Special Issue on Generative Artificial Intelligence for Design, Manufacturing Processes, and Materials Systems: Part I

Abstract

Generative artificial intelligence (AI) refers to the domain of AI systems designed to generate new information and artifacts by sampling from complex distributions captured from the data they were trained on. Encompassing techniques such as generative adversarial networks, variational autoencoders, diffusion models, large language models (LLMs), and vision-language models (VLMs), generative AI is fundamentally revolutionizing the engineering domain. Moving beyond traditional descriptive and predictive modeling, these technologies have demonstrated an exceptional capacity to enable the end-to-end creation of design solutions, optimize high-dimensional complex systems, and synthesize deep, multimodal insights into engineering problems. By leveraging multimodal data, such as textual, visual, and physical signals, generative AI can be used to address critical tasks ranging from conceptual design ideation and specification generation to digital prototyping, predictive maintenance, process optimization, and simulation analysis. Leveraging these advanced capabilities allows engineers to push the boundaries of what is possible in modern engineering design and manufacturing fields. This special issue consolidates cutting-edge research on the applications of generative AI in a broad range of engineering contexts.The rapid adoption of generative AI in engineering also raises critical research questions. How can we ensure the reliability and trustworthiness of AI-generated designs? How can we effectively fuse multimodal data in complex manufacturing environments? And how do we adapt these models to specialized engineering domains where data is often scarce, proprietary, or highly technical? This special issue aims to address these challenges, gathering contributions that explore the integration and impact of generative AI across design, manufacturing processes, and material systems. The team of guest editors issued a call for papers focusing on topics including the application of LLMs/VLMs in engineering, generative modeling for design and topology optimization, multimodal data fusion in manufacturing, generative AI for temporal data, and the augmentation of generative models through explainable AI frameworks.The first part of this special issue comprises 10 papers that fall into three primary thematic groups: (1) large language models in engineering design and knowledge retrieval, which covers how LLMs facilitate the reuse of knowledge and design components, complex document comprehension, and material selection; (2) generative and surrogate modeling for design and materials systems, focusing on how generative frameworks aid in design optimization under scarce data and enable multimodal data fusion; and (3) AI and generative models for smart manufacturing and operations, examining the impact of generative decision models on temporal data, human behavior modeling, and robotic task planning. Collectively, these contributions offer critical insights into both the immense potential and the current limitations of generative AI in transforming design and manufacturing research and practice. In the following, we provide brief summaries of the papers in this first part of the special issue, grouped by the aforementioned themes.The first set of papers explores how large language models can be utilized to automate and enhance complex engineering design processes, knowledge extraction, and critical decision-making tasks such as material selection.Wenqiang Yuan, Kaidi Wang, Jianfeng Lu, et al., in the paper titled “A MBSE and LLM Driven Method for Intelligent Generation of Aerial Bomb Design Solution,” tackle the persistent challenges of knowledge reuse and design agility in model-based systems engineering (MBSE). The authors propose an intelligent generation framework that synergistically combines MBSE methodologies with LLMs. By constructing hierarchical knowledge structures to emulate expert cognitive processes, the framework bridges abstract tactical requirements with highly technical, component-specific summaries. Experimental results demonstrate that their proposed method significantly outperforms generic chain-of-thought models and GraphRAG approaches in terms of requirement satisfaction and solution validity, thereby enhancing the agility of the design process.In a related study focused on design reuse, Ekrem Bilgehan Uyar, Cemil Gokce, Ali Ergin Gursoy, and Tugba Taskaya Temizel present “Design Reuse via Automated Requirements-Driven Retrieval: A Framework for Electronic Hardware.” To overcome the hurdles of retrieving electronic hardware components in confidentiality-driven and low-resource environments, the paper introduces a structured framework that combines expert-guided metadata curation, retrieval methods, and LLMs to support hardware design reuse. It also contributes R3SET, a dataset linking 338 real-world hardware requirements to 92 reusable circuit blocks from 8 open-source projects, supplemented with synthetic distractor blocks to reflect realistic retrieval conditions. Overall, the work demonstrates the promise of requirements-driven retrieval as a practical foundation for more efficient and explainable electronic hardware reuse workflows.Engineering rule books and technical requirements are highly specialized, and general-purpose LLMs often fail to reason over them accurately without extensive fine-tuning. To address this need for specialized knowledge extraction, Haoyang Xie and Feng Ju propose a principled agent-based design paradigm for interpreting complex engineering documents in their paper titled “Agent-Based Framework for Engineering Document Comprehension with Large Language Models.” The authors present an agentic framework that allows engineers to rapidly retrieve and reason over this specialized technical information under data-scarce conditions, negating the need for continuous model fine-tuning and providing a reliable retrieval-augmented generation strategy tailored to specific engineering tasks.Finally, Megan Y. Ying, Daniele Grandi, Allin Groom, and Christopher McComb investigate a fundamental design challenge in “Aligning Agents with Experts in Material Selection: Prompting, Size, and Reasoning.” Material selection is a complex, open-ended task requiring designers to balance diverse stakeholder demands, costs, and performance trade-offs. Recognizing discrepancies between standalone LLM outputs and human expert recommendations, the researchers compared traditional standalone LLMs against agentic AI frameworks equipped with a reasoning process and external information retrieval tools. Through an exploration of various prompting strategies and model sizes, the paper highlights the capabilities of agentic systems in mimicking expert decision-making and provides guidance on the most effective approaches for deploying LLMs in material selection workflows.This group of papers covers research on how advanced generative modeling approaches, including diffusion models and generator networks, can be leveraged for high-dimensional design optimization and complex