Zhenghui Sha

Abstract This guest editorial introduces Part I of the Special Issue on Generative AI for Design, Manufacturing Processes, and Materials Systems. By moving beyond traditional predictive modeling, generative AI enables end-to-end design creation, complex system optimization, and the synthesis of multimodal insights. Despite its promise, the rapid adoption of generative AI introduces critical challenges regarding design reliability, data fusion, and model adaptation in specialized, data-scarce engineering environments. To address these challenges, this issue compiles ten cutting-edge research papers organized into three central themes. The first theme focuses on leveraging LLMs for engineering design and knowledge retrieval, highlighting innovations in material selection, complex document comprehension, and electronic hardware reuse. The second theme explores generative and surrogate modeling, demonstrating how these frameworks aid in high-dimensional design optimization and multimodal data fusion in manufacturing. The final theme examines generative models in smart manufacturing, emphasizing robotic task planning, human-machine collaboration, and temporal data anomaly detection. Collectively, these contributions establish a robust foundation for future AI-driven engineering innovations.

Large language models (LLMs) have recently enabled automatic generation of parametric computer-aided design (CAD) programs from natural language. However, real-world CAD workflows are inherently iterative and require reliable editing rather than one-shot model synthesis. In this work, we propose LLM4CAD-Editor, an LLM-based intent-aware framework for instruction-guided CAD editing based on a structured domain-specific language (LLM4CAD-DSL). The symbolic representation of LLM4CAD-DSL enables robust geometric modification through a feature-level entity selection mechanism, allowing models to reference geometry via feature names instead of coordinates, thus transforming fragile coordinate-based reasoning into natural language-based reasoning that many LLMs can handle. We construct a multimodal CAD editing dataset with over 35,139 instruction-program pairs via DSL-based augmentation and vision-language instruction synthesis, covering functional-, operation-, and parameter-level editing intents. To validate the work, we fine-tuned a 32B-parameter language model for DSL editing generation. Experimental results show high parsing accuracy for parameter-level edits (96.3%) and strong intent satisfaction rates of 82% for functional instructions. The model also achieves an average Intersection-over-Union (IoU) of 0.935 for parameter-level edits, 0.871 for operation-level edits, and 0.708 for functional-level edits, while the corresponding average editing distances are 0.176, 0.579, and 2.859, respectively. Comparative studies further demonstrate a significant improvement in editing robustness by 1.4x over Python-based CAD scripting approaches. These results confirm that LLM4CAD-Editor can reliably perform both low-level parameter modifications and high-level functional edits, maintaining high accuracy and low structural errors across diverse editing tasks.

Large language models (LLMs) have recently enabled automatic generation of parametric computer-aided design (CAD) programs from natural language. However, real-world CAD workflows are inherently iterative and require reliable editing rather than one-shot model synthesis. In this work, we propose LLM4CAD-Editor, an LLM-based intent-aware framework for instruction-guided CAD editing based on a structured domain-specific language (LLM4CAD-DSL). The symbolic representation of LLM4CAD-DSL enables robust geometric modification through a feature-level entity selection mechanism, allowing models to reference geometry via feature names instead of coordinates, thus transforming fragile coordinate-based reasoning into natural language-based reasoning that many LLMs can handle. We construct a multimodal CAD editing dataset with over 35,139 instruction-program pairs via DSL-based augmentation and vision-language instruction synthesis, covering functional-, operation-, and parameter-level editing intents. To validate the work, we fine-tuned a 32B-parameter language model for DSL editing generation. Experimental results show high parsing accuracy for parameter-level edits (96.3%) and strong intent satisfaction rates of 82% for functional instructions. The model also achieves an average Intersection-over-Union (IoU) of 0.935 for parameter-level edits, 0.871 for operation-level edits, and 0.708 for functional-level edits, while the corresponding average editing distances are 0.176, 0.579, and 2.859, respectively. Comparative studies further demonstrate a significant improvement in editing robustness by 1.4x over Python-based CAD scripting approaches. These results confirm that LLM4CAD-Editor can reliably perform both low-level parameter modifications and high-level functional edits, maintaining high accuracy and low structural errors across diverse editing tasks.

Abstract Manufacturing processes are inherently dynamic and uncertain, with varying parameters and nonlinear behaviors, making robust control essential for maintaining quality and reliability. Traditional control methods often fail under these conditions due to their reactive nature. Model predictive control (MPC) has emerged as a more advanced framework, leveraging process models to predict future states and optimize control actions. However, MPC relies on simplified models that often fail to capture complex dynamics, and it struggles with accurate state estimation and handling the propagation of uncertainty in manufacturing environments. Machine learning (ML) has been introduced to enhance MPC by modeling nonlinear dynamics and learning latent representations that support predictive modeling, state estimation, and optimization. Yet, existing ML-driven MPC approaches remain deterministic and correlation-focused, motivating the exploration of generative ML. Generative ML offers new opportunities by learning data distributions, capturing hidden patterns, and inherently managing uncertainty, thereby complementing MPC. This review highlights five representative methods and examines how each has been integrated into MPC components, including predictive modeling, state estimation, and optimization. By synthesizing these cases, we outline the common ways generative ML can systematically enhance MPC and provide a framework for understanding its potential in diverse manufacturing processes. We identify key research gaps, propose future directions, and use a representative case to illustrate how generative ML-driven MPC can extend broadly across manufacturing. Taken together, this review positions generative ML not as an incremental add-on but as a transformative approach to reshape predictive control for next-generation manufacturing systems.

