Researchers at the IMDEA Materials Institute developed a machine-learning model to predict the fire resistance of epoxy resins, replacing slow laboratory testing with digital screening. The tool accelerates the design of safer materials for construction, automotive, and aerospace sectors by accurately forecasting how polymer blends will behave in a fire.
Epoxy resins are standard in heavy industry, but their flammability restricts use in critical applications. Manufacturers typically add phosphorus-based flame retardants to mitigate this risk, but developing effective formulas requires slow and expensive physical testing.
"The traditional development of efficient flame retardants involves a design, synthesis and laboratory testing process that is slow, costly and highly sensitive to experimental conditions," said Dr. Qiong Tan, a postdoctoral researcher at IMDEA. The new model bypasses physical synthesis for initial screening by training on data from 510 epoxy composite samples containing these additives.
Predicting fire behavior
The algorithm analyzes the molecular structure of the flame retardants and the specific formulation of the resin. It predicts two standard industry metrics: the UL-94 vertical flammability rating and the Limiting Oxygen Index (LOI).
The UL-94 test measures burning behavior, afterflame time and flame-dripping characteristics under a standardized ignition source. The LOI calculates the minimum oxygen concentration required to sustain combustion.
A unified classification framework
The system goes beyond raw predictions by organizing results into a four-tier classification scale. Materials are categorized as excellent, good, moderate or poor based on their flame-retardant performance.
This framework gives engineers direct guidance when selecting polymers, allowing them to prioritize the most promising candidates before moving to the lab. The model's accuracy was confirmed through external case studies.
Prof. De Yi Wang leads the High-Performance Polymers and Fire Retardants Research Group at IMDEA. The team plans to grow the database to include other polymer types and flame retardant chemistries. Research groups adopting similar methods often rely on specialized training, such as an AI Learning Path for Research Scientists, to integrate machine learning into experimental workflows.
Why this matters for science and research professionals
This model demonstrates how machine learning can compress years of physical material testing into a digital screening process. For research scientists, it offers a blueprint for training predictive models on specific, high-value experimental datasets like flammability metrics.
Applying these methods to other material properties could reduce the time and cost required to bring new, safe polymers to market. Professionals looking at similar applications can review resources for AI for Science & Research to see how machine learning integrates into material design.
Your membership also unlocks:
