Page 28 - IJAMD-2-3

International Journal of AI for
Materials and Design AI applications in composite materials

in defect detection. The approach demonstrates a good temperature profile for CFRP curing, resulting in a 40%
balance between real-time performance and accuracy, reduction in curing time compared to traditional methods.
optimizing the AFP part quality and enhancing the Furthermore, the study employed Bayesian optimization
overall process efficiency. In addition, the integration of to adjust the mold thickness based on the geometry of the
decision support systems in AFP would be beneficial, as CFRP part, improving heat transfer and preventing local
these systems assist in defect detection, segmentation, and overheating. This approach demonstrates the potential of
precision assessment, further streamlining the process. reinforcement learning as autonomous AI optimizes both
In terms of process optimization, a theory-guided process time and energy consumption, thereby significantly
probabilistic ML (TGML) approach has been proposed enhancing the efficiency of the composite curing process.
for minimizing process-induced deformations (PIDs) in 4.3. Summary of automation models for composite
composite materials. This method provides a reliable materials
148
prediction model for optimizing key process parameters,
such as layup design and curing cycles, using minimal These advancements highlight the growing role of AI in
experimental data. By utilizing TGML, the study identified automating and optimizing the composite manufacturing
an optimal layup configuration and curing cycle, achieving process. To date, only a limited number of studies have
significant reductions in PIDs and production costs. The applied autonomous AI in composite materials research.
study demonstrates how ML can be applied to optimize With continuous improvements in process monitoring,
complex manufacturing processes without the need for modeling, and optimization, ML/DL-driven approaches
extensive experimental trials. are paving the way for more efficient, reliable, and cost-
effective production of composite materials. Table 3
In the domain of composite curing, deep reinforcement provides a summary of the applications of automation
learning has been employed to optimize temperature models in the field of composite materials, with the
profiles and tooling during the curing process. 76,127,130 advantages and limitations.
Würth et al. demonstrated the application of PINNs
91
for cost optimization of the thermochemical curing 5. Conclusion and future perspectives
process of a composite plate. The use of PINNs resulted In this paper, the innovative potential of AI in the design,
in speed improvements by over 500 times compared with manufacturing, and analysis of composite materials is
conventional finite element simulations, demonstrating
their potential to significantly reduce simulation times investigated through a systematic study of predictive,
and enhance process efficiency. Similarly, Humfeld et al. generative, and automation models.
149
optimized the air temperature profile for composite curing Predictive models have shown exceptional performance
subject to process constraints, such as cure duration in predicting the physical properties of composite
and maximum part and air temperatures, using a PINN materials, microstructure analysis, and design parameter
framework. Their multiple neural network framework estimation, utilizing techniques including DNNs, CNNs,
consisted of four neural networks, each representing a transfer learning, and PINNs. Generative models play
different process variable, and a series of transfer learning a crucial role in the design and optimization of new
stages with increasing complexity to ensure training materials, microstructure design, and the discovery of
convergence. Through reinforcement learning-based novel materials, offering innovative solutions in the field
process control, Szarski and Chauhan improved the air of composite materials based on various techniques,
146

Table 3. Applications, advantages, and limitations of automation models for composite materials
Applications Advantages Limitations
Void detection in laminates using CNN, defect detection in AFP - High defect detection accuracy - Difficulty in detecting subtle defects
143
using CNN, delamination characterization using transformer, - Automation of the detection process - Challenges in data acquisition
141
144
RTM mold filling prediction using PixelRNN, and defect detection - Improved production quality and - Requirement for integration with
142
using GAN/diffusion/NF 126,131,146 reduced time/cost manufacturing processes
AFP process monitoring using ML and thermal imaging, curing - Optimization of manufacturing - Complexity of system integration
147
148
process optimization with TGML, curing profile optimization with efficiency and precision - Sensitivity to external environment
reinforcement learning, and composite curing optimization using - Reduction of human error variations
146
multi-stage PINN 149 - Lower production costs
Abbreviations: AFP: Automated fiber placement; CNN: Convolutional neural network; GAN: Generative adversarial network; ML: Machine learning;
NF: Normalizing flow; PINN: Physics-informed neural network; RNN: Recurrent neural network; RTM: Resin transfer modeling;
TGML: Theory-guided probabilistic machine learning.
Volume 2 Issue 3 (2025) 22 doi: 10.36922/IJAMD025210016