Abstract Computer-aided design (CAD) tools empower designers to design and modify 3D models through a series of CAD operations, commonly referred to as a CAD sequence. In scenarios where digital CAD files are inaccessible, reverse engineering (RE) has been used to reconstruct 3D CAD models. Recent advances have seen the rise of data-driven approaches for RE, with a primary focus on converting 3D data, such as point clouds, into 3D models in boundary representation (B-rep) format. However, obtaining 3D data poses significant challenges, and B-rep models do not reveal knowledge about the 3D modeling process of designs. To this end, our research introduces a novel data-driven approach based on representation learning to infer CAD sequences from product images, coined as Image2CADSeq. These sequences can then be translated into B-rep models using a solid modeling kernel. Unlike B-rep models, CAD sequences offer enhanced flexibility to modify individual steps of model creation, providing a deeper understanding of the construction process of CAD models. One unique contribution of this study is the development of a multilevel evaluation framework for model assessment, so the predictive performance of the Image2CADSeq model can be rigorously evaluated. The model was trained on datasets generated using a proposed data synthesis pipeline, and two different neural network architectures were explored to optimize the Image2CADSeq performance. The experimental results show that our methods are promising in data-driven reverse engineering of 3D CAD models (CAD sequences) from 2D images.

Cooperative 3D printing (C3DP) is an additive manufacturing paradigm where a swarm of robotic 3D printers work cooperatively in a shared environment to fabricate continuous parts. Reliable operation requires both accurate per-printer kinematic calibration and cross-printer spatial alignment. This paper presents an automatic vision-based XY calibration workflow for C3DP using ArUco fiducials and low-cost monocular cameras. The method performs intra-printer kinematic calibration and inter-printer alignment through peer-to-peer observations without fixed global infrastructure. In a two-printer Selective Compliance Assembly Robot Arm (SCARA) Fused Filament Fabrication (FFF) testbed, the automatic workflow reduced total calibration time from 157.19 min (manual) to 36.49 min while improving positional consistency and print accuracy. For individual-printer artifacts, the mean Euclidean error was 0.03 ± 0.02 mm, whereas cooperative artifacts exhibited a mean Euclidean error of 0.078 ± 0.002 mm. These results show that practical and repeatable C3DP calibration can be achieved with low-cost vision hardware.

Modern artificial intelligence (AI)-assisted design methods, such as generative design (GD), have expanded concept generation and evaluation far beyond traditional approaches. This chapter presents a comprehensive product concept generation and evaluation framework that integrates HFE and mechanical performance across both the divergent and convergent stages of conceptual design. It aims to eliminate infeasible concepts that fail to meet human-centered design requirements and identify preferred alternatives using AI algorithms early in the process. One of the major challenges of applying data-driven GD approaches to the early design stages is the pursuit of creativity. The chapter describes the proposed methodology for iteratively refining design concepts to satisfy ergonomic requirements alongside users' preferences. It employs an occupant packaging evaluation template that uses 3D digital manikin models to assess geometric factors affecting driver ergonomics, such as the impact of roof height on headroom.

Abstract Modern factories, despite advancements in automation and digitalization, are still limited in adapting to different products due to their design for specific purposes and dependence on specialized supply chains. This rigidity limits their adaptability and resilience to disruptions. This paper introduces a comprehensive framework for Heterogeneous Swarm Manufacturing (HSM) that enables a swarm of different manufacturing robots to dynamically reconfigure and cooperatively execute hybrid manufacturing tasks. The framework encompasses processes including task division, task allocation, dynamic scheduling, and path planning, addressing the complexities of interdependence between these processes. To validate the framework, we developed a testbed featuring four types of robot-3D printing robots, laser processing robots, transport robots, and assembly robots—operating on a modular floor tile system and conducted two proof-of-concept case studies by manufacturing a large-scale Razorback Logo and a fully functional mini-electric vehicle (Mini-EV). The results demonstrated the effectiveness of the framework in guiding the planning of heterogeneous swarm manufacturing. The results show the promise of HSM to improve the adaptability and resilience of manufacturing systems, paving the way for more agile, general-purpose factories.

Abstract This editorial is the first in a series of editorials exploring the future of engineering design from the perspective of thought-leaders within the engineering design community. Five early- and mid-career researchers were invited to present their perspectives at a meeting of the Design Society in March 2025 on the campus of the Georgia Institute of Technology. Jessica Menold focused on teaming and collaboration in the context of artificial intelligence (AI); Kosa Goucher-Lambert on design cognition; Astrid Layton on sustainable and resilient design; Mohsen Moghaddam on virtual reality (VR)/augmented reality (AR)/extended reality (XR) in engineering design; and Zhenghui Sha on the design of complex sociotechnical systems. Each presentation was followed by roundtable discussions of the challenges and opportunities posed by the speaker. The presentations and audience discussions are summarized in this editorial, along with a brief overview of some of the opportunities for further work. The goal of this editorial series is to inform the broader community of the progress, challenges, and opportunities associated with important themes within our research community and to offer a starting point for those who seek to investigate these topics and continue to advance the state-of-the-art in our field.

