Abstract This article proposes a design optimization strategy aimed at improving the performance of a linear belt drive system. The elasticity of the belt and nonlinear disturbances causes undesirable vibrations and reduces the trajectory tracking accuracy, affecting the precise positioning of the drive and reducing its efficiency. These issues are addressed by developing kinematics and dynamics models of the drive, which include position-dependent stiffness, damping, and nonlinear frictions. Based on these models, design optimization and control strategies are devised to enhance the efficiency of the system. To analyze and further enhance the performance of the system under actual operating conditions, a virtual prototype (VP) model of this drive is developed using the multi-body dynamics simulations tool. This technique reduces the time and cost of prototyping by enabling design iterations without the need for physical prototyping. The results obtained from the VP and analytical modeling are validated experimentally to ensure their accuracy and effectiveness in capturing the behavior of the drive. In the end, a reliable and cost-effective solution is provided for designing a high performance linear belt drive system.
This article introduces a novel three-degree-of-freedom (3-DOF) dual-mode parallel manipulator (DMPM) capable of switching between 2R1T (2-DOF rotation and 1-DOF translation) and 3T (3-DOF translation) motion modes. The kinematic architecture comprises three identical PU(R)U legs, each incorporating a bistable passive rotary joint (R) that enables tool-free mode transitions without external actuation. Screw-theory-based mobility analysis confirms that the bistable joint angle governs constraint-topology transformation between modes. Multiobjective particle swarm optimization (MOPSO) combined with the technique for order preference by similarity to ideal solution (TOPSIS) was employed to optimize workspace volume and global dexterity index (GDI) across both operational modes. Experimental validation demonstrates that the optimized design achieves positioning accuracy below 0.85-mm root mean squared (RMS) error compared to over 16.81 mm for nonoptimized configurations--an approximately 20-fold improvement-with translational repeatability below 0.04 mm. Payload experiments further verify that the spring-based bistable locking maintains joint stability under operational loads without unintended mode transitions. The DMPM is well suited to applications alternating between orientation-intensive and translation-intensive tasks, such as precision electronics assembly and automated surface inspection.
The subjective evaluation of early-stage engineering designs, such as concept sketches, traditionally relies on human experts. However, expert evaluations are time-consuming, expensive, and sometimes inconsistent. Recent advances in vision-language models (VLMs) offer the potential to automate design assessments, but it is crucial to ensure that these artificial intelligence (AI) "judges" perform on par with human experts. This work introduces in-context learning (ICL)-enhanced VLM judges and a comprehensive statistical framework (including agreement, error, correlation, statistical difference checks, equivalence testing, and top-set overlap) to rigorously assess AI-expert equivalence. Across two case studies, we show that reasoning-enabled VLMs are the strongest-performing AI judges. They consistently outperform two-third trained novices across all metrics, and for measures such as uniqueness, creativity, and drawing quality, they approach expert-equivalent performance. In specific cases, they even exceed expert-expert agreement, attaining lower mean absolute error and higher rank correlations than the expert baseline. These findings suggest that, on certain statistical tests, AI judges are not only approaching expert-expert equivalence but in some cases surpassing it.
This article presents a new adaptive sequential sampling strategy for surrogate models based on the maximin distance criterion space filling of the system response quantity (SRQ). The strategy enables sequential adaptive sampling of an experimental design while attempting to cover SRQ and input spaces at each stage of the adaptive sequential surrogate model construction process. The proposed adaptive sampling strategy selects a new sample point from a pool of candidate design points based on dual maximin distance designs in the SRQ and input spaces, with a clear sequential screening mechanism that first ranks candidates by predicted SRQ space dispersion and then optimizes input space space filling. The proposed criterion for the sample selection balances both filling predicted SRQ space of the surrogate model and the input space in a computationally efficient heuristic framework. This article adopts polynomial chaos Kriging as the surrogate model. The superiority of the SRQ-Mm method over pure SRQ space filling and input space filling is discussed in detail through theoretical analysis and numerical simulations. The proposed strategy is also compared with the widely used EIGF and MiVor adaptive approaches. The numerical results confirm its superiority over existing adaptive sampling approaches in terms of surrogate model accuracy, computational efficiency, and robustness.
