Topology optimization seeks the optimal material distribution under specific loads, boundary conditions, and design constraints to maximize performance or minimize material usage. Deep learning-based approaches have been studied in this process by generating optimized topologies. However, existing data-driven topology optimization methods often exhibit limited physical consistency, insufficient generalization across loading and boundary conditions, and a trade-off between prediction accuracy and computational efficiency. To address these issues, we propose Multi-Stage Contrastive cGANs (MSC-GANs) for topology optimization. The inputs of MSC-GANs consist of physical information from the initial design domain, specifically the applied loads and volume fraction. Expressiveness is enhanced by minimizing iteration losses through an optimized multi-stage generator network and incorporating a physics-informed loss function. By integrating an Improved Residual Network (IResNet) into the generator, the computation time is reduced by 40
The growing complexity and introduction of intelligence in industrial products has led to an unprecedented growth in the complexity of their design organizations. In this context, the industrial paradigm of the Extended Enterprise allows to better leverage the expertise of the different actors, and to create strong industrial synergies. However, the growing number of stakeholders also carries a number of difficulties, especially in the information exchange domain. Actors of varying - and sometimes competing - interests exchange product information at the interface of the subsystems they develop. These exchanged information present different levels of importance for each actor, along with different levels of confidentiality or Intellectual Property issues. Information that is both important for the development of adjacent subsystems and confidential for some actors will thus be qualified of critical, in the sense that their exchange during later design phases present high risks of causing delays and errors. The early identification of such critical information in the design process allows the actors to adapt and plan information exchange accordingly, thus strongly contributing to securing the project. In order to ease this early identification, we present a tool-supported, structured and formal method. The CIPHER Method is based on Systems Engineering principles, with a recursive stakeholder cartography and a dual evaluation of information importance and confidentiality. These evaluations are normalized using a regression function and used to compute actor-to-actor and information-specific criticality indices. A prototype tool was developed to support the method, and its application was demonstrated through a workshop involving experts from an urban railway project. Applied to a simplified urban railway development project, the method identified 27 exchanged information items and 16 directed information interfaces between six stakeholders. The computed criticality indices ranged from 0.097 to 2.027, allowing the identification of four highly critical interfaces requiring particular attention. A usability evaluation involving 16 participants also yielded positive results, with the highest scores obtained for Learning (mean = 2.13), followed by Appreciation and Efficiency (mean = 1.94). These results highlight the method’s ability to identify high-risk information exchanges early in the design process and support proactive risk mitigation.
High-precision industrial manufacturing requires reliable and automated surface defect inspection to improve product quality and reduce the limitations of manual visual inspection. This study presents a deep learning-based framework for the classification and semantic segmentation of surface defects in machine tool components using the Metallic Surface Defect Dataset (GC10-DET). Three convolutional neural network (CNN) models (AlexNet, ResNet18 and VGG16) were compared for defect classification, and three semantic segmentation models (U-Net, Fully Convolutional Network (FCN) and DeepLabV3) were studied for defect localization. A Lévy Arithmetic Algorithm (LAA) was employed to transform the hyperparameter optimization process and make the model training process more effective and robust. The experimental results showed that AlexNet achieved the best classification results in terms of validation accuracy (0.997) and F1 score (0.977). For semantic segmentation, FCN outperformed the other evaluated architectures, achieving an Intersection over Union (IoU) of 0.725 and a Dice coefficient of 0.834, while maintaining near real-time computational performance suitable for automated industrial inspection. Comprehensive quantitative and qualitative analyses, including confusion matrices, learning curves, computational efficiency, and comparison with recent state-of-the-art studies, demonstrate the effectiveness of the proposed evaluation framework. The results offer practical suggestions when adopting appropriate deep learning architectures for industrial surface defect inspection, and present a possible tradeoff between segmentation performance and computational requirements.