not-yet-known not-yet-known not-yet-known unknown Civil infrastructure is a system through which humans inter- act with each other and their environment, making it essen- tial for community well-being. Such systems, however, de- grade over time and become less usable, either due to pro- gressive wear or abrupt damage. Therefore, it becomes nec- essary to allocate resources to maintain the infrastructure and thereby ensure the community’s well-being. However, the relationship between infrastructure condition and com- munity well-being is unclear and challenging to quantify. This is especially true in locations that are impacted by climate change or other significant forcing factors. Current compu- tational models often struggle to predict cascading nonlin- ear events that occur during system failures. This paper in- troduces a computational method to quantitatively analyze the impact of infrastructure on community well-being, com- bining agent-based modeling (ABM) and network robustness measurement. We specifically examine the town of Utqiaġvik, Alaska, where permafrost thaw due to global warming threat- ens a substantial amount of infrastructure. The results show that the breakdown of critical infrastructure progressively weak- ens community access to essential resources, leading to much lower robustness. Additionally, our simulations indicate that smaller household sizes and redundant infrastructure design prove beneficial for sustaining resource accessibility and fos- tering close social connections within the community. These insights offer valuable guidance for understanding the com- plex systems interplay among communities, infrastructure, and the environment, thereby informing strategies to build more sustainable and resilient systems in remote areas such as Utqiaġvik.

Abstract Computer-aided design (CAD) tools empower designers to design and modify 3D models through a series of CAD operations, commonly referred to as a CAD sequence. In scenarios where digital CAD files are inaccessible, reverse engineering (RE) has been used to reconstruct 3D CAD models. Recent advances have seen the rise of data-driven approaches for RE, with a primary focus on converting 3D data, such as point clouds, into 3D models in boundary representation (B-rep) format. However, obtaining 3D data poses significant challenges, and B-rep models do not reveal knowledge about the 3D modeling process of designs. To this end, our research introduces a novel data-driven approach based on representation learning to infer CAD sequences from product images, coined as Image2CADSeq. These sequences can then be translated into B-rep models using a solid modeling kernel. Unlike B-rep models, CAD sequences offer enhanced flexibility to modify individual steps of model creation, providing a deeper understanding of the construction process of CAD models. One unique contribution of this paper is the development of a multi-level evaluation framework for model assessment, so the predictive performance of the Image2CADSeq model can be rigorously evaluated. The model was trained on a specially synthesized dataset, and various neural network architectures were explored to optimize the performance. The experimental and validation results show the great potential of our model in data-driven reverse engineering of 3D CAD models from 2D images.

Abstract This paper presents methods used to design and develop a smart hand tool system that takes advantage of low-cost sensing, machine learning, and real-time monitoring to optimize tool usage and improve human-tool interaction. A multidisciplinary team took a user-focused approach, balancing engineering design, prototyping, and testing with qualitative research and quantitative analysis to derive user needs and requirements. A prototype sensor unit (PSU) that can be adapted to various tools was developed to enable the real-time acquisition of data on motion, power consumption, orientation, and user activity. The prototype was experimentally validated on multiple tool types, and ML-enabled features including skill assessment, task recognition, battery life prediction, load estimation, and anomaly detection were developed and tested. Skill assessment ranking of user proficiency based on a Skill Index Score (SIS) correlated well with GD&T-based evaluations. Task recognition algorithms achieved over 77% accuracy, while battery life prediction closely matched real usage data. Load estimation was found to provide force predictions with an average error of ±1.18 N, and anomaly detection identified deviations such as excessive force and tool stoppages. These features were processed online using PSU and edge computing features. The results demonstrate the feasibility of further developing AI-enhanced power tools with real-time monitoring and performance evaluation, paving the way for advances in human-tool collaboration, skill development, training, and next-generation smart manufacturing applications.

Abstract The optimal allocation of electric vehicle (EV) charging resources is crucial for advancing national electrification and de-carbonization plans, prompting extensive research into efficient placement strategies for EV charging infrastructure. While many aim to maximize the coverage of charging resources based on demand, typical approaches adopt uniform grid cells to divide the region of interest for analysis. However, these methods often overlook spillover effects (i.e., dependencies) between subregions, where high charging demand in one sub-region spreads to surrounding areas. To address this limitation, we develop a novel bipartite network-based design decision-making framework for optimal placement and allocation of EV charging stations, including the number and type of chargers. The proposed framework introduces two key innovations. First, it includes a new partition method integrating Voronoi diagrams with K-means clustering to mitigate the spillover effects inherent in grid-based methods. This method aggregates charging demand by accounting for points of interest (POIs) and traffic flow through K-means clustering and then partitions the region of interest into Voronoi cells based on the clustered centroids. Second, the framework adopts the choice modeling philosophy and uses a bipartite network model to represent “customer” nodes (i.e., EV drivers in a service zone defined by a Voronoi cell) and “product” nodes (i.e., charging stations). A link is established between a station and its corresponding service zone if it lies within the zone. For service zones without a station, links are created to the nearest station based on the shortest driving distance calculated from the real-world transportation network, incorporating driving costs. With such a choice modeling method, the charging demand can be explicitly represented to support optimal resource allocation, including the location of stations and the number and type of chargers. To demonstrate and validate the proposed framework, we formulate an optimization problem to maximize the coverage of public EV charging resources in Austin, Texas, while minimizing the driving cost and total expenses, subject to budget and power grid constraints.