Sustainability objectives in the aeronautical industry push companies toward novel aircraft propulsive systems (e.g., hybrid electric or hydrogen-based). Coupled with reliability constraints for certification, this creates a potentially highly complex propulsive architecture design. Thus, tools and methods are needed to explore this growing system design space efficiently. Several critical features have been identified by system architects in one of the leading companies in this domain: consideration of parallel functional chains for safety and redundancy reasons, systematic exploration of the design space, and the possibility to systematically generate from functional domain onto the physical domain of the system architecture. The article proposed a Design Structure Matrix (DSM)-based system architecture generation and design methodology with the objective to consider both parallel and branching system architectures, while supporting this systematic generation from the functional to the physical domain. The methodology was compared with previous case studies. Moreover, it has been tested and validated in the industrial setting. Both the generation process and the constraints were tested for the design of aircraft propulsive systems. The aim of this research is to propose a first step allowing for systematic exploration of system architectures that can be further integrated into existing simulation workflows, enabling the performance of propulsion architectures to be estimated systematically and more quickly than with manual analyses.
Abstract For many modern design problems, e.g., in clinical device design, evaluating designs prior to experimental testing typically requires high-fidelity multiphysics models that may involve fluid–structure interaction (FSI), multiple spatial and temporal scales, and dependencies on open systems. Such simulations can take hours or days, making them impractical for incorporation into computational optimization protocols. Surrogate models that provide rapid assessments are therefore essential. Traditional data-driven surrogates are fast but offer limited interpretability, while reduced-order models capture physical mechanisms but depend on simplifying assumptions and prior knowledge. Mechanistic learning (MxL) addresses these limitations by combining machine learning with reduced-order modeling to create a surrogate that retains mechanistic insight while achieving high computational efficiency. Here, we propose to integrate MxL-based surrogates into a computational optimization protocol. We showcase this methodology in the design of hydrocephalus shunt systems. Starting from a 3D finite element FSI model, we motivate the use of a fast surrogate model in which fluid stresses on the FSI boundary are combined with machine learning techniques to predict solid deformation. The resulting MxL surrogate is integrated into a hybrid optimization algorithm, reducing evaluation time from 3 days to 10 min and enabling tractable optimization of a high-dimensional design space. The resulting designs are ultimately tested in the full FSI model to confirm their improved efficacy. This methodology is applicable to engineering design problems characterized by computationally expensive multiphysics models, particularly those involving dominant FSI effects.
Engineering rulebooks and technical standards contain multimodal information like dense text, tables, and illustrations that are challenging for retrieval augmented generation (RAG) systems. Building upon the DesignQA framework (Doris, A. C., Grandi, D., Tomich, R., Alam, M. F., Ataei, M., Cheong, H., and Ahmed, F., 2025, "Designqa: A Multimodal Benchmark for Evaluating Large Language Models' Understanding of Engineering Documentation," J. Comput. Inf. Sci. Eng., 25(2), p. 021009. 10.1115/1.4067333), which relied on full-text ingestion and text-based retrieval, this work establishes a multimodal ColPali-enhanced retrieval and reasoning framework (MCERF), a system that couples a multimodal retriever with large language model reasoning for accurate and efficient question answering from engineering documents. The system employs ColPali, which retrieves both textual and visual information, and multiple retrieval and reasoning strategies: (i) hybrid lookup mode for explicit rule mentions, (ii) vision to text fusion for figure- and table-guided queries, (iii) high-reasoning LLM mode for complex multi modal questions, and (iv) SelfConsistency decision to stabilize resp onses. The modular framework design provides a reusable template for future multimodal systems regardless of the underlying model architecture. Furthermore, this work establishes and compares two routing approaches: a single-case routing approach and an agent-based system, both of which dynamically allocate queries to optimal pipelines. Evaluation on the DesignQA benchmark illustrates that this system improves average accuracy across all tasks with a relative gain of +32.6% from baseline RAG best results, which is a significant improvement in multimodal and reasoning-intensive tasks without complete rulebook ingestion. This shows how vision-language retrieval, modular reasoning, and adaptive routing enable scalable document comprehension in engineering use cases. MCERF is publicly available online.
