摘要 : Traditional Simultaneous Localization and Mapping (SLAM) systems assume static environments, and their localization accuracy degrades significantly when encountering dynamic objects. Although dynamic SLAM based on semantic information can handle simple dynamic objects, it fails to address the impact of pseudo-dynamic objects and unrecognized dynamic objects on pose estimation. To address these issues, we propose a DC-SLAM system, comprehensively eliminating two types of dynamic features while preserving valid static features. The system implements a dual-category mechanism: for active dynamic features, it combines a dynamic object detection network with dense optical flow for detection and removal, preventing excessive elimination of static features' for passive dynamic features, the system first determines their attributes using multi-view geometry, then clusters all feature points based on depth information and object detection categories, and finally optimizes dynamic properties through Mahalanobis distance analysis of outlier similarity. This approach compensates for the limitations of multi-view geometry, enabling more effective suppression of passively dynamic objects. Additionally, an octree mapping module is developed to assist mobile robots in scene understanding for practical applications. Extensive experiments on the TUM dataset, Bonn dataset, and real-world dynamic scenes verify DC-SLAM's effectiveness. The results demonstrate significant improvements, compared with ORB-SLAM2, DC-SLAM reducing Absolute Trajectory Error (ATE) by 98.86 % and Relative Pose Error (RPE) by 97.28 %, while enabling the reliable construction of octree maps to enhance spatial understanding.

摘要 : Mass lumping of flexure beams plays a critical role in the dynamic compliance matrix method (DCM) for both kinetostatic and vibration analyses of compliant mechanisms and structures. The primary objective of this study is to derive and compare various mass matrix formulations for use in the DCM. Continuous and diagonal lumped mass matrices of flexure hinges and beams as well as their local-to-global coordinate transformation formulations are derived, discussed and clarified. By virtue of mass lumping and grounding, the schemes of compliance matrix summation for serial beam chains and stiffness/mass matrix summation for parallel branches are developed considering damping effects. The performance of different mass matrices is evaluated through case studies, which highlight the variation in performance with different mass matrices. The established mass matrix library of general flexure hinges and beams enhances the robustness of the DCM, and thus offers a straightforward performance-oriented analysis tool for compliant mechanisms and beam structures in terms of dynamic compliance.

摘要 : The growing demand for electrostatic capacitors in extreme conditions highlights the urgent need for polymer dielectric films with high breakdown strength ( E b ), high discharge energy density ( U e ), and outstanding high-temperature stability. Herein, a high-temperature stable capacitive composite film based on poly(vinylidene fluoride-co-chlorotrifluoroethylene) (P(VDF-CTFE)) is proposed by synergizing cross-linking and doping strategies. Specifically, P(VDF-CTFE) is engineered to form a cross-linking network and subsequently doped with surface-modified BNNs (BNNs-OH). By harnessing the synergistic effect between cross-linking and BNNs-OH doping, one can effectively restrict molecular mobility, disrupt the growth of crystalline domains, and inhibit the propagation of electrical trees and defects. This dual modification not only enhances the structural integrity of the polymer matrix but also improves its breakdown strength, high-temperature stability, and energy storage capabilities. The resultant composite film delivers a high discharge energy density up to 14.1 Jcm −3 at 25 °C and 13.59 Jcm −3 at 150 °C, validating its distinguished temperature stability over a wide temperature range. This study presents a facile strategy to develop advanced polymer dielectric films for harsh operating environments where both performance and durability are crucial.