Abstract Engineering design optimization of complex systems often involves varying operating conditions, e.g., the same design of a cold-spray nozzle under different manufacturing configurations with varying pressures and stand-off distance. This requires robust design methods to simultaneously ensure high performance and the stability of such a high performance. This paper presents a constrained Bayesian Optimization (BO) framework for robust design, aimed at optimizing performance and stability across varying operation conditions or configurations. The proposed methodology leverages condition-specific Gaussian Process (GP) submodels to estimate the nonlinear relationships between design parameters and performance using a newly developed robust Expected Improvement (EI) acquisition function. This function aggregates GP submodels from different configurations accounting for the worst-case scenario, a strategy that selects the minimum posterior mean and maximum posterior variance across the submodels to prioritize the most challenging scenario from all operating conditions. In addition, a penalty term based on the variance of the posterior mean is added to the new EI acquisition function to improve the consistency of the design performance. To demonstrate its effectiveness, we applied the framework to the design optimization of cold spray nozzles, where the objective is to find dimension variables of the nozzle that maximize particle impact velocity while ensuring a stable performance under four operating conditions. The results demonstrate that the proposed approach achieves high performance, with a 13% improvement over the original design. Furthermore, the performance variance is reduced from 204.97 (m/s)2 to 43.06 (m/s)2, significantly improving the design consistency. These results highlight its potential for broader applications in robust engineering design optimization.

Abstract The creation of manufacturable and modifiable 3D shapes using Computer-Aided Design (CAD) remains a predominantly manual and time-consuming process, hindered by the complexity of boundary representations in 3D solids and the lack of intuitive design tools. This paper introduces TransformCAD, a CAD generation model that leverages both image and natural language descriptions as input to generate CAD sequences, producing editable 3D representations relevant to engineering design. TransformCAD incorporates a fine-tuned Contrastive Language-Image Pre-Training (CLIP) model to process multimodal input and employs two prediction branches—sketch and extrude—to enhance the parsing rate of CAD generation. Extensive evaluations demonstrate that TransformCAD outperforms existing models in terms of parsing rate, Chamfer distance, minimum matching distance, and Jensen-Shannon divergence. Furthermore, by analyzing the impact of training data, we show that TransformCAD exhibits strong potential for accurately generating long-sequence CAD models, which correspond to higher-complexity designs. Moreover, real-world 3D object images taken by a smartphone are used to validate TransformCAD’s practicability, demonstrating its effectiveness in industrial applications. To the best of our knowledge, this is the first attempt at generating 3D CAD models integrating both image and natural language input. TransformCAD expands the boundaries of automated CAD modeling, enabling a more flexible and intuitive design process that bridges visual perception and structured command-based representations.

Abstract For team-based decisions under uncertainty, collaboration plays a crucial role in improving decision quality where information sharing between teammates helps reducing individual risk. Understanding how teams adapt and coordinate under uncertainty is essential for designing better decision-support systems in engineering and organizational contexts. This study investigates how information sharing within a team affects human decisions under uncertainty. We present the results of an online team-based experiment where 20 participants with engineering backgrounds engaged in a black-box 1-D function optimization game under two distinct conditions: no collaboration (no information shared between team members) and limited collaboration (with partial information shared between team members). Each game included 8 rounds in which participants strategically sampled input values to maximize an unknown function with limited budget while competing against another team. We analyze behavioral changes in response time, accuracy, effort, and overall performance. Results show that participants in the collaborative condition achieved lower solution quality, but achieving higher average payoffs with fewer attempts. Response time also increased significantly in the collaborative condition, indicating slower but more thoughtful decision-making in collaborative teams. These findings suggest that collaboration led to more cautious, efficient strategies and resource conservation. Our experiment highlights the inherent trade-offs between speed, precision and efficiency in team settings and offers insights for designing collaborative decision systems in uncertain environments.

Abstract In this study, we pose the following question: How can we leverage the geometric information of a part to guide the partition of a print job for cooperative 3D printing (C3DP)? To enable a print job to be cooperatively fabricated by multiple robots, traditional partitioning in C3DP has historically been robot-centric, meaning that the print job is divided according to the robot resources (e.g., the number of robots) and their physical constraints (e.g., arm reachability and kinematics). However, often the user of the C3DP system does not know how to leverage the existing resources, as assigning too many robots can lead to diminishing, or even negative returns in terms of certain performance metrics (e.g., makespan). Thus, we are interested in the inverse problem, where the partition is geometry-centric and inherently suggests what the robot resources should be, given a printing layer. Our hypothesis is that there exists valuable information embedded in the geometric and topological structure of the printing layer that can provide a natural way to partition it, and therefore suggest the number of robots needed for the process, including their location in the workspace. This work elicits a significant paradigm shift from robot-centric to geometry-centric C3DP partitioning. We demonstrate that, when considering the medial axis transform (MAT) of the printing layer boundary, we can take advantage of its corresponding radius function to find the most important subset of the medial axis that induces a natural domain decomposition. We achieve this by iteratively constructing a set of vertices in the branches of the medial axis that faithfully capture the important local geometric features of the printing part. The degree of the selected vertices is then used to infer the most natural number of robots to print that immediate vicinity. Finally, given the suggested number of robots, Voronoi sites are optimally sampled on the boundary of the layer, and the final partition is created from the resulting Voronoi tessellation. The proposed framework is layer-wise, as the cross-section of a part may vary significantly both geometrically and topologically along its height. We show a range of numerical results, demonstrating that our methodological framework offers a robust and intuitive way to decompose the geometric domain of each layer for C3DP. Finally, the proposed medial axis-based methodology is process-agnostic and can be used in different multi-robot coverage problems, including with other additive manufacturing technologies.