This study presents a deployment-aware framework for explainable remaining useful life prediction of turbofan engines, integrating metaheuristic hyperparameter optimisation with deep learning and SHAP-based explainability. Long Short-Term Memory (LSTM) models are optimised using a genetic algorithm (GA) and particle swarm optimisation (PSO), while an untuned bidirectional LSTM (BiLSTM) serves as the architecturally stronger reference. This asymmetric design quantifies the regime-dependent conditions under which automated tuning of a single-direction LSTM substitutes for the additional cost of a bidirectional architecture. The framework is evaluated on the NASA C-MAPSS benchmark across all four sub-datasets (FD001–FD004) under an engine-wise validation split that eliminates window-level data leakage and across N=10 independent runs analysed with Kruskal-Wallis and Dunn post-hoc tests. Results demonstrate that the substitution is regime-dependent. Under single-condition operation (FD001), the mean RMSE values converge within 14.74–14.86 cycles with no significant differences ( p=0.545 ); PSO-LSTM attains the numerically lowest mean ( 14.736 ± 0.489 cycles) and BiLSTM has the lowest variance (CV = 1.51% ). GA-LSTM achieves the lowest mean RMSE under dual-fault operation ( 14.015 ± 0.636 cycles on FD003). Under multi-condition operation (FD002, FD004), BiLSTM delivers the lowest mean RMSE with minimal variability (30.83 and 35.28 cycles), while metaheuristic-tuned LSTMs collapse to the mean-prediction baseline in ten of ten runs on at least one subset. SHAP attributions are reshaped by engine complexity: HPC outlet pressure (Ps30, phi), exhaust gas temperature (T50) and core rotational speed (Nc) dominate under single-condition operation, while operating-condition descriptors emerge as primary contributors under multi-condition dynamics. The framework provides an automated, interpretable, and statistically validated approach for turbofan engine prognostics.
This study presents an innovative framework comprising experimental and computational studies aimed at developing and evaluating Al6061 hybrid metal matrix composite (MMCs) spur gears strengthened with WS2 and TiO2 reinforcements for power transmission applications. Machine learning (ML)-based material design strategy is implemented to build a predictive model based on literature datasets to forecast mechanical and tribological characteristics as a function of reinforcement. Multi-response optimization using the concept of desirability is employed to pinpoint the best MMC compositions. Five different hybrid compositions are fabricated by the stir casting method and characterized in terms of tensile, hardness, flexural, impact, and wear testing. Scanning electron microscopy (SEM) analysis with energy-dispersive spectroscopy (EDS) was performed to characterize microstructure. The composite with 3 wt.
Detailed 3D kinematics provide crucial insights for evaluating fall injury mechanisms and impact severity; nevertheless, their application remains understudied. While traditional 3D motion capture systems (marker-based and inertial) are highly effective for laboratory research, their real-world applicability is limited. Consequently, real-world monitoring frequently relies on alternative systems that prioritize usability over kinematic accuracy. Recently, advanced AI algorithms have enabled 3D human pose estimation from standard RGB cameras, potentially extending the applicability of 3D fall kinematics analysis in real-world settings. However, their reliability in highly dynamic, occlusion-prone fall scenarios has yet to be established. This study validates two RGB vision-based motion capture frameworks, the multi-camera OpenCap (OC) and the monocular SAM3D Body (3DB), against a reference OpenSim musculoskeletal pipeline driven by an inertial system (i.e., Xsens) during simulated falls. Nine healthy participants performed forward, backward, and lateral falls, and their kinematic trajectories were compared using the Mean Absolute Error (MAE) across 24 rotational and 3 translational degrees of freedom. Results indicate highly comparable tracking capabilities, both yielding a global rotational MAE of 13°. Lower-body kinematics demonstrated moderate errors (<10°), whereas upper-body tracking exhibited higher discrepancies (>15°). Regarding pelvic translation, OC consistently outperformed 3DB (6.1 cm vs. 8.5 cm). Despite non-negligible error margins, these findings confirm that these vision-based algorithms hold great potential in 3D fall kinematics research. Remarkably, the monocular 3DB system performed comparably to the multi-camera OC framework, highlighting the feasibility of further streamlining RGB markerless motion capture through single-camera setups.
