Search Results (1,970)

Monocular Depth Estimation (MDE) remains a challenging problem due to texture ambiguity, occlusion, and scale variation in real-world scenes. While recent deep learning methods have made significant progress, maintaining structural consistency and robustness across diverse environments remains difficult. In this paper, we propose DAR-MDE, a novel framework that combines an autoencoder backbone with a Multi-Scale Feature Aggregation (MSFA) module and a Refining Attention Network (RAN). The MSFA module enables the model to capture geometric details across multiple resolutions, while the RAN enhances depth predictions by attending to structurally important regions guided by depth-feature similarity. We also introduce a multi-scale loss based on curvilinear saliency to improve edge-aware supervision and depth continuity. The proposed model achieves robust and accurate depth estimation across varying object scales, cluttered scenes, and weak-texture regions. We evaluated DAR-MDE on the NYU Depth v2, SUN RGB-D, and Make3D datasets, demonstrating competitive accuracy and real-time inference speeds (19 ms per image) without relying on auxiliary sensors. Our method achieves a $\delta$ < 1.25 accuracy of 87.25% and a relative error of 0.113 on NYU Depth v2, outperforming several recent state-of-the-art models. Our approach highlights the potential of lightweight RGB-only depth estimation models for real-world deployment in robotics and scene understanding. Full article

Five-axis machining centers are essential for manufacturing complex, high-precision parts. However, their accuracy is significantly affected by thermally induced geometric errors, also known as thermo-elastic errors. This paper presents a comprehensive approach to thermal characterization and its potential application in predictive modeling on a five-axis machine tool demonstrator, showcasing the capabilities of a novel dynamic R-test measurement method. Based on a previously developed and validated dynamic R-test measurement method that enables the rapid, volumetric acquisition of machine deviations during continuous movement, detailed experimental investigations were conducted under various single- and combined-axis loading scenarios. The extensive dataset and detailed error information provided by the dynamic R-test method enabled thorough analysis and correlation of thermo-elastic errors, including translational and rotational errors, with temperature and control-internal axis data. A well-established phenomenological model based on PT1 transfer functions is used, detailing its input variables and parameter determination methods. The model’s predictive capability was rigorously validated against independent datasets, demonstrating significant reductions in primary errors (up to 70% in maximum residual and 80% in RMSE). This study identifies the most influential error types and their correlation with thermal loads. This confirms the feasibility of robustly predicting thermo-elastic behavior and enhancing the volumetric accuracy of five-axis machine tools, particularly by leveraging the detailed error insights enabled by the dynamic R-test. Full article

To address the inadequate balance between exploration and exploitation of existing algorithms in complex solution spaces, this paper proposes a novel mathematical metaheuristic optimization algorithm—the Perpendicular Bisector Optimization Algorithm (PBOA). Inspired by the geometric symmetry of perpendicular bisectors (the endpoints of a line segment are symmetric about them), the algorithm designs differentiated convergence strategies. In the exploration phase, a slow convergence strategy is adopted (deliberately steering particles away from the optimal region defined by the perpendicular bisector) to expand the search space; in the exploitation phase, fast convergence refines the search process and improves accuracy. It selects 4 particles to construct line segments and perpendicular bisectors with the current particle, enhancing global exploration capability. The experimental results on 27 benchmark functions, compared with 15 state-of-the-art algorithms, show that the PBOA outperforms others in accuracy, stability, and efficiency. When applied to 5 engineering design problems, its fitness values are significantly lower. For H-type motion platforms, the PBOA-optimized platform not only achieves high unidirectional motion accuracy, but also the average synchronization error of the two Y-direction motion mechanisms reaches ±2.6 × 10⁻⁵ mm, with stable anti-interference performance. Full article