An adaptive sampling acquisition criterion is developed to improve global surrogate model accuracy for computationally expensive engineering simulations using Gaussian process (GP) regression. The proposed approach combines integrated mutual information (IMI), which quantifies expected global information gain over a reference set, with GP predictive uncertainty that characterizes local model uncertainty. These complementary quantities are normalized over the candidate set and integrated through a dynamically updated weighting mechanism to form a hybrid entropy-uncertainty (HEU) acquisition criterion, in which the blending weight is sequentially adjusted based on feedback from newly observed samples to adaptively balance global information gain and local uncertainty during sampling. The performance of HEU is evaluated under constrained evaluation budgets using several analytical benchmark function of varying dimensionality together with a two-dimensional steady-state temperature-field reconstruction problem, with identical initial designs and fixed sampling budgets across methods. Statistical reliability is assessed through repeated independent runs. To characterize convergence behavior under prescribed accuracy requirements, a dual-threshold samples-to-threshold metric based on two global performance measures is introduced. Samples to threshold is defined as the first iteration at which the accuracy criteria are both satisfied within a fixed sampling budget. Results indicate that HEU achieves competitive and stable global error reduction relative to representative adaptive sampling strategies, demonstrating consistent sample efficiency under constrained budgets.
Microneedle patches are emerging minimally invasive devices that deliver drugs using micron-scale needles to penetrate the stratum corneum of the skin, offering significant advantages such as minimal pain, convenient administration, and low risk of infection. However, microneedle patches are typically applied by thumb pressing, making it difficult to limit the maximum input force, which can cause needle damage or even nerve contact causing pain. This article proposes a force-limiting applicator for microneedle patches, which limits the maximum input force through a sheet-based contact-aided compliant force-limiting mechanism (CCFM). The sheet-based CCFM consists of a compliant sheet and a contact rod (with smooth contact surface). As the contact surface contacts the compliant sheet, causing it to deform, the reaction force gradually increases. When the contact surface separates from the compliant sheet, the deformation disappears, leading to a sudden reduction in the reaction force. The sheet-based CCFM contains no electronics, making the applicator easy to miniaturize, reusable, and cost-effective. In addition, by designing different dimensions of the compliant sheet and contact surface, the applicator can satisfy various force-limiting requirements. The chained pseudo-rigid-body model (CPRBM) is used to predict the deformation of the compliant sheet in contact with linear and curved contact surfaces. The model is validated using finite element analysis (FEA), with maximum relative errors of 6.61% and 6.79%, respectively. The reaction force is obtained using the Lagrange multiplier method. For the two different contact surfaces, the maximum relative errors between the analytical and FEA results are 9.3% and 6.8%. In addition, physical prototypes are manufactured and experimental tests are performed to further verify the feasibility of the design. Finally, the design procedures, the limitation of the analytical model, and future work are discussed.
Soft grippers, used in applications such as food handling and assistive devices, leverage multiple soft fluidic actuators (SFAs) for safe and compliant grasping. Designing SFAs is challenging because they must satisfy multiple functional requirements while operating outside the principles of rigid machine design, as they undergo large deformations and exhibit material nonlinearity. Because fabricating numerous design candidates is costly, computational tools have emerged to expedite the search for optimal designs. However, existing computational tools do not focus on SFA design optimization for state-specific grasping, where actuators are optimized for a particular deformation dictated by the intended use case. Moreover, many existing tools support a limited range of performance metrics and optimization modes. Here, we present PneuGrasp, an open-source tool for the design optimization of SFAs according to a user-specified grasping task. The tool can analyze design candidates across multi-functional combinations of seven performance metrics, including the understudied metrics of grasping force, actuation speed, and actuation energy. In addition, PneuGrasp supports three optimization modes that together provide parameter intuition and shorten optimization time. Through a series of examples evaluating over 1000 design candidates, we demonstrate that PneuGrasp can identify optimized designs that outperform our baseline. For instance, one design achieved a 60% reduction in maximum strain and a 52% reduction in actuation volume, while another showed a 405% decrease in a combined durability-grasping-force performance score. We fabricated and tested over 30 actuators across five distinct designs, demonstrating PneuGrasp's relative prediction capabilities. PneuGrasp can be found online.
