Page 25 - IJAMD-2-3

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

3.4. Summary of generative models for composite also actively control, execute, and optimize manufacturing
materials processes. Unlike predictive or generative models that
The applications of generative models in composite primarily focus on producing outputs from given inputs or
materials research are summarized in Table 2, along with generating new data, automation models are characterized
the advantages and limitations of each type of ML method. by their ability to autonomously manage dynamic
workflows through continuous feedback, control strategies,
Although the use of generative models is appealing and real-time optimization. This includes decision-making
for discovering new designs and materials, limitations tasks, such as defect detection, process monitoring, and
include the difficulty in manufacturing the optimized quality control, as well as direct process control, adaptive
microstructures, considering the extremely stochastic manufacturing scheduling, robotic motion planning,
nature of composite materials. The design process should and real-time process optimization. These models are
be constrained by the manufacturability of the generated particularly used in automated manufacturing processes,
composite structure, and additive manufacturing and robotic control systems, and autonomous driving, playing
automated technologies could be used in combination a critical role in improving both efficiency and quality. 138-140
with generative models to create more accurately tailored
structures. Composite materials are manufactured through
complex processes that often demand significant time and
Moreover, while transformer-based generative models, cost. Moreover, manual production can lead to variations
such as GPT-like architectures, Image GPT, and vision in material properties, making defect detection and quality
transformer-based diffusion generators, 134-136 have not control crucial to maintaining high standards. In addition,
yet been widely applied in composite materials research, the ability to automatically adjust process parameters in
their potential in the field is promising. The ability of real-time is crucial in accommodating the sensitivity of
transformer-based generative models to capture complex composites to environmental and operational conditions.
patterns and long-range dependencies makes them suitable Therefore, AI automation – encompassing both decision
for applications such as microstructure generation and logic and process execution – is emerging as a key technology
cross-modal composite design. Future research exploring for enhancing efficiency, consistency, and quality in the
the adaptation of these foundation models could open new production of composite materials. The manufacturing
directions for AI-driven composite materials research and methods and applications utilizing automation models
further enhance the generative design workflow. covered in this review are shown in Figure 11A.

4. Automation models for composite 4.1. Machine learning/deep learning-driven quality
materials control

Automation models refer to AI technologies that not only Defect detection and quality control are essential to
analyze data in real time and make automated decisions but maintain consistent quality and prevent significant

Table 2. Applications, advantages, and limitations of generative models for composite materials

Machine Applications Advantages Limitations
learning method
Variational - Composite layup generation and optimization 117 - Generative and - Low quality of generated data, especially for
autoencoders -Shape memory polymer property optimization 115 inference models high-resolution data or complex patterns
-Metamaterial design for impact mitigation 116
Generative -Microstructure generation for fiber composites 123 - High resolution - Potential occurrence of mode collapse,
adversarial - Engineered cementitious composites: multi-objective results resulting in repetitive pattern generations
networks material optimization 124 - Effectiveness
-Topology optimization for periodic structures 125 for complex
-Defect detection in composites 126 multi-modal
distributions
Diffusion models -Microstructure generation 128 - High quality and - High Computational demands and need for
-Data amplification 131 diversity of generated large training datasets
samples
Normalizing flows Microstructure generation with targeted property 133,137 - Likelihood - Complex training processes, sensitivity to
-
estimation for a model architecture, and scalability challenges
given sample

Volume 2 Issue 3 (2025) 19 doi: 10.36922/IJAMD025210016