To address the issues of low efficiency, high cost of manual harvesting, and the lack of mechanized harvesting technology and equipment for high-stem Chrysanthemum coronarium, a self-propelled orderly harvester was designed to perform key harvesting operations such as row alignment, clamping and cutting, orderly conveying, and collection. Based on the analysis of agronomic requirements for cultivation and mechanized harvesting needs, the overall structure and working principle of the machine were described. Meanwhile, the key components such as the reciprocating cutting mechanism and orderly conveying mechanism were structurally designed and theoretically analyzed. The main structural and operating parameters of the harvester were determined based on the geometric and kinematic conditions of high-stem Chrysanthemum coronarium during its movement along the conveying path, as well as the mechanical model of the conveying process. In addition, a three-factor, three-level Box-Behnken field experiment was also conducted with the experimental factors including the machine’s forward, cutting, and conveying speed, and evaluation indicators like harvesting loss rate and orderliness. A second-order polynomial regression model was established to analyze the relationship between the evaluation indicators and the factors using the Design-Expert 13 software, which revealed the influence patterns of the machine’s forward speed, reciprocating cutter cutting speed, conveying device speed, and their interaction influence on the evaluation indicators. Moreover, the optimal parameter combination, obtained by solving the optimization model for harvesting loss rate and orderliness, was forward speed of 260 mm/s, cutting speed of 250 mm/s, and conveying speed of 300 mm/s. Field test results showed that the average harvesting loss rate of the prototype was 4.45% and the orderliness was 92.57%, with a relative error of less than 5% compared to the predicted values. The key components of the harvester operated stably, and the machine was capable of performing cutting, orderly conveying, and collection in a single pass. All performance indicators met the mechanized orderly harvesting requirements of high-stem Chrysanthemum coronarium. Full article

Cylindrical gears are used extensively due to their significant advantages including high efficiency, high load-bearing capacity, and long lifespan. However, the machining accuracy of cylindrical gears is significantly affected by motion errors and force-induced errors of machine tools. In this study, a motion error model of the machine tools was established based on multi-body system theory and homogeneous coordinate transformation method, quantifying the contributions and variation patterns of 12 key errors in the A and B-axes to workpiece geometric errors. Then, by using the stiffness analytical model and the spatial meshing theory, the influence of the force-induced elastic deformation of the shaft of rolling wheel and the springback of the workpiece tooth flank on the geometric error was revealed. Finally, taking the through rolling of a spur cylindrical gear with a module of 1.75 mm, a pressure angle of 20°, and 46 teeth as an example, the force-induced elastic deformation model of the shaft was verified by the rolling tests. Results show that for 40CrNiMo steel, the total profile deviation, total helix deviation, and single pitch deviation in the X-direction caused by rolling forces are 32.48 µm, 32.13 µm and 32.13 µm, respectively, with a maximum contact rebound is δ_c = 28.27. The relative error between theoretical and measured X-direction spindle deformation is 8.26%. This study provides theoretical foundation and experimental support for improving the precision of rolling process. Full article

Numerical simulation of metal forming processes is finding increasingly wide applications in advanced industry for the optimization of material processing conditions and prediction of process parameters, finally delivering a reduction of production costs. This work presents a comparison between simulation results of radial shear rolling (RSR) of VT3-1 titanium alloy (Ti-Al-Mo-Cr-Fe-Si) and results of experimental RSR at 1060 °C, 980 °C, and 900 °C in one, three, and five passes, respectively. The digital model (DM) demonstrates a high convergence of the calculation results (calculation error of less than 5%) with the actual geometric parameters of the experimental bars, their surface temperature, and rolling time during the experiment, which indicates a good potential for its application in the selection of deformation modes. Based on the simulation and experimental data, the conditions providing for the formation of differently sized grains in the bar cross-section have been identified. All of the as-rolled bars exhibit a gradient distribution of macrostructure grain size number (GSN), from the smallest one at the bar surface (2–4) to the greatest one in the center (4–6). The macrostructure GSN correlates with the workpiece temperature, which is the highest in the axial zone of the bars, and with the experimentally observed high plastic strain figures in the surface layers. It was found that, depending on the temperature conditions and reduction ratio per pass, any minor change in the values of process parameters can lead to the formation of macrostructures with different grain size numbers. Full article