Abstract Cooperative 3D printing (C3DP) is an emerging field in Swarm Manufacturing (SM) that allows larger format objects to be printed in parallel for improved efficiency without compromising print quality. Our previous work with in-situ monitoring of C3DP includes detecting stringing defects and 2D image matching in fused filament fabrication 3D printing. However, the lack of true in-situ analysis impeded proper closed-loop control. In this study, we improve the in-situ monitoring framework by utilizing the Canny edge detection used for 2D image matching in our previous work to detect warping, layer splitting, and interstitial gaps. After testing four different methods, including bounding edge comparison, closest point comparison, corner point comparison, and K-nearest neighbor regression comparison, we were able to accurately measure warping effects using corner points of the detected contours. This corner point comparison method can effectively determine the severity of warping, allowing us to have closed-loop control of a heated bed that can successfully decrease warping effects. The proposed methods can also detect interstitial gaps on a y-axis basis, and continuously collect data to communicate G-code commands to the printers for recalibration and stopping fatal prints. The computer vision-based monitoring framework presented in this paper laid a good foundation to improve the C3DP process and provided insight into the in-situ monitoring of various other forms of additive manufacturing.

We present Traveling Cellsman , an approach for creating a parameterization for task scheduling and collision avoidance with Cooperative 3D printing (C3DP). The parameterization is based on the distribution of work between robots (partition), which allows the robots to navigate through their printing tasks effectively while also allowing for collision avoidance with other robots. The parameterization provides straightforward optimization of makespan. Inspired by the multiple traveling salesman problem (MTSP), we schedule tasks by first clustering tasks together based on a parameterization of the partition. The clustered tasks can then be ordered for printing. Numerical results indicate that our clustering approach finds an optimal solution faster than the non-clustered approach for minimizing the pause and movement time of the robots. Physical results also show that optimization allows for faster printing time as compared to non-optimized or slicer-based methods for generating a printing schedule. While we demonstrate our method using C3DP, it is generally applicable to other multi-robot task scheduling problems where collision may occur.

ABSTRACT: Rational decision-making is crucial in the later stages of engineering system design to allocate resources efficiently and minimize costs. However, human rationality is bounded by cognitive biases and limitations. Understanding how humans deviate from rationality is critical for guiding designers toward better design outcomes. In this paper, we quantify designer rationality in competitive scenarios based on utility theory. Using an experiment inspired by crowd-sourced contests, we show that designers employ varied search strategies. Some participants approximate a Bayesian agent that aimed to maximize its expected utility. Those with higher rationality reduce uncertainty more effectively. Furthermore, rationality correlates with both the proximity to optimal design and design iteration costs, with winning participants exhibiting greater rationality than losing participants.

ABSTRACT: Design decision-making under competition is a critical challenge in real-world engineering design. These challenges are compounded by bounded rationality, where cognitive limitations and imperfect information influence decision-making strategies. To address these issues, we develop a game-theoretic research platform to investigate team-based design under competition. This platform abstracts and simulates real-world competitive design scenarios through controlled experiments. It features a user-friendly interface to collect behavioral data, which supports the analysis of team and individual strategies. Additionally, we validated the platform through a pilot study, demonstrating its ability to capture realistic design features and generate meaningful insights into competitive design behaviors.

ABSTRACT: Engineering design has recently undergone a paradigm shift led by generative artificial intelligence (AI). The Generative Design (GD) paradigm utilizes generative AI tools (e.g., large language models) to define the objective space and computationally exploit the design space. This is a drastic shift from the roles of human designers in the Traditional Design (TD) paradigm which consists of manual design-objective space co-evolution, and has created a research gap for Generative Design Thinking (GDT): how a designer thinks and cognitively approaches the design process during GD. To fill this gap, we propose the Paradigmatic Design Thinking Model which uniquely defines design thinking as situated within three factors (Design Cognition, Design Tools, and Design Methodology) and use it to explain design thinking in two paradigms: Traditional Design Thinking and Generative Design Thinking.

Abstract Cooperative 3D printing (C3DP) is an emerging field in Swarm Manufacturing (SM) that allows larger format objects to be printed in parallel for improved efficiency without compromising print quality. Our previous work in C3DP has presented a real-time process monitoring framework for C3DP, capable of detecting defects in fused deposition modeling (FDM) 3D printing. However, the annotation accuracy and image-matching scores in side-view error detection are too low to achieve effective closed-loop control. In this study, we improve the in-situ monitoring framework by integrating the Canny edge detector for enhanced image-matching accuracy. Also, we retrained the computer vision model based on YOLOv8 with over 7000 images utilizing image argumentation, achieving an increase of 25% compared to the 276 images used initially. With this improvement, the system has gained robustness, adaptability, and reliability across diverse camera perspectives and error conditions in real-world applications. Therefore, this research will show the improvement from the original model with a side-by-side in-situ analysis, demonstrating the increased effectiveness of real-time detection and adjustment in cooperative 3D printing. The results will highlight the potential of this enhanced system to be adapted across various applications in the additive manufacturing industry, ultimately moving toward autonomous, high-precision 3D printing systems in complex manufacturing environments.

