Machine learning as a surrogate for FEM: Predicting mechanical properties of tyres

Yang Pei, Bing Han, Dileep Kumar, Scott Adams, Sui Yang Khoo, Michael Norton, Abbas Z. Kouzani,
Machine learning as a surrogate for FEM: Predicting mechanical properties of tyres,
Advanced Industrial and Engineering Polymer Research,
Volume 8, Issue 4,
2025,
Pages 499-515,
ISSN 2542-5048,
https://doi.org/10.1016/j.aiepr.2025.08.003.
(https://www.sciencedirect.com/science/article/pii/S2542504825000399)
Abstract: Accurately characterizing the mechanical properties of tyre materials remains a critical challenge due to their heterogeneous composition and anisotropic behaviour. This study introduces a comprehensive modelling pipeline that integrates composite-informed Finite Element Methods (FEM) with supervised machine learning techniques to enable robust prediction of the mechanical properties of waste tyre materials. Incorporating composite-structure principles into FEM enables the development of a high-fidelity, well-parameterised simulation model capable of generating diverse synthetic datasets. These datasets are used to train and evaluate a range of predictive models, including deep learning architectures such as Long Short-Term Memory (LSTM) and Feedforward Neural Networks (FNN), as well as traditional machine learning algorithms such as Extreme Gradient Boosting (XGBoost). The results demonstrate that deep learning approaches consistently outperform conventional machine learning methods in predictiion accuracy, with XGBoost identified as the most effective among traditional machine learning approaches. Experimental validation confirms the reliability and physical relevance of the proposed framework. Sensitivity and data-efficiency analyses indicate that 500–800 simulated samples are sufficient for accurate predictions. Furthermore, directional dependency in mechanical behaviour is captured, with peak Young's modulus observed along orientations aligned with the tyre's rolling direction. This study establishes a validated, machine-learning-based surrogate modelling approach as a fast and reliable alternative to FEM analysis of complex waste tyre composites for predicting their mechanical properties. Beyond tyre recycling, the methodology provides a scalable and transferable framework for the mechanical characterization of other anisotropic and heterogeneous materials in engineering design and sustainable material reuse.
Keywords: FEM; Machine learning; Surrogate modelling; Composite materials; Young's modulus; Tyres