The tolerancing of complex features remains a significant challenge, as these geometries are not classified as conventional features of size under current standards. Geometric tolerances, including form, orientation, and position tolerances, are defined to enable designers to control the multiple degrees of freedom of a feature and ensure that it remains within its prescribed tolerance zone. The automatic assessment of complex forms, such as cyclic-symmetric profiles, can considerably facilitate their geometric tolerancing processes. To advance tolerancing analysis for complex polygonal geometries, this paper investigates the challenges associated with such profiles in comparison with conventional features. A methodology is proposed to analyze polygonal profiles by establishing analogies with features of size while integrating machine learning techniques for classification and tolerances specification. The proposed methodology applies regression models as a promising approach for automating tolerance assignment and evaluation for complex geometries. The developed model demonstrates high classification accuracy (0.97–0.99) in identifying geometric-related problems. A key originality of this work lies in the integration of two area and tolerance-based factors. This introduces an innovative framework for the geometric specification and analysis of cyclic-symmetric profiles, which can also be extended to enhance the capability analysis of these complex geometries. The proposed approach is illustrated and validated through two case studies involving the assembly of cyclic-symmetric profiles parts and an isolated trilobe part using 3D tolerancing methods, such as domain techniques, under well-defined assumptions. Results showed that the incorporation of learning algorithms significantly improves the robustness and efficiency of the proposed geometric analysis methodology. Finally, the paper discusses the advantages and limitations of the proposed methodology and highlights its potential integration into advanced geometric tolerancing frameworks for complex profiles. Geometric Evaluation approach for complex profile.
Rotate vector (RV) reducers are widely applied in robotics and precision machinery due to their high positioning precision and strong load-bearing capacity. As a core component of the RV reducer, the cycloid gear directly affects the reducer’s performance. Moreover, an inappropriate cycloidal profile may lead to force concentration, and even cause damage to the mechanical system. Adopting the RV-40E type reducer as a research object, this study proposes a segmented modification approach for the cycloidal profile based on the meshing pressure angle. The standard profile is retained in the working section to ensure conjugate meshing, while the non-working sections (addendum and dedendum) are modified with pressure-angle-based backlashes. The working and non-working sections are smoothly connected via boundary conditions. An optimization model based on a particle swarm algorithm is developed using the minimum pressure angle and bearing distribution coefficient as the optimization objectives. The single- and multi-objective optimization results are solved through iterative computation, and the theoretical optimization results are further validated by ANSYS finite element simulation. The results illustrate that the segmented modified profile achieves a more uniform meshing force distribution, and the minimum pressure angle decreases by 13.9 load distribution coefficient increases by 2.8
This paper presents a design automation framework that integrates geometric reconstruction, constraint modeling, and manufacturable solid generation for irregular, unstructured geometries. The paper also evaluates the use of AI/ML methods to assist with image-to-point cloud conversion and point cloud-to-surface reconstruction. The framework addresses a key gap in design automation: translating geometric data obtained from 3-D scans into parameterized, constraint-satisfying CAD models that can be fabricated using additive manufacturing. Our approach combines data-driven surface reconstruction with constraint-aware generative modeling to ensure that the resulting designs perform well in form, fit, and function. Quantitative evaluations compare analytical and hybrid reconstruction methods based on curvature-derived smoothness and manufacturability metrics, demonstrating the impact of constraint integration on geometric fidelity and design efficiency. The proposed methodology generalizes to diverse domains that require customized geometries, such as wearables, tooling, and adaptive interfaces, and is validated through an anatomically inspired case study. By bridging geometric data processing, computational modeling, and design synthesis, this work advances the development of constraint-integrated generative design frameworks within engineering design automation.
Industry 4.0 utilizes Additive Manufacturing (AM) due to its ability to offer flexibility and reduce waste; however, quality remains a significant obstacle: problems such as porosity, cracking, delamination, and dimensional defects limit its use in critical sectors. Nevertheless, no existing review offers a systematic classification of defects that encompasses all processes. This research addresses this gap by examining 165 academic works using the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) methodology, with support from an Artificial Intelligence tool (LLaMA3). The results indicate that defects fall into four categories: porosity and inclusions (45
Additive manufacturing (AM), particularly fused deposition modeling (FDM), enables the fabrication of complex geometries through layer wise material deposition directly from digital models. Despite its flexibility, process productivity remains strongly dependent on toolpath planning, especially during the infill phase. In practice, infill toolpath generated by computer-aided manufacturing (CAM) systems often introduces geometric discontinuities, leading to frequent velocity fluctuations and dynamic limitations that degrade programmed feed rates. This work presents a dynamic analysis of infill deposition in FDM, focusing on the interaction between toolpath topology and machine kinematic constraints. A physics-based modeling framework is developed to simulate nozzle trajectories and estimate the resulting actual feed rates under bounded velocity, acceleration, and jerk constraints. The proposed model captures the influence of path curvature, segment transitions, and directional changes on motion smoothness and trajectory execution. A simulation tool is implemented to evaluate multiple infill patterns with respect to their induced kinematic load and printing efficiency. The results reveal that certain CAM generated patterns lead to significant reductions in deposition feed rate due to high frequency deceleration phases and jerk induced limitations, particularly in complex geometries. Conversely, continuous and curvature optimized toolpaths exhibit improved dynamic behavior, enabling higher average feed rates and reduced process time. Experimental findings demonstrate that infill pattern design is a critical factor for enhancing AM performance, and that incorporating machine dynamics into toolpath generation can significantly improve productivity and print quality. This study provides a quantitative framework for evaluating CAM generated infill strategies based on machine dynamic constraints.
