Gaussian Process Regression-based Knowledge Distillation Framework for Simultaneous Prediction of Physical and Mechanical Properties of Epoxy Polymers

Sindu B.S, Jan Hamaekers · 2026

Read Gaussian Process Regression-based Knowledge Distillation Framework for Simultaneous Prediction of Physical and Mechanical Properties of Epoxy Polymers by Sindu B.S, Jan Hamaekers (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-16925.txt.

Introduction

Sindu and Hamaekers predict many epoxy polymer properties at once with a distilled learning framework. Epoxy polymers are widely used, but curated datasets are scarce and their structure is complex. Their method trains single property Gaussian process models as teachers. A unified neural network student then learns from all teachers together. The target property is fed in as an input, so the student can exploit links between properties. Features come from SMILES strings through RDKit. The authors report better accuracy than conventional models.

Catalog record

CREATORSindu B.S, Jan Hamaekers
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE25802
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.16925

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Gaussian Process Regression-based Knowledge Distillation Framework for Simultaneous Prediction of Physical and Mechanical Properties of Epoxy Polymers Sindu B.S, Jan Hamaekers Abstract Epoxy polymers are widely used due to their multifunctional properties, but machine learning (ML) applications remain limited owing to their complex 3D molecular structure, multi-component nature, and lack of curated datasets. Existing ML studies are largely restricted to simulation data, specific properties, or narrow constituent ranges. To address these limitations, we developed an informed Gaussian Process Regression-based Knowledge Distillation (GPR-KD) framework for predicting multiple physical (glass transition temperature, density) and mechanical properties (elastic modulus, tensile strength, compressive strength, flexural strength, fracture energy, adhesive strength) of thermoset epoxy polymers. The model was trained on experimental literature data covering diverse monomer classes (9 resins, 40 hardeners). Individual GPR models serve as teacher models capturing nonlinear feature-property relationships, while a unified neural network student model learns distilled knowledge across all properties simultaneously. By encoding the target property as an input feature, the student model leverages cross-property correlations. Molecular-level descriptors extracted from SMILES representations using RDKit create a physics-informed model. The framework combines GPR interpretability and robustness with deep learning scalability and generalization. Comparative analysis demonstrates superior prediction accuracy over conventional ML models. Simultaneous multi-property prediction further improves accuracy through information sharing across correlated properties. The proposed framework enables accelerated design of novel epoxy polymers with tailored properties. 1 Introduction Epoxy polymers are thermoset materials exhibiting with multi-functional properties such as high strength, excellent adhesion, effective electrical insulation, low shrinkage, chemical and solvent resistance, etc. They are widely used in several industries like aerospace, marine, automotive, infrastructure, electrical and electronics for specific functional requirement. For instance, adhesive strength, fracture toughness and weight are the major requirements for use in aerospace applications [ 15 ] ; resistance to moisture, salinity and varied temperature cycles is the major requirement in marine applications; viscosity, low shrinkage and corrosion resistance are the requirements for its application in paint and coatings [ 15 ] ; insulation resistance is the major requirement for electrical applications [ 42 ] ; adhesive strength, durability, tensile and fracture properties are the requirements for infrastructural applications [ 38 ] . Hence, care should be taken to design epoxy polymers with required functionalities so that it can be used for the intended purposes. The basic constituents of epoxy polymers are resin and hardener. When these two compounds undergo an irreversible reaction, a dense, three dimensional network is formed. However, the functionalities and the properties of epoxy polymers depend upon several factors like the constituent resin and hardener types, their composition, curing conditions and degree of polymerization. The conventional trial-and-error-based experimental investigations limit the development of high performance epoxy polymers with multi-functional characteristics. A few novel approaches were developed in the recent times to establish constituent-structure-property relationship of epoxy polymers and to predict its properties which can enrich the design process. Results from experimental nanoindentation [ 54 , 32 ] tests and scratch induced deformation tests [ 28 ] were used to establish a correlation with the intrinsic mechanical properties. Computational molecular dynamics (MD) simulations were performed to determine the mechanical, physical and transport properties of epoxy polymers [ 45 ] . MD simulations using Reactive Interface Force Field (IFF-R) were used to determine the properties of epoxy polymers when subjected to large deformations [ 30 ] ; to get the experimental results that can be obtained using Differential Scanning Calorimetry (DSC) [ 31 ] by utilizing information such as heat of formation and activation energy. MD simulations were also used to identify the influence of certain molecular interactions on mechanical and transport properties of polymers [ 31 , 44 ] , to correlate the functionality of monomers (di-, tri-, tetra-) with the mechanical properties like yield stress, elastic modulus and Poisson’s ratio [ 37 ] . ML-based approaches are being widely used for design, optimization and property prediction of homopolymers and copolymers due to the availability of large databanks [ 1 , 53 , 21 , 47 ] . However, their use in thermoset epoxy polymers is very limited due to the involvement of two or more constituent ingredients, complex three-dimensional structure and lack of large curated datasets. Recently, attempts are being made for composition optimization, property prediction and design of high performance epoxy polymers using ML-based approaches. Results from MD simulations in conjunction with Neural Networks (NNs) was used to optimize the composition of epoxy polymers with performance characteristics like elastic modulus, tensile strength, elongation at break, glass transition temperature ( \mathrm{Tg} ) [ 16 ] and self-healing properties [ 26 ] . The composition of shape memory epoxy polymers (SMP) was also optimized using NNs using the results from experimental investigations [ 24 ] . Unified ML model based on Support Vector Regression (SVR) was proposed to predict the \mathrm{Tg} of homo-, hetero- and cross linked epoxy- polymers based on the constituent monomers and its descriptors [ 11 ] . ML ensemble model based on Gradient Boosting Regression (GBR) and Kernel Ridge Regression (KRR) was used to predict the \mathrm{Tg} of epoxy polymers by correlating the molecular descriptors of resin (16 types) and hardener (19 types) with experimental (94 combinations) DMTA measurements and the most important descriptors affecting the \mathrm{Tg} were further identified using Lasso regression [ 27 ] . Materials genome approach in conjunction with attention- and gate-augmented graph convolutional networks, multilayer perceptrons and transfer learning was used to identify the gene structures responsible for different properties and to design epoxy polymers with high strength, modulus and toughness [ 12 ] . Similarly, ML-based approach was developed to predict the elastic modulus and yield strength of epoxy polymers using the basic structural features of monomers, thereby, enabling feature-based prediction of their properties [ 43 ] . ML-based convolutional model was employed for discovery of new thermoset SMPs with high recovery stress from a newly constituted compositional space [ 52 , 51 ] . Though several attempts have been made to use ML for design of epoxy polymer or prediction of properties, most of the approaches seem to be limited as the data for the model are obtained from computational simulations or the model has been developed to predict a specific property of epoxy polymers or with very limited set of constituents. With this in mind, we develop a ML model which can be used to predict multiple physical (glass transition temperature and density) and mechanical (elastic modulus, tensile strength, compressive strength, flexural strength, fracture energy and adhesive strength) properties of two-component, thermoset epoxy polymers merely with the help of its constituent ingredients (type of resin and hardener and its proportion). We develop a GPR-based knowledge distillation framework (GPR-KD) for this purpose in which individual GPR models are trained as teacher models for each target property, and the the knowledge learned from these teachers is distilled into a unified student model for property prediction. The constituent resin and hardener are first encoded using a label encoding scheme to numerically represent their chemical identities. These encoded variables are then used together with the relevant process parameters as inputs to the model, allowing it to learn correlations between material composition, processing conditions, and the resulting physical and mechanical properties (cf. Section 2.1 ).

Read the complete text