摘要 : Functionally graded materials (FGMs) are widely used in high-end fields like aerospace and energy for their customizable gradient properties, yet accurate detection of subtle defects in their inhomogeneous structures remains a key challenge for conventional non-destructive testing (NDT) techniques. To address this, this study proposes a multi-task learning-based phased array ultrasonic testing (PAUT) system for FGM defect inspection, featuring a multi-task neural network integrating CNN, RNN, and ensemble learning, plus gradient-corrected acoustic modeling, multi-scale feature extraction, and 3D reconstruction. A physic dataset was built based on Ti6Al4V-ZrO 2 FGM acoustic properties, incorporating gradient-induced wave distortion and Gaussian/speckle synthetic noise. The system's CNN extracts B-scan spatial features and LSTM captures A-scan temporal dependencies, enabling synergistic defect localization and quantification via a combined loss function optimized by Pareto multi-objective strategy. Experimental results show high detection accuracy for different size defects. Transfer learning adapts it to Al 2 O 3 -Ni FGMs and trained/validated on 5 defect types with Bayesian uncertainty quantification ensuring reliability. This work provides a physics-informed solution for FGM inspection, overcoming single-modal NDT and homogeneous-material model limitations, and supports intelligent testing system generalization in FGM-based high-end manufacturing.

摘要 : Process planning of laser bending is crucial for precision manufacturing yet remains challenging for multi-path and multi-layer scanning due to complex thermo-mechanical interactions. This study presents an expert system with a framework integrating finite element simulation and machine learning to predict angular distortion in laser scanning processes. The methodology employs an artificial thermal strain method as a computationally efficient alternative to traditional thermo-mechanical finite element analysis for rapid training data generation. A joint prediction framework is proposed, consisting of a BP Prediction Model trained on ATS-generated data and an Error-Compensation Model calibrated against high-fidelity simulations. This framework achieves prediction accuracy within ± 4.33 % while significantly reducing computational costs compared to conventional methods. The developed expert system can output scanning strategies ranked according to recommendation rules based on the specified desired deformation and error tolerance and has been validated and analyzed. This research provides a practical solution for intelligent process planning in laser manufacturing applications, effectively balancing computational efficiency with prediction accuracy.

摘要 : To explore the mechanism of electric field on the interface evolution and heat transfer during bubble rising, the behavior of bubble-induced heat transfer under external electric field is investigated based on an incompressible smooth particle hydrodynamics − finite volume method (ISPH-FVM) coupling method. In the ISPH-FVM coupling method, ISPH particles are used to track the motion of bubble, while FVM grids are responsible for solving the coupling equation of electric field and fluid dynamics. By incorporating the electric field force as the source term into the momentum equation, the integration of the electric field model and the ISPH-FVM coupling method is realized. To verify the accuracy of the electric field force model in the present coupling method, firstly, the tensile deformation of the suspended bubble under vertical electric field is simulated. Then, the coupling method is employed to simulate the single thermal bubble rising under vertical and horizontal electric field, and the influence of different physical parameters on the thermal bubble rising is deeply analyzed. Finally, the influence of vertical electric field on the coaxial thermal bubble coalescence is investigated.

摘要 : For AlSi7Mg components produced by selective laser melting (SLM) and subsequent friction stir welding (FSW), the interaction of manufacturing defects from both processes is a critical factor governing fatigue failure. However, existing predictive models rarely integrate both additive manufacturing inherent defects and welding‐induced flaws. Furthermore, the formation and evolution mechanisms of defects arising from the coupled FSW-SLM process remain insufficiently characterized. This study addresses this gap by investigating the underlying mechanisms and developing an enhanced predictive model. X-ray computed tomography (CT) and detailed fractographic analysis were used to quantify, respectively, the global and critical defect populations and to correlate these features with fatigue performance across specimens processed under varied FSW parameters. Then SHapley Additive exPlanations (SHAP) values were employed based on a random forest model to rank the influence of individual defect attributes on fatigue life. To capture the combined morphological effects of multiple defects, an improved area parameter area mod was introduced. Finally, area mod within a modified El Haddad probabilistic framework was embedded to demonstrate its superior predictive accuracy for fatigue limits across diverse defect distributions.

