Prediction of Temperature and Moisture Concentration in Autoclave-Cured Epoxy Resin Using Physics-Informed Neural Networks

Abstract:

Epoxy-based composites used in the aerospace industry are highly sensitive to moisture absorption, which can lead to porosity formation during the curing process and compromise structural integrity. Therefore, accurate prediction of temperature fields, degree of cure, and moisture concentration is essential for process optimization and defect mitigation. However, classical numerical approaches for solving the coupled governing equations are computationally expensive, limiting their applicability in real-time analyses and optimization strategies. In this work, Physics-Informed Neural Networks (PINNs) are investigated for predicting the transient thermal behavior, cure kinetics, and moisture concentration in an epoxy composite laminate during autoclave curing. Two PINNs are developed: the first solves the coupled transient heat transfer and cure kinetics equations in a compositetooling system, while the second predicts the moisture concentration field in the laminate using the temperature information provided by the first network. Different network architectures are evaluated, and their performance is compared with numerical solutions obtained via the Finite Volume and Finite Element Methods. The results demonstrate that PINNs accurately reproduce temperature profiles, degree of cure, and moisture concentration, achieving high coefficients of determination, while also providing significant computational efficiency advantages during the prediction stage. These findings highlight the potential of PINNs as a robust and efficient tool for modeling complex coupled phenomena in composite manufacturing processes.

• Periodical: Defect and Diffusion Forum (Volume 451)
• Pages: 75-85
• DOI: https://doi.org/10.4028/p-12AeQc
• Online since: April 2026
• Authors: Wancley O. Pedruzzi, Igor F. Tosi, William Pablo Montes Quiroz, Raffaele D'Elia, Julio Cesar Sampaio Dutra, Wellington B. da Silva*
• Keywords: Automotive Composites, PINNs, Porosity, Void Growth

References

[1] Andrea Dei Sommi, Francesca Lionetto, Giuseppe Buccoliero, and Alfonso Maffezzoli. The ef fect of absorbed moisture and resin pressure on porosity in autoclave cured epoxy resin. Polymer Composites, 45(17):15793–15803, August 2024.

DOI: 10.1002/pc.28870

[2] Andrea Dei Sommi, Giuseppe Buccoliero, Francesca Lionetto, Fabio De Pascalis, Michele Nacucchi, and Alfonso Maffezzoli. A finite element model for the prediction of porosity in autoclave cured composites. Composites Part B: Engineering, 264:110882, September 2023.

DOI: 10.1016/j.compositesb.2023.110882

[3] M. Raissi, P. Perdikaris, and G.E. E Karniadakis. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differ ential equations. Journal of Computational Physics, 378:686–707, 2019.

DOI: 10.1016/j.jcp.2018.10.045

[4] Maziar Raissi, Paris Perdikaris, and George Em Karniadakis. Physics Informed Deep Learning (Part I): Data-driven Solutions of Nonlinear Partial Differential Equations, 2017.

[5] Maziar Raissi, Paris Perdikaris, and George Em Karniadakis. Physics Informed Deep Learning (Part II): Data-driven Discovery of Nonlinear Partial Differential Equations, 2017.

[6] Jingjing Liu, Yefeng Liu, and Qichun Zhang. A weight initialization method based on neural network with asymmetric activation function. Neurocomputing, 483:171–182, April 2022.

DOI: 10.1016/j.neucom.2022.01.088

[7] Yash Srivastava, Vaishnav Murali, and Shiv Ram Dubey. A Performance Evaluation of Loss Functions for Deep Face Recognition, page 322–332. Springer Singapore, 2020.

[8] S.H. Shabbeer Basha, Shiv Ram Dubey, Viswanath Pulabaigari, and Snehasis Mukherjee. Im pact of fully connected layers on performance of convolutional neural networks for image clas sification. Neurocomputing, 378:112–119, February 2020.

DOI: 10.1016/j.neucom.2019.10.008

[9] Qi Xu, Ming Zhang, Zonghua Gu, and Gang Pan. Overfitting remedy by sparsifying regulariza tion on fully-connected layers of cnns. Neurocomputing, 328:69–74, February 2019.

DOI: 10.1016/j.neucom.2018.03.080

[10] Musa R. Kamal. Thermoset characterization for moldability analysis. Polymer Engineering amp; Science, 14(3):231–239, March 1974.

DOI: 10.1002/pen.760140312

[11] Panagiotis I. Karkanas and Ivana K. Partridge. Cure modeling and monitoring of epoxy/amine resin systems. i. cure kinetics modeling. Journal of Applied Polymer Science, 77(7):1419–1431, 2000.

DOI: 10.1002/1097-4628(20000815)77:7<1419::aid-app3>3.0.co;2-n

[12] Zaharaddeen Karami Lawal, Hayati Yassin, Daphne Teck Ching Lai, and Azam Che Idris. Physics-informed neural network (pinn) evolution and beyond: A systematic literature review and bibliometric analysis. Big Data and Cognitive Computing, 6(4):140, 2022.

DOI: 10.3390/bdcc6040140

[13] Xavier Glorot and Yoshua Bengio. Understanding the difficulty of training deep feedforward neural networks. In Yee WhyeTehandMikeTitterington, editors, Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, volume 9 of Proceedings of Machine Learning Research, pages 249–256, Chia Laguna Resort, Sardinia, Italy, 13–15 May 2010. PMLR.

[14] Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization, 2014.

[15] J. L.Kardos, M. P.Duduković,and R. Dave. Void growth and resin transport during processing of thermosetting — Matrix composites, page 101–123. Springer Berlin Heidelberg, 1986.

DOI: 10.1007/3-540-16423-5_13

[16] Y. Ledru, G. Bernhart, R. Piquet, F. Schmidt, and L. Michel. Coupled visco-mechanical and diffusion void growth modelling during composite curing. Composites Science and Technology, 70(15):2139–2145, December 2010.

DOI: 10.1016/j.compscitech.2010.08.013