This study proposes a selection and stacking ensemble-based method to facilitate the determination of the covariance function type for Gaussian process regression. The proposed method operates at the model ensemble level and involves the use of multiple Gaussian process regression models with different types of covariance functions as base learners in a stacking ensemble with a final Gaussian process regression model as the meta-learner. First, the Pearson correlation coefficients between the leave-one-out predicted responses from each candidate base learner and the actual responses are computed and sorted in ascending order, after which the interquartile range (IQR) is calculated, and the candidate base learners that fall below the lower 1.5 & times; IQR are removed. Afterward, the adjacent gaps between the Pearson correlation coefficients that correspond to the remaining candidate base learners are calculated and sorted in ascending order, and some of the remaining candidate base learners are further removed according to the gaps that lie above the upper 1.5 & times; IQR. Finally, a newly constructed Gaussian process regression model with a linear covariance function is used as the meta-learner for final predictions. To validate the effectiveness of the proposed method, six analytical test functions, three engineering datasets, and one simulation case are used for a performance study along with three representative approaches. The results demonstrate that the proposed method achieves competitive accuracy and generalization ability. Furthermore, its performance is evaluated across four widely used open-source toolkits for Gaussian process regression, and the results confirm the robustness of the method.
Abstract The number of manufacturing jobs in the US has been consistently increasing, driven by a rapidly evolving industrial landscape and the implementation of a new strategic plan. At the same time, concerns have emerged about the problem-solving abilities of engineering students, who represent the future workforce. This highlights the need for systematic evaluation and deeper insight into how these students approach problem-solving. In this article, we introduce a virtual reality (VR)-based manufacturing environment combined with a data-driven analytical workflow to evaluate engineering students’ problem-solving performance. Within the VR system, students complete assembly tasks to build car toys that meet specific design criteria. During the process, we capture real-time eye-tracking data, reflecting the spatial and temporal dynamics of their visual attention and assembly actions. We extract latent features from this data via a long short-term memory-based supervised representation learning for problem-solving performance evaluation. Our approach outperforms the traditional performance metrics-based evaluation by capturing the nonlinear dynamics of the in situ problem-solving process. Experimental results, including benchmarking against alternative architectures and ablation analyses, show that the learned feature representations yield the clearest distinctions in categorizing students' problem-solving performance among the evaluated methods when integrating full behavioral input. The proposed evaluation framework holds broader potential for improving problem-solving assessments across various manufacturing systems and workforce training programs.
Robotic grippers with integrated sensing capabilities exhibit significant potential in interactive manipulation tasks. However, existing studies typically concentrate tactile sensors at the fingertips, overlooking the critical role of the palm during grasping, and thus the design of sensor-integrated palms remains insufficiently explored. To address this issue, this article proposes a robotic gripper based on a multi-spherical-joints self-adaptive palm structure. By strategically combining multiple levels of spherical joints, the palm passively conforms to the object during grasping and readily accommodates embedded orientation sensors. Using the measured joint pose angles in combination with a surface-fitting algorithm, the gripper can rapidly reconstruct the surface model of the object. Experimental results demonstrate that the proposed perception method is accurate and reliable, and that the palm structure exhibits excellent compliance with objects of various shapes, providing a solid reference for future designs of sensor-integrated robotic palms.