Milepost-assisted positioning estimates the distance between a vehicle-mounted camera and a milepost as a reference position for autonomous driving. However, the accuracy of monocular metric depth estimation is compromised by camera installation angle, milepost inclination, and image occlusions. To solve the problems, this paper proposes a two-stage monocular metric depth estimation with boundary calibration and geometric optimization. In the first stage, the method detects a milepost in one frame of a video and computes a metric depth map of the milepost region by a monocular depth estimation model. In the second stage, in order to mitigate the effects of road surface undulation and occlusion, we propose geometric optimization with road plane fitting and a multi-frame fusion strategy. An experiment using pairwise images and depth measurement demonstrates that the proposed method exceeds other state-of-the-art methods with an absolute relative error of 0.055 and root mean square error of 3.421. Full article

The abnormal tension change in one cable in a cable-stayed bridge indicates cable damage, so it is necessary to obtain the benchmark of the cable tension. After establishing the regression model of the mapping between the temperature-induced cable tension and the bridge temperature field or other data, the regression value can be used as the benchmark. To improve the regression model, the geometric compatibility and mechanical equilibrium must be jointly considered. Therefore, two data groups, which contain the bridge temperature field and the regression values of the temperature-induced deflection of the main girder, are input into the deep learning neural networks. Time lags exist between the temperature features and the temperature-induced cable tension, but are not significant between the temperature-induced deflection and tension. So one neural network module, which receiving the regression values of the temperature-induced deflection, is composed of Convolutional Neural Networks (CNNs). The other neural network module, which receives the temperature features, is composed of stacked CNN and Long Short-Term Memory (LSTM). Finally, several convolution kernels will integrate the array output from the two modules into one regression value of the temperature-induced cable tension. By combining the input data and the composite neural networks, the R² of the regression models of the temperature-induced cable tension is more than 0.95, and the error of the regression values is less than 0.3 kN. In the future, if the nonlinearity at the curve inflection point and the complexity in data distribution could be solved, the stability of the model may be improved. Full article

In Global Navigation Satellite System (GNSS)-denied environments, navigation using signals of opportunity (SOP) from Low Earth Orbit (LEO) satellites is considered a feasible alternative. Compared with single-constellation systems, multiple-constellation LEO systems offer improved satellite visibility and geometric diversity, which enhances positioning continuity and accuracy. To allocate weights among heterogeneous observations, prior studies have employed the Helmert variance component estimation (HVCE) method, which iteratively determines relative weight ratios of different observation types through posterior variance estimation. HVCE enables error modeling and weight adjustment without prior noise information but is highly sensitive to outliers, making it vulnerable to their impact. This study proposes a Robust HVCE-based dual-constellation weighted positioning method. The approach integrates prior weighting based on satellite elevation, observation screening based on characteristic slopes, HVCE, and IGG-III robust estimation to achieve dynamic weight adjustment and suppress outliers. Experimental results over a 33.9 km baseline demonstrate that the proposed method attains Two-Dimensional (2D) and Three-Dimensional (3D) positioning accuracies of 12.824 m and 23.230 m, corresponding to improvements of 29% and 16% over conventional HVCE weighting, respectively. It also outperforms single-constellation positioning and equal-weighted fusion, confirming the effectiveness of the proposed approach. Full article