摘要 : Janus fabrics with unidirectional water transport properties are widely applied in personal moisture management. Although existing Janus fabrics (e.g., Janus Cotton fabric) can transport sweat away from the skin side, the sweat often fails to evaporate and be removed timely, which inevitably leads to sweat accumulation and may cause discomfort. In this study, we propose a Janus Coolmax fabric (J-Coolmax) designed by asymmetric single-side spraying of a thermoplastic polyurethane (TPU) solution onto Coolmax fabric. As a result, the microstructures of Coolmax fibers not only enable J-Coolmax to achieve rapid unidirectional liquid transport (∼0.2 s), but also facilitate extensive sweat diffusion (three times the area of Janus cotton fabric (J-Cotton)) and rapid/efficient evaporation. Moreover, the skin temperature under our J-Coolmax is reduced by a sustained difference of 1.6 °C compared to that of J-Cotton in dynamic cooling process. In addition, the J-Coolmax demonstrates excellent durability, which can withstand 200 home washes and 400 abrasion cycles. This work improves the issues of limited evaporation efficiency and persistent sweat adhesion on the skin surface, offering a scalable strategy for next-generation Janus textiles.

摘要 : Transfer learning has become one of the most effective techniques to reduce the supervision cost of learning tasks, and has been applied in various domains. However, how to accurately and efficiently transfer knowledge between different domains is a challenging issue. Some previous pilot work can evaluate the transfer performance of different domains, but the practical performance of its application cannot be guaranteed when it is applied to the engineering domain. To tackle this issue, we focused on the task of surface defect detection in industrial engineering. The proposed method Defect-deepscore (D-deepscore) could quickly and accurately evaluate models from different source domains on a target domain, and then select the source domain model without any fine-tuning process. D-deepscore takes the parameters from deeper layers in the convolutional neural network, which are further processed by dimensionality reduction and information correlation analysis. In the experiments we demonstrate that finetuning the commonly used ImageNet pre-trained model is not necessarily the best choice and transfer learning from defect dataset will be more effective. Then, we evaluate the multiple pretrained models which were trained on multiple surface defect datasets, the results show that there is a strong correlation between D-deepscore's model evaluation scores and the classification accuracy. By comparing with existing SOTA (State-of-the-art) methods that focus on model transfer learning performance, D-deepscore improves the evaluation accuracy by 49.9 % over the best previous work. The proposed D-deepscore could provide a fast selection of the best pre-trained model for industrial defect detection tasks, which ultimately leads to improved detection performance.

摘要 : In this paper, we present an innovative method for the fabrication of multilayer microfluidic analytical devices (μPADs) in a single piece of paper. It involves hand-tearing a single piece of industrial oil filter paper (Sc-y102) into multiple layers. Subsequently, we employ a customized rubber stamp to print PDMS hydrophobic barriers on each layer, finally creating a functional and multilayered 3D microfluidic paper-based analytical device (3D μPAD). Parameter investigations showed an optimal resolution of 1200 μm for the hydrophilic channel and 840 μm for the hydrophobic barrier. Through using hand-tearing method, we can separate a single sheet of paper into up to six layers, each with a uniform thickness of ∼100 μm. Based on this method, we design and create a variety of single-layer 2D μPADs and multilayer 3D μPADs, and all derived from a single sheet of paper. Notably, the hydrophobic PDMS barriers, outperformed the conventional wax barriers in terms of chemical resistance, are able to effectively block the penetration of solvents and surfactants. Then, a 3D μPAD aimed for the timed detection of glucose is developed, which features an innovative multilayered design that precisely controls the timing of fluid flow to different detection zones. Finally, we designed a multilayer 3D μPAD that can achieve accurate and reliable blood typing by utilizing red blood cell agglutination and vertical flow filtration through single-sheet multiple microchannels. This hand-tearing approach expands the fabrication process for monolithic μPADs and provides a new way of realizing more complex and multifunctional device designs, may triggering more in-depth investigations into the monolithic devices and expanding research in the field.