Managing design complexity is a critical challenge in modular product families, especially when diverse module variants must be combined via standard interfaces. This study proposes a methodology for interface standardization in modular product families by modeling design relationships at the module variant level. Unlike conventional approaches that assume fixed or single interfaces, the proposed framework explicitly considers multiple interface configurations and their impact on design complexity. Two types of complexity are defined: standardization effort, which refers to the coordination effort required to develop standard interfaces, and integration effort, which reflects the structural complexity arising from integrating module variants with interfaces. Based on these measures, the interface design problem is formulated as a decision framework that balances the trade-off between standardization and integration efforts. A case study on an automotive front chassis system demonstrates the applicability of the approach, showing how different interface configurations lead to distinct complexity outcomes and how Pareto-efficient solutions can be identified. The results provide practical insights into interface design strategies, highlighting that the optimal level of standardization depends on the dominant source of complexity.
Reliability-based design optimization (RBDO) faces severe challenges due to its nested double-loop structure and high computational cost associated with reliability analysis. For this purpose, this article proposes a deep generative modeling framework for the RBDO problem using conditional normalizing flow (cNF). The proposed framework consists of a training phase and an optimization phase. In the training phase, a cNF is constructed and trained to learn an invertible mapping between the response distribution of the performance function and a standard Gaussian base distribution, conditioned on the design parameters. The flow model is built by composing multiple transformation layers, including affine coupling layers and neural spline layers. In the optimization phase, once the flow model has been trained, the response distributions at any given design point can be efficiently evaluated using the change-of-variable formula, enabling direct and exact computation of the associated failure probabilities without additional performance function evaluations. As a result, the original RBDO problem is reformulated as a deterministic optimization problem, in which the probabilistic constraints are evaluated through the trained flow model. Consequently, standard deterministic optimization algorithms can be readily employed to search for the optimal design point. Four numerical examples are presented to demonstrate the effectiveness of the proposed method.
Minimizing the energy consumption of robot manipulators is essential to address both environmental and economic challenges. This study introduces a novel design concept for manipulators that reduces energy use in both static and dynamic operating modes. The key idea is to employ straight-line guiding linkages as the manipulator's actuating system. This approach enables the elimination of static loads on the actuators without fixing the manipulator's overall center of mass. Instead, the center of mass follows a rectilinear horizontal trajectory, maintaining constant potential energy. As a result, the system requires fewer counterweights and experiences a smaller increase in total moving mass. Two manipulator architectures are presented, inspired by the Scott-Russell and four-bar mechanisms. For dynamic operation, an optimal design method is developed to minimize input torques. By carefully tuning the counterweight parameters, the manipulator achieves energy-efficient performance across a family of "Pick-and-Place" trajectories, each executed according to a "Bang-Bang" motion control law. The results clearly illustrate the transition between static and dynamic modes and demonstrate a substantial reduction in input torque in both cases. The proposed method is generalizable to other manipulator types and provides an effective framework for optimizing robotic energy performance.
Generative design has emerged as an effective approach for exploring complex engineering design spaces beyond the limitations of conventional optimization, particularly for thermal management systems with high-dimensional geometric variability; however, its application to liquid-cooled cold plates remains challenging under asymmetric thermal loading conditions. To address this issue, this study proposes an enhanced cross-attention conditional diffusion framework in which physics-based conditional variables are injected into the U-Net denoiser via multi-scale cross-attention and feature-wise linear modulation, enabling condition-aware geometry generation throughout the diffusion process and thereby preserving high-fidelity geometric and physical consistency. Training data are obtained using multi-objective topology optimization to generate a diverse set of cold-plate geometries together with their corresponding pressure drop Delta p and average surface temperature difference T under asymmetric thermal loading. This dataset is then used to train the proposed diffusion model, generating new cold-plate geometries under prescribed physical conditions. Promising design results are efficiently identified using a ResNet18-based surrogate model for rapid thermal-hydraulic performance evaluation, avoiding reliance on extensive full-order simulations. Overall, the proposed framework integrates physics-based optimization with generative modeling to provide an efficient and effective approach for designing high-performance liquid-cooled cold plates in asymmetric thermal environments.