In the ever-evolving landscape of engineering, the fusion of network science and graph theories has emerged as a dynamic force, revolutionizing the way we represent, design, model, and optimize complex systems. Networks, defined by nodes and edges, are particularly effective in modeling the interaction and interdependency among individual entities in complex systems. Networks have become the cornerstone for comprehending the intricate relationships underlying a myriad of engineering domains. From transportation networks optimizing urban mobility, power grids ensuring energy efficiency and resilience, and social networks shaping human interactions to biological networks inspiring human-engineered system design, the application of network science and graphs in engineering spans a vast spectrum of disciplines. This special issue is dedicated to promoting the dissemination of knowledge related to complex networks in engineering systems and design and highlighting the latest advances at the intersection of network science, graph theory, and engineering.

Abstract Motion planning for mobile robots in dynamic and uncertain environments (e.g., in multirobot manufacturing) is challenging due to the stochastic nature of the problem. One common approach is to construct an initial plan to guide the robots, and as information is collected during execution, adjustments are made in real time to account for the impact of uncertainties. This approach, while feasible, leaves the burden of dynamic collision avoidance on controllers, which may not find collision-free and optimal control inputs fast enough. Additionally, the computational burden is exacerbated as the dimensionality of the workspace and the number and geometric complexity of obstacles increase. This article presents a novel probabilistic roadmap (PRM)-based offline motion planner for mobile robots traveling under uncertainty. The planner considers arbitrarily shaped holonomic robots in an environment with multiple static and dynamic obstacles. Since PRM is graph-based, we model the uncertainty by treating edge costs as general probability density functions whose exact profiles are related to the actuation characteristics of a mobile robot. The risk of success (i.e., no collision) per each action in the plan is lower-bounded by a user-defined value, allowing an informed choice between solution safety and quality. Simulations in various scenarios with both static and dynamic obstacles, and configuration spaces of different dimensions, show the effectiveness and flexibility of the planner, including scenarios contemplating prioritized multirobot planning. Finally, we show that, under practical conditions, the proposed planner can provide time-optimal and globally risk-bounded solutions.

Abstract Swarm manufacturing is an emerging manufacturing paradigm that employs a heterogeneous swarm of robots to accomplish complex hybrid manufacturing tasks. Cooperative 3D printing (C3DP), a specialized form of swarm manufacturing, enables multiple printers to collaboratively produce large-scale parts, addressing key tradeoffs in additive manufacturing, such as size, speed, quality, and cost. A fundamental challenge in C3DP is ensuring collision-free, time-optimal printing in a shared workspace. This is a complex problem that can be influenced by factors such as the number of printers, part geometry, printer positioning, mobility, and kinematics. In this article, we present SafeZone*, a collision-free and scalable C3DP framework that optimizes printing time by co-considering the geometry (area and shape) and topology (space-connectivity) of a shared workspace during layer partitioning. We first establish a conceptual framework to mathematically represent the topology of a layer through partition graphs. Then, we use a Voronoi tessellation within a constrained optimization framework to control the partition graph and minimize makespan. The Voronoi sites are associated with printer locations, allowing the framework to integrate physical constraints and facilitating solutions for systems with robotic manipulators. Physical testing in a four-printer scenario with robotic arms confirms that SafeZone* enables collision-free printing, resulting in a printing time reduction of 44.63% when compared to the single-printer scenario. Finally, numerical studies reveal trends in the optimal solutions concerning the chromatic number of their resulting partition graphs and the distribution of the printing areas among printers.

Abstract Despite the power of large language models (LLMs) in various cross-modal generation tasks, their ability to generate 3D computer-aided design (CAD) models from text remains underexplored due to the scarcity of suitable datasets. Additionally, there is a lack of multimodal CAD datasets that include both reconstruction parameters and text descriptions, which are essential for the quantitative evaluation of the CAD generation capabilities of multimodal LLMs. To address these challenges, we developed a dataset of CAD models, sketches, and image data for representative mechanical components such as gears, shafts, and springs, along with natural language descriptions collected via Amazon Mechanical Turk. By using CAD programs as a bridge, we facilitate the conversion of textual output from LLMs into precise 3D CAD designs. To enhance the text-to-CAD generation capabilities of GPT models and demonstrate the utility of our dataset, we developed a pipeline to generate fine-tuning training data for GPT-3.5. We fine-tuned four GPT-3.5 models with various data sampling strategies based on the length of a CAD program. We evaluated these models using parsing rate and intersection over union (IoU) metrics, comparing their performance to that of GPT-4 without fine-tuning. The new knowledge gained from the comparative study on the four different fine-tuned models provided us with guidance on the selection of sampling strategies to build training datasets in fine-tuning practices of LLMs for text-to-CAD generation, considering the trade-off between part complexity, model performance, and cost.