The precise identification, classification, sorting, and rapid and accurate quality assessment of soybean seeds are extremely important in terms of the continuity of agricultural production, varietal purity, seed processing, protein extraction, and food safety. Currently, commonly used methods for the identification and quality assessment of soybean seeds include morphological analysis, chemical analysis, protein electrophoresis, liquid chromatography, spectral analysis, and image analysis. The use of image analysis and artificial intelligence is the aim of the presented research, in which a method for the automatic classification of soybean varieties, the assessment of the degree of damage, and the identification of geometric features of soybean seeds based on numerical models obtained using a 3D scanner has been proposed. Unlike traditional two-dimensional images, which only represent height and width, 3D imaging adds a third dimension, allowing for a more realistic representation of the shape of the seeds. The research was conducted on soybean seeds with a moisture content of 13%, and the seeds were stored in a room with a temperature of 20–23 °C and air humidity of 60%. Individual soybean seeds were scanned to create 3D models, allowing for the measurement of their geometric parameters, assessment of texture, evaluation of damage, and identification of characteristic varietal features. The developed 3D-CNN network model comprised an architecture consisting of an input layer, three hidden layers, and one output layer with a single neuron. The aim of the conducted research is to design a new, three-dimensional 3D-CNN architecture, the main task of which is the classification of soybean seeds. For the purposes of network analysis and testing, 22 input criteria were defined, with a hierarchy of their importance. The training, testing, and validation database of the SB3D-NET network consisted of 3D models obtained as a result of scanning individual soybean seeds, 100 for each variety. The accuracy of the training process of the proposed SB3D-NET model for the qualitative classification of 3D models of soybean seeds, based on the adopted criteria, was 95.54%, and the accuracy of its validation was 90.74%. The relative loss value during the training process of the SB3D-NET model was 18.53%, and during its validation process, it was 37.76%. The proposed SB3D-NET neural network model for all twenty-two criteria achieves values of global error (GE) of prediction and classification of seeds at the level of 0.0992. Full article

Due to the rapid advancements in coordinate measuring systems, data processing software, and additive manufacturing (AM) techniques, it has become possible to create copies of existing models through the reverse engineering (RE) process. However, the lack of precise estimates regarding the accuracy of the RE process—particularly at the measurement, reconstruction, and computer-aided design (CAD) modeling stages—poses significant challenges. Additionally, the assessment of dimensional and geometrical errors during the manufacturing stage using AM techniques limits the practical implementation of product replicas in the industry. This paper provides an estimation of the errors encountered in the RE process and the AM stage of various models. It includes examples of an electrical box, a lampshade for a standing lamp, a cover for a vacuum unit, and a battery cover. The geometry of these models was measured using a GOM Scan 1 (Carl Zeiss AG, Jena, Germany). Following the measurement process, data processing was performed, along with CAD modeling, which involved primitive detection, profile extraction, and auto-surface methods using Siemens NX 2406 software (Siemens Digital Industries, Plano, TX, USA). The models were produced using a Fortus 360-mc 3D printer (Stratasys, Eden Prairie, MN, USA) with ABS-M30 material. After fabrication, the models were scanned using a GOM Scan 1 scanner to identify any manufacturing errors. The research findings indicated that overall, 95% of the points representing reconstruction errors are within the maximum deviation range of ±0.6 mm to ±1 mm. The highest errors in CAD modeling were attributed to the auto-surfacing method, overall, 95% of the points are within the average range of ±0.9 mm. In contrast, the lowest errors occurred with the detect primitives method, averaging ±0.6 mm. Overall, 95% of the points representing the surface of a model made using the additive manufacturing technology fall within the deviation range ±0.2 mm on average. The findings provide crucial insights for designers utilizing RE and AM techniques in creating functional model replicas. Full article

This paper presents a computational design method for generating discrete shell structures using sets of equivalent discrete members. This study addresses the challenge of reducing the geometrical variety in discrete shell elements by introducing a method to design and optimize constituent members considering their similarity, approximation of the double-curved architectural surface, and buildability. First, we employed a relaxation-based computational form-finding method to generate a discrete topology with planar quad faces and an approximated smooth, double-curved surface. Then, we perform clustering and optimization based on face similarities concerning the minimization of deviations from the smooth surface approximation, and the dihedral angle between the planes of neighboring elements and their optimal intersection plane. The proposed approach can reduce the geometrical differences in discrete shell elements while satisfying the user-defined error threshold. We demonstrated the viability of our method on various structured topologies with different boundary conditions, support settings, and total face counts, while explicitly controlling inter-element facing angles for assembly ready contacts. This enables mold-based prefabrication with repeatable components. Full article