data fusion.Bingran Wang, Seongha Jeong, Sebastiaan P. C. van Schie, et al., in the paper titled “Knowledge-Guided Generative Surrogate Modeling for High-Dimensional Design Optimization under Scarce Data,” introduce RBF-Gen, a surrogate modeling framework for design optimization in settings where data are limited but domain knowledge is available. The method combines an overcomplete radial basis function representation with a generative modeling strategy, using expert-informed loss terms to guide the surrogate toward physically meaningful solutions while preserving consistency with scarce training data. Through structural optimization examples and a semiconductor manufacturing case study, the paper shows that integrating engineering knowledge in this way can improve surrogate accuracy and optimization performance in data-scarce regimes, while also highlighting the promise of knowledge-guided surrogate modeling for complex engineering design problems.Suk Ki Lee, Fatemeh Elhambakhsh, and Hyunwoong Ko present “Diffusion Modeling-Based Generative Multimodal Data Fusion for Aerosol Jet Electronics Printing.” Aerosol jet printing (AJP) features inherent process uncertainties and complex spatiotemporal dynamics that cannot be fully captured by single-modality sensors. To overcome this, the authors propose a diffusion-based generative data fusion framework that integrates data from multiple sensing modalities. This method learns the joint distribution of multimodal AJP sensing data to synthesize fused 2D representations, which are then used to construct comprehensive 3D surface representations of printed features. This work creates a data-driven foundation for further construction of accurate digital twins in additive manufacturing.The final set of papers focuses on applying AI, LLMs, and generative decision models to improve manufacturing line operations, conduct temporal data analysis, and optimize human-machine collaboration in factory settings.In the paper titled “Task and Motion Planning with Large Language Models Enhanced by Spatially-Temporally Aware Tools for Smart Manufacturing,” Shuo Liu and Youyi Bi address a significant challenge in the deployment of LLMs in smart manufacturing regarding their inherent limitations in spatial and temporal reasoning, which are crucial for coordinating robotic actions. While LLMs excel at comprehension, coordinating robots for task and motion planning requires precise physical grounding. The authors develop a tool-augmented LLM framework tailored for multitype and interleaved manipulation tasks, such as material loading and unloading with robotic manipulators. By providing the LLM with spatially and temporally aware tools, the framework substantially improves the LLM's success rate and generalizability, particularly in complex, long-horizon scheduling scenarios. This work not only provides a practical solution for robust robotic control but also establishes a powerful paradigm for grounding generative AI models in the physical reality of manufacturing.Chen-Wei Guo, Omar Ashour, Christian López, and Conrad Tucker, in the paper titled “Estimation of Control Intervals for Reinforcement Learning-Based Manufacturing-Line Control,” explore the use of generative decision models for optimizing coordinated operations. Efficient control of manufacturing lines requires complex routing, worker allocation, and scheduling decisions that traditional model-based approaches struggle to scale. The authors investigate the application of reinforcement learning (RL) as an alternative, focusing deeply on the crucial design choice of the policy control interval. To move beyond heuristic or arbitrary selection, they introduce a novel, layout-aware estimator derived from the critical path method and a probability mass function over transit times. By aligning the decision frequency with the intrinsic timescale of material flow through the line, their method provides a reproducible, system-informed choice for the control interval. Experiments on a complex manufacturing line simulation reveal a nonmonotonic relationship between interval length and policy performance, with the estimated interval consistently achieving an interior optimum. This work offers a simple yet powerful plug-and-play procedure for improving the robustness and efficiency of RL controllers, highlighting the crucial role of temporal scale in deploying generative AI for complex operations management.Melinda T. Mudzurandende, Katherine A. Flanigan, and Christopher McComb present “Characterizing Sequential Patterns of Human Behavior in Advanced Manufacturing.” To design adaptive, human-centered manufacturing systems, it is essential to understand the sequential nature of human behavior in advanced manufacturing contexts. The authors utilize a large-scale dataset of annotated human activity as workers engaged with a wire arc additive manufacturing machine. By applying generative sequential models, including hidden Markov models and Markov chains, they systematically analyze how behavioral predictability and structure vary across different sampling frequencies. Their findings reveal that human workflows are temporally persistent and can be accurately described by a remarkably small number of latent modes, providing a pathway for better human-machine collaboration.Finally, in the paper titled “Hybrid Temporal Modeling and Generative Augmentation for Imbalanced Multivariate Time Series Anomaly Detection in Engineering Systems,” Ahmed Shoyeb Raihan, Farzana Islam, Zhichao Liu, Srinjoy Das, and Imtiaz Ahmed address the critical challenge of fault detection in engineering systems. Multivariate time series data often feature anomalies that are rare, severely imbalanced, and embedded in long contexts. The authors propose a hybrid deep learning architecture named DA-GAT-Net, which combines an anomaly detector, GAT-Net, with a generative augmentation strategy using CTGAN. This framework successfully captures long-term dependencies, identifies critical time-steps, and mitigates the severe class imbalance, resulting in highly robust anomaly detection capabilities.This group of 10 papers forms the first part of this special issue. Together, they have demonstrated the immense promise of generative AI methodologies to advance design exploration, facilitate knowledge reuse, optimize complex manufacturing processes, aid engineering decision-making, and forge a deeper understanding of human behavior. While challenges remain in data acquisition, domain adaptation, and physical grounding, the rigorous frameworks, benchmarks, and models presented in these articles underscore the rapid development and maturation of the field. The upcoming second part of the special issue will include additional papers. Their findings and insights will be analyzed and synthesized in our next editorial article accompanying them.As generative AI continues to evolve, its successful integration into complex engineering environments will require ongoing innovation and cross-disciplinary collaboration. We hope this special issue serves as a robust foundation for further exploration and inspires future developments in AI-driven engineering applications and innovation.