Our comprehensive ML investigation employed both our polymer descriptors and those generated by the RDkit library. Despite the challenges of limited experimental data, our models demonstrate the potential to accurately predict FI and cone calorimetry outcomes, which could be instrumental in designing safer polymers. Additionally, we developed POLYCOMPRED, a module integrated into the cloud-based MatVerse platform, providing an accessible, web-based interface for flammability prediction. This work provides not only the predictive modeling of polymer flammability but also an interactive analysis tool for the discovery and design of new materials with tailored fire-resistant properties.

Problem

• Predict flammability metrics of polymers using machine learning
• Overcome limited experimental data with synthetic polymer generation
• Develop web-based tool for flammability prediction and material design

Innovation

• Machine learning predicts polymer flammability metrics
• Synthetic Data Vault augments experimental dataset
• POLYCOMPRED enables web-based flammability prediction