Against the backdrop of the apparel industry’s gradual digital transformation, traditional human skeleton recognition and 3D modeling clothing design technologies suffer from challenges such as low efficiency and insufficient accuracy. To achieve efficient and precise batch recognition of human skeletons and 3D modeling clothing design, this study proposes the spatio-temporal channel feature shrinkage network-graph convolution network. The study first constructs a multi-modal fusion acquisition architecture, synchronously collecting data using an arrayed camera cluster and millimeter-wave radar, followed by dynamic calibration and other preprocessing steps. Subsequently, the spatio-temporal channel feature shrinkage network is enhanced through modules such as multi-scale feature fusion and adaptive channel attention to optimize skeleton recognition. Finally, by integrating graph convolutional neural networks, a two-layer graph architecture was used to correlate skeleton and garment parameters, with a dynamic constraint subnetwork embedded to fit the pattern. Experiments indicated that the model achieved 95.74
Product process coupling remains a major limitation in mechatronic system development, as nearly 80 Integrated CAD to process framework based on MBSE and AI
Under the wave of digitalization, the clothing industry is facing profound changes. Consumers have an increasingly strong demand for personalized customized clothing. However, the traditional clothing customization model is cumbersome, communication is poor and costly. At the same time, although augmented reality and artificial intelligence technologies have been applied in the field of clothing customization, they are mostly used in isolation, with problems such as poor accuracy and inaccurate prediction. This paper aims to develop an interactive clothing customization platform combining AR and AI. The research is carried out by constructing a body shape recognition model based on deep learning, a virtual clothing dynamic adaptation algorithm, a consumer preference prediction engine, and a real-time feedback and optimization suggestion system. The experimental results show that the body shape recognition accuracy of this platform is 92.3
Wearable bioelectronic systems are increasingly advancing toward continuous, non-invasive monitoring of physiological and biochemical signals; however, their widespread deployment is constrained by mechanical mismatch with soft tissues, limited long-term biocompatibility, and dependence on external power sources. In this context, silk fibroin has emerged as a promising biomaterial due to its tunable mechanical properties, biodegradability, optical transparency, and intrinsic compatibility with biological environments. Despite significant progress, current research remains fragmented across individual sensing domains, with a lack of a unified multiphysics integration framework that can effectively couple optical, thermal, electrical, and electrochemical transduction mechanisms within a single platform. This gap limits the rational design of fully integrated and energy-autonomous wearable systems. This review critically analyzes recent developments in SF-based wearable bioelectronics, focusing on material design, nanocomposite engineering, device architectures, and multiphysics sensing mechanisms involving quantum dots, phase change materials, and pyroelectric components. By systematically comparing existing material systems and device configurations, the review highlights how interfacial interactions and hierarchical structures govern overall device performance. Key findings indicate that while nanomaterial integration significantly enhances sensitivity, conductivity, and multifunctionality, critical challenges persist in long-term stability under physiological conditions, toxicity of certain nanomaterials (particularly heavy metal-based quantum dots), thermal management inefficiencies, and scalability of fabrication approaches. Moreover, multiphysics sensing systems often suffer from signal interference and lack standardized performance evaluation protocols, limiting cross-study comparability. Future research must therefore focus on developing scalable manufacturing strategies, ensuring long-term biocompatibility and environmental stability, establishing standardized benchmarking methods, and achieving real-time clinical validation. By bridging material-level innovations with system-level integration, this review provides a structured framework for advancing SF-based multiphysics bioelectronic platforms toward practical, self-powered, and clinically relevant wearable technologies.