Abstract We present NoodlePrint, a generalized computational framework for maximally concurrent layer-wise cooperative 3D printing (C3DP) of arbitrary part geometries with multiple robots. NoodlePrint is inspired by a recently discovered set of helically interlocked space-filling shapes called VoroNoodles. Leveraging this unique geometric relationship, we introduce an algorithmic pipeline for generating helically interlocked cellular segmentation of arbitrary parts followed by layer-wise cell sequencing and path planning for cooperative 3D printing. Furthermore, we introduce a novel concurrence measure that quantifies the amount of printing parallelization across multiple robots. Consequently, we integrate this measure to optimize the location and orientation of a part for maximally parallel printing. We systematically study the relationship between the helix parameters (i.e., cellular interlocking), the cell size, the amount of concurrent printing, and the total printing time. Our study revealed that both concurrence and time to print primarily depend on the cell size, thereby allowing the determination of interlocking independent of time to print. To demonstrate the generality of our approach with respect to part geometry and the number of robots, we implemented two cooperative 3D printing systems with two and three printing robots and printed a variety of part geometries. Through comparative bending and tensile tests, we show that helically interlocked part segmentation is robust to gaps between segments.

The immense number of parameters and deep neural networks make large language models (LLMs) rival the complexity of human brains, which also makes them opaque "black box" systems that are challenging to evaluate and interpret. AI Psychometrics is an emerging field that aims to tackle these challenges by applying psychometric methodologies to evaluate and interpret the psychological traits and processes of artificial intelligence (AI) systems. This paper investigates the application of AI Psychometrics to evaluate the psychological reasoning and overall psychometric validity of four prominent LLMs: GPT-3.5, GPT-4, LLaMA-2, and LLaMA-3. Using the Technology Acceptance Model (TAM), we examined convergent, discriminant, predictive, and external validity across these models. Our findings reveal that the responses from all these models generally met all validity criteria. Moreover, higher-performing models like GPT-4 and LLaMA-3 consistently demonstrated superior psychometric validity compared to their predecessors, GPT-3.5 and LLaMA-2. These results help to establish the validity of applying AI Psychometrics to evaluate and interpret large language models.

Abstract The evolution of multimodal large language models (LLMs) capable of processing diverse input modalities (e.g., text and images) holds new prospects for their application in engineering design, such as the generation of 3D computer-aided design (CAD) models. However, little is known about the ability of multimodal LLMs to generate 3D design objects, and there is a lack of quantitative assessment. In this study, we develop an approach to enable LLMs to generate 3D CAD models (i.e., LLM4CAD) and perform experiments to evaluate their efficacy where GPT-4 and GPT-4V were employed as examples. To address the challenge of data scarcity for multimodal LLM studies, we created a data synthesis pipeline to generate CAD models, sketches, and image data of typical mechanical components (e.g., gears and springs) and collect their natural language descriptions with dimensional information using Amazon Mechanical Turk. We positioned the CAD program (programming script for CAD design) as a bridge, facilitating the conversion of LLMs’ textual output into tangible CAD design objects. We focus on two critical capabilities: the generation of syntactically correct CAD programs (Cap1) and the accuracy of the parsed 3D shapes (Cap2) quantified by intersection over union. The results show that both GPT-4 and GPT-4V demonstrate great potential in 3D CAD generation by just leveraging their zero-shot learning ability. Specifically, on average, GPT-4V outperforms when processing only text-based input, exceeding the results obtained using multimodal inputs, such as text with image, for Cap 1 and Cap 2. However, when examining category-specific results of mechanical components, the prominence of multimodal inputs is increasingly evident for more complex geometries (e.g., springs and gears) in both Cap 1 and Cap 2. The potential of multimodal LLMs to improve 3D CAD generation is clear, but their application must be carefully calibrated to the complexity of the target CAD models to be generated.

Abstract Network-based analyses have effectively understood customer preferences through interactions between customers and products, particularly for tailored product design. However, research applying this analysis to diverse customers with varied preferences is limited. This paper introduces a market-segmented network modeling approach, guided by customer preference, to explore heterogeneity in customers’ two-stage decision-making process: consideration-then-choice. In heterogeneous markets, customers with similar characteristics or purchasing similar products can exhibit different decision-making processes. Therefore, this method segments customers based on preferences rather than just characteristics, allowing for more accurate choice modeling. Using joint correspondence analysis, we identify associations between customer attributes and preferred products, characterizing market segments through clustering. We then build individual bipartite customer–product networks and apply the exponential random graph model to compare the product features influencing customer considerations and choices in various market segments. Using a US household vacuum cleaner survey, our method detected different customer preferences for the same product attribute at different decision-making stages. The market-segmentation model outperforms the non-segmented benchmark in prediction, highlighting its accuracy in predicting varied customer behaviors. This study underscores the vital role of preference-guided segmentation in product design, illustrating how understanding customer preferences at different decision stages can inform and refine design strategies, ensuring products align with diverse market needs.