Abstract Artificial intelligence (AI)-driven surrogate modeling has emerged as an effective alternative to physics-based simulations for 3D design, analysis, and manufacturing. These models use data-driven techniques to predict physical quantities that traditionally require computationally expensive simulations. However, the scarcity of labeled CAD-to-simulation datasets has motivated the development of self-supervised and foundation models, in which geometric representation learning is performed offline and later adapted to downstream tasks using limited labeled data. While promising, existing approaches often struggle in applications that require accurate preservation of fine-scale geometric details. This work introduces a self-supervised geometric representation learning method designed to capture fine-scale geometric features from non-parametric 3D models. Unlike traditional end-to-end surrogate models, the proposed approach decouples geometric feature extraction from downstream physics prediction by learning a latent representation guided solely by geometric reconstruction losses. Key components include near-zero-level signed distance field sampling and a batch-adaptive attention-weighted loss function, which together enhance sensitivity to subtle yet physically influential geometric variations. The proposed method is validated through two case studies involving high-dimensional design parameter regression, achieving coefficients of determination exceeding 0.98, as well as structural mechanics tasks that demonstrate strong few-shot prediction performance for reaction forces and deformation fields. Comparisons with parametric surrogate models further illustrate our method's ability to bridge geometric and physics-based representations, providing an effective surrogate modeling solution in data-scarce settings.
To address the challenge of trajectory deviation during automated tape laying (ATL) on variable-curvature inclined surfaces in aerospace applications, and to overcome the limitations of existing correction mechanisms in terms of flexibility and stability, this study proposes a novel swing-translation hybrid correction (STHC) mechanism. First, based on an analysis of functional requirements for tape correction, the design requirements for the STHC mechanism are identified. Second, a synthesis method for swing-translation hybrid motion mechanisms based on trajectory guidance supplemented by virtual constraint enhancement is proposed. Based on this approach, a 2R2P-RP-P configuration with a symmetric structure and favorable stability is designed. Subsequently, the kinematic performance of different links in this configuration as actuated joints is analyzed to identify the driving joints that meet the requirements for motion symmetry, and then design a compact STHC mechanism. Then, a kinematic theoretical model of the STHC mechanism in correcting the position of the tape is established, and its correctness is verified through finite element simulation. Finally, a prototype of the STHC mechanism is developed and integrated into an ATL application platform. Experimental results demonstrate its excellent motion performance and tape correction capability. This study provides a novel solution for the automated forming of composite materials on variable-curvature inclined surfaces, highlighting its significant practical application value.
Abstract Addressing the high cost of building optimization models in conceptual mechanical design, this article presents MecSOAgent, a closed-loop multiagent framework that leverages large language models (LLMs) and an auditable mechanical system optimization graph (MSOG) to translate natural-language requirements into executable optimization scripts and SysML 2.0 physical system models. Domain knowledge is injected via retrieval-augmented generation, while the MSOG persists a traceable mapping from requirements to physical entities, design variables, constraints, objectives, and solution algorithms, enabling consistency checking and engineering audit. To improve reliability, MecSOAgent introduces a deterministic intermediate representation based on a design structure matrix (DSM) that converts relation matrices into incremental updates of the Knowledge Graph, and decouples multiobjective preference modeling from solving via analytic hierarchy process (AHP) with consistency checks. A checklist-based reviewer reconciles graph artifacts with solver outputs to trigger iterative repair. A case study on redundancy configuration optimization for an airborne hydraulic actuation system, along with multidimensional comparative experiments against end-to-end LLMs and general-purpose programming agents, demonstrates that MecSOAgent effectively mitigates semantic drift and physical-logic deficiencies. By automatically instantiating system-level optimization models and generating verifiable artifacts, the proposed framework approaches human-expert reliability, offering a practical path toward Model-Based Systems Engineering-oriented automation of mechanical system optimization design.