The on-orbit geometric calibration accuracy of high-resolution space cameras directly affects the application value of Earth observation data. Conventional on-orbit geometric calibration methods primarily rely on ground calibration fields, making it difficult to simultaneously achieve high precision and real-time monitoring. To address this limitation, we, in collaboration with Tsinghua University, propose a high-precision, real-time, on-orbit geometric calibration system based on active optical monitoring. The proposed system employs reference lasers to integrate the space camera and the star tracker into a unified optical system, enabling real-time monitoring and correction of the camera’s exterior orientation parameters. However, during on-orbit operation, the space camera is subjected to a complex thermal environment, which induces thermal deformation of optical elements and their supporting structures, thereby degrading the measurement accuracy of the geometric calibration system. To address this issue, this article analyzes the impact of temperature fluctuations on the focal plane, the reference laser unit, and the laser relay folding unit and proposes athermalization design optimization schemes. Through the implementation of a thermal-compensated design for the collimation optical system, the pointing stability and divergence angle control of the reference laser are effectively enhanced. To address the thermal sensitivity of the laser relay folding unit, a right-angle cone mirror scheme is proposed, and its structural materials are optimized through thermo–mechanical–optical coupling analysis. Finite element analysis is conducted to evaluate the thermal stability of the on-orbit geometric calibration system, and the impact of temperature variations on measurement accuracy is quantified using an optical error assessment method. The results show that, under temperature fluctuations of 5 °C for the focal plane and the reference laser unit, 1 °C for the laser relay folding unit, and 2 °C for the star tracker, the maximum deviation of the system’s measurement reference does not exceed 0.57″ (3σ). This enables long-term, stable, high-precision monitoring of exterior orientation parameter variations and improves image positioning accuracy. Full article

The turning process is the main machining task in brake disc production, and the PCBN tool is the most suitable type of cutting tools in the machining of brake discs made of cast iron. The edge geometric optimization of the PCBN tool is the key factor to obtain a better tool performance. In this paper, the cutting simulation for the machining of cast iron with PCBN tool of grade HNMN120712 was established, which exhibits a simulation error lower than 10.8%. The optimal turning parameters were obtained by the equal material removal rate method. The edge geometric parameters were optimized in two stages: firstly, the optimal edge radius was obtained as 30 μm by the comprehensive normalization analysis of the cutting temperature and stress, and then, the chamfer width and angle were further optimized to 0.1 mm and 15°. At finally, the optimized PCBN tool was prepared and tested in the machining of brake discs; the results indicate that the designed tool exhibits an excellent tool performance with 3.4 times the tool life of the conventional tool. Full article

Digital ink Chinese characters (DICCs) written by international students often contain various errors and irregularities, making the recognition of these characters a highly challenging pattern recognition problem. This paper designs a graph convolutional network with agent attention (GCNAA) for recognizing DICCs written by international students. Each sampling point is treated as a vertex in a graph, with connections between adjacent sampling points within the same stroke serving as edges to create a Chinese character graph structure. The GCNAA is used to process the data of the Chinese character graph structure, implemented by stacking Block modules. In each Block module, the graph agent attention module not only models the global context between graph nodes but also reduces computational complexity, shortens training time, and accelerates inference speed. The graph convolution block module models the local adjacency structure of the graph by aggregating local geometric information from neighboring nodes, while graph pooling is employed to learn multi-resolution features. Finally, the Softmax function is used to generate prediction results. Experiments conducted on public datasets such as CASIA-OLWHDB1.0-1.2, SCUT-COUCH2009 GB1&GB2, and HIT-OR3C-ONLINE demonstrate that the GCNAA performs well even on large-category datasets, showing strong generalization ability and robustness. The recognition accuracy for DICCs written by international students reaches 98.7%. Accurate and efficient handwritten Chinese character recognition technology can provide a solid technical foundation for computer-assisted Chinese character writing for international students, thereby promoting the development of international Chinese character education. Full article