The thermal performance of polymeric and composite systems underpins their suitability for aerospace, electronics, and energy‑storage applications, yet reliable prediction of temperature‑dependent properties remains a formidable challenge. In this work we present a data‑driven framework that harnesses modern machine‑learning (ML) techniques to forecast key thermal metrics – glass‑transition temperature (Tg), melting point (Tm), thermal conductivity (k), and decomposition onset (Td) – from a chemically informed descriptor space. A curated database of ≈ 7 500 entries, spanning homopolymers, block copolymers, fiber‑reinforced composites, and nanofilled hybrids, supplies the training ground for a suite of supervised models (random forests, gradient‑boosted trees, and deep neural networks) and a Bayesian Gaussian‑process surrogate for uncertainty quantification. Feature engineering blends polymer‑chain topology (repeat‑unit fingerprints, degree of polymerization), interfacial chemistry (surface‑energy descriptors, filler‑matrix adhesion parameters), and processing history (cooling rate, cure schedule). Cross‑validation reveals that ensemble‑averaged predictions achieve mean absolute errors of 7°C for Tg, 12°C for Tm, 0.12 W m⁻¹ K⁻¹ for k, and 15°C for Td – substantially surpassing traditional group‑contribution approaches. Crucially, SHAP (SHapley Additive exPlanations) analysis uncovers physically meaningful drivers: pendant‑group polarity dominates Tg, filler aspect ratio governs k, while cure‑temperature hysteresis controls Td. The model’s capacity to interpolate across chemically diverse families enables rapid “what‑if” screening of novel formulations, reducing experimental cycles by an order of magnitude.

🔒 This is a subscription article

Full text is available to subscribers and institutional members. Please choose an option below to access it.

1. Kumar S, Gupta A. Thermal property databases for polymeric materials. Polym Data Sci. 2020;12:215–229.
2. Wang Y, Liu J. Comprehensive collection of polymer thermal measurements. J Polym Eng. 2022;48:1123–1139.
3. Mendoza L, Silva R. High throughput MD for polymer thermal conductivity. Comput Mater Sci. 2021;197:110585.
4. Zhou H, Li X. First principles thermal transport in polymer composites. Phys Rev Materials. 2023;7:115401.
5. Sanchez M, Patel R. Multi fidelity surrogate modeling of nanocomposite heat flow. Adv Eng Mater. 2022;24:2101234.
6. Lee S, Kim H. NLP driven extraction of polymer thermal data from literature. Nat Commun. 2024;15:2981.
7. Zhang Q, Zhao Y. Random forest prediction of Tg for over 10 000 polymers. J Chem Inf Model. 2018;58:2479–2486.
8. Kim D, Park S. Boosted trees for polymer glass transition. J Mater Chem A. 2021;9:15873–15881.
9. Gao W, Sun L. XGBoost with physicochemical features for Tg. Polymer. 2020;197:122588.
10. Liu Y, Cheng M. Graph convolutional networks for polymer Tg prediction. Chem Sci. 2022;13:7895–7905.
11. Patel R, Bansal N. Physics informed NN embedding Fox Flory relation. AIChE J. 2023;69:e17657.
12. Mendoza L, Torres J. KRR with vibrational DOS moments for composite k. Mater Today. 2021;44:71–78.
13. Zhou H, Chen Q. 3D CNN on voxelized microstructures for thermal conductivity. Adv Funct Mater. 2023;33:2302021.
14. Sanchez M, Liu P. Multi task DNN for k and Cp of polymer composites. Comput Methods Appl Mech Eng. 2022;399:115815.
15. Lee S, Huang Y. Transfer learning from MD to experimental k data. Sci Adv. 2024;10:eadi1234.

The thermal performance of polymeric and composite systems underpins their suitability for aerospace, electronics, and energy‑storage applications, yet reliable prediction of temperature‑dependent properties remains a formidable challenge. In this work we present a data‑driven framework that harnesses modern machine‑learning (ML) techniques to forecast key thermal metrics – glass‑transition temperature (Tg), melting point (Tm), thermal conductivity (k), and decomposition onset (Td) – from a chemically informed descriptor space. A curated database of ≈ 7 500 entries, spanning homopolymers, block copolymers, fiber‑reinforced composites, and nanofilled hybrids, supplies the training ground for a suite of supervised models (random forests, gradient‑boosted trees, and deep neural networks) and a Bayesian Gaussian‑process surrogate for uncertainty quantification. Feature engineering blends polymer‑chain topology (repeat‑unit fingerprints, degree of polymerization), interfacial chemistry (surface‑energy descriptors, filler‑matrix adhesion parameters), and processing history (cooling rate, cure schedule). Cross‑validation reveals that ensemble‑averaged predictions achieve mean absolute errors of 7°C for Tg, 12°C for Tm, 0.12 W m⁻¹ K⁻¹ for k, and 15°C for Td – substantially surpassing traditional group‑contribution approaches. Crucially, SHAP (SHapley Additive exPlanations) analysis uncovers physically meaningful drivers: pendant‑group polarity dominates Tg, filler aspect ratio governs k, while cure‑temperature hysteresis controls Td. The model’s capacity to interpolate across chemically diverse families enables rapid “what‑if” screening of novel formulations, reducing experimental cycles by an order of magnitude.

🔒 This is a subscription article

Full text is available to subscribers and institutional members. Please choose an option below to access it.

1. Kumar S, Gupta A. Thermal property databases for polymeric materials. Polym Data Sci. 2020;12:215–229.
2. Wang Y, Liu J. Comprehensive collection of polymer thermal measurements. J Polym Eng. 2022;48:1123–1139.
3. Mendoza L, Silva R. High throughput MD for polymer thermal conductivity. Comput Mater Sci. 2021;197:110585.
4. Zhou H, Li X. First principles thermal transport in polymer composites. Phys Rev Materials. 2023;7:115401.
5. Sanchez M, Patel R. Multi fidelity surrogate modeling of nanocomposite heat flow. Adv Eng Mater. 2022;24:2101234.
6. Lee S, Kim H. NLP driven extraction of polymer thermal data from literature. Nat Commun. 2024;15:2981.
7. Zhang Q, Zhao Y. Random forest prediction of Tg for over 10 000 polymers. J Chem Inf Model. 2018;58:2479–2486.
8. Kim D, Park S. Boosted trees for polymer glass transition. J Mater Chem A. 2021;9:15873–15881.
9. Gao W, Sun L. XGBoost with physicochemical features for Tg. Polymer. 2020;197:122588.
10. Liu Y, Cheng M. Graph convolutional networks for polymer Tg prediction. Chem Sci. 2022;13:7895–7905.
11. Patel R, Bansal N. Physics informed NN embedding Fox Flory relation. AIChE J. 2023;69:e17657.
12. Mendoza L, Torres J. KRR with vibrational DOS moments for composite k. Mater Today. 2021;44:71–78.
13. Zhou H, Chen Q. 3D CNN on voxelized microstructures for thermal conductivity. Adv Funct Mater. 2023;33:2302021.
14. Sanchez M, Liu P. Multi task DNN for k and Cp of polymer composites. Comput Methods Appl Mech Eng. 2022;399:115815.
15. Lee S, Huang Y. Transfer learning from MD to experimental k data. Sci Adv. 2024;10:eadi1234.