Abstract System design has been facing the challenges of incorporating complex dependencies between individual entities into design formulations. For example, while the decision-based design framework successfully integrated customer preference modeling into optimal design, the problem was formulated from a single entity’s perspective, and the competition between multiple enterprises was not considered in the formulation. Network science has offered several solutions for studying interdependencies in various system contexts. However, efforts have primarily focused on analysis (i.e., the forward problem). The inverse problem still remains: How can we achieve the desired system-level performance by promoting the formation of targeted relations among local entities? In this study, we answer this question by developing a network-based design framework. This framework uses network representations to characterize and capture dependencies and relations between individual entities in complex systems and integrate these representations into design formulations to find optimal decisions for the desired performance of a system. To demonstrate its utility, we applied this framework to the design for market systems with a case study on vacuum cleaners. The objective is to increase the sales of a vacuum cleaner or its market share by optimizing its design attributes, such as suction power and weight, with the consideration of market competition relations, such as inter-brand triadic competition involving three products from different brands. We solve this problem by integrating an exponential random graph model (ERGM) with a genetic algorithm. The results indicate that the new designs, which consider market competition, can effectively increase the purchase frequency of specific vacuum cleaner models and the proposed network-based design method outperforms traditional design optimization.

Abstract The evolution of multimodal large language models (LLMs) capable of processing diverse input modalities (e.g., text and images) holds new prospects for their application in engineering design, such as the generation of 3D computer-aided design (CAD) models. However, little is known about the ability of multimodal LLMs to generate 3D design objects, and there is a lack of quantitative assessment. In this study, we develop an approach to enable two LLMs, GPT-4 and GPT-4V, to generate 3D CAD models (i.e., LLM4CAD) and perform experiments to evaluate their efficacy. To address the challenge of data scarcity for multimodal LLM studies, we created a data synthesis pipeline to generate CAD models, sketches, and image data of typical mechanical components (e.g., gears and springs) and collect their natural-language descriptions with dimensional information using Amazon Mechanical Turk. We positioned the CAD program (programming script for CAD design) as a bridge, facilitating the conversion of LLMs’ textual output into tangible CAD design objects. We focus on two critical capabilities: the generation of syntactically correct CAD programs (Cap1) and the accuracy of the parsed 3D shapes (Cap2) quantified by intersection over union. The results show that both GPT-4 and GPT-4V demonstrate potential in 3D CAD generation. Specifically, on average, GPT-4V outperforms when processing only text-based input, exceeding the results obtained using multimodal inputs, such as text with image, for Cap 1 and Cap 2. However, when examining category-specific results of mechanical components, while the same trend still holds for Cap 2, the prominence of multimodal inputs is increasingly evident for more complex geometries (e.g., springs and gears) in Cap 1. The potential of multimodal LLMs in enhancing 3D CAD generation is clear, but their application must be carefully calibrated to the complexity of the target CAD models to be generated.

Abstract In multi-agent Bayesian optimization for Design Space Exploration (DSE), identifying a communication network among agents to share useful design information for enhanced cooperation and performance, considering the trade-off between connectivity and cost, poses significant challenges. To address this challenge, we develop a distributed multi-agent Bayesian optimization (DMABO) framework and study how communication network structures/connectivity and the resulting cost would impact the performance of a team of agents when finding the global optimum. Specifically, we utilize Lloyd’s algorithm to partition the design space to assign distinct regions to individual agents for exploration in the distributed multi-agent system (MAS). Based on this partitioning, we generate communication networks among agents using two models: 1) a range-limited model of communication constrained by neighborhood information; and 2) a range-free model without neighborhood constraints. We introduce network density as a metric to quantify communication costs. Then, we generate communication networks by gradually increasing the network density to assess the impact of communication costs on the performance of MAS in DSE. The experimental results show that the communication network based on the range-limited model can significantly improve performance without incurring high communication costs. This indicates that increasing the density of a communication network does not necessarily improve MAS performance in DSE. Furthermore, the results indicate that communication is only beneficial for team performance if it occurs between specific agents whose search regions are critically relevant to the location of the global optimum. The proposed DMABO framework and the insights obtained can help identify the best trade-off between communication structure and cost for MAS in unknown design space exploration.

Abstract Swarm manufacturing (SM) is an emerging manufacturing paradigm that employs a heterogeneous swarm of robots to accomplish complex hybrid manufacturing tasks. Cooperative 3D Printing (C3DP), a special form of swarm manufacturing, uses multiple printers to print large-scale parts cooperatively and aims to tackle key challenges in the additive manufacturing industry, such as trade-offs among size, speed, quality, and cost. A fundamental challenge in C3DP is how to achieve collision-free, time-efficient printing when multiple printers operate in a shared workspace. This is a complex problem since the solution may depend on a myriad of factors, such as the number of printers, part geometry, printer positioning, mobility, and kinematics, or whether the printing path pre-determined. In this paper, we present SafeZone, a collision-free and scalable C3DP framework that aims to minimize printing time by considering both the geometry and topology (space-connectivity) of the resulting workspace when segmenting the part layer. To achieve this, we use a guided Voronoi tessellation that can only produce degree-3 partitions, which we show to have optimal scheduling properties based on the chromatic number of the resulting partition graph. The sites of the Voronoi tessellation are constrained to only lie on the boundary of their convex hull, thus facilitating collision-free operation in C3DP systems with robotic arms. We demonstrate through physical testing in a 4-printer scenario with SCARA arms that SafeZone can produce collision-free prints, resulting in a printing time reduction of 44.63% when compared to the single-printer scenario. Finally, we show how the partition created by our methodology has a printing time reduction of 22.83% when compared to a naive choice which does not consider workspace topology.