This study presents a hybrid methodology integrating Lean Manufacturing tools with Design Thinking principles to address axis misalignment through constraint-based fixture design and statistical alignment modelling. Current State Mapping using Value Stream Mapping identified process inefficiencies and root causes. Lean tools (5S, Kaizen, Poka-Yoke, SMED, and PDCA) were combined with Design Thinking stages (Empathize, Define, Ideate, Prototype, and Test) to develop a geometrically constrained self-aligning fixture. Feature misalignment was quantified using best-fit statistical fitting of cylindrical primitives, while fixture positioning was modeled as a geometric constraint satisfaction problem. Prototypes were iteratively refined using operator feedback and validated under real production conditions. Results demonstrated significant improvements: assembly time decreased by 32
To meet consumers’ diverse demands for automotive design, this study proposes a research and design strategy based on online review data and Kansei Engineering (KE) theory. This strategy analyzes user review data to first extract KE intention needs and automotive design focus points from user reviews, then uses Social Network Analysis (SNA) theory and the Kano model to categorize KE vocabulary, and employs Quantitative Theory Type I (QTTI) to establish a mapping relationship between KE vocabulary and automotive design. Based on this mapping relationship, automotive designs that align with users' emotional intentions are created. Practical research was conducted on sedans, and the results showed that when users purchase sedans, they are more inclined to choose car designs with sporty, handsome, slick, and grand kansei intentions, among which they show the most significant preference for sports designs. The QTTI analysis indicates that the design factors C93 (55.745), C61 (15.125), C21 (15.418), and C82 (12.700) significantly impact the intention for sports designs. With the design goal of sports kansei imagery, nine car design schemes were designed and tested using the Semantic Differentiation (SD) method, and all schemes met the users’ expectations for sports kansei intentions. Therefore, this strategy can effectively provide a solid theoretical basis and support for subsequent car design.
Longitudinal Torsional Ultrasonic Grinding (LTUG) is an effective machining technique for reducing the grinding force of Ti-6Al-4V and improving surface integrity. However, existing mechanical modeling studies on LTUG of Ti-6Al-4V typically predict only the average grinding force and do not account for the stochastic distribution of abrasive grains or the contact–separation behavior induced by ultrasonic vibration. Under the background, the LTUG force model for Ti-6Al-4V was established by considering the random distribution of grains. Firstly, the net cutting time model of LTUG was established and applied to analyze the dynamic contact–separation characteristics between abrasive grains and the workpiece. Then, the wheel surface morphology model was built based on the random distribution of positions and geometries grain, and the criteria for distinguishing cutting stages during LTUG were investigated. Subsequently, the single-grain grinding force model under LTUG of titanium alloys was developed. By considering these factors, the LTUG force model was established. The results demonstrate that the proposed LTUG force model achieves high prediction accuracy, with errors ranging from 2.7–16.1
High-altitude regions of India viz., Leh and Ladakh present unique challenges of extreme weather, hilly terrain, rarefied atmosphere, and small construction period, resulting in high manpower wages and materials cost. Traditional construction practices include manual production of mud blocks offering sustainable construction practices. However, the manually cast blocks exhibit low strength, non-uniformity, and lower production. To overcome these limitations, the present study is dedicated to a modular mud block machine for production of superior quality stabilised compressed mud blocks for high-altitude regions, addressing the challenges of construction in remote, hilly, and rugged terrains. The machine employs a mechanised production system with 2 blocks per cycle and up to 800–1000 blocks per 8-hours shift production with prepared mix design. The machine is designed as portable and modular, operating without the need for electricity. With an overall weight of ∼ 120 kg, it can be disassembled into sub-assemblies weighing less than 25 kg each, ensuring ease of transport and handling as confirmed through extensive design analysis. The machine can be assembled and dismantled within 20 min for operation and requires minimal maintenance, making it ideal for construction in high-altitude regions. The developed solution addresses the limitations of low-quality, manually cast mud blocks by offering superior quality, mechanised, and higher production rate in challenging environments of high-altitude regions.
