Search Results (350)

We present a reward-free control framework for balancing and steering a two-wheeled scooter using a diffusion-based world model. Rather than engineering a reward, we specify goals directly in observation space: target values (e.g., zero roll and zero yaw error) are pinned through a continuous mask, and classifier-free guidance amplifies the goal signal during trajectory generation. Because the mask is continuous at inference, goals can be traded off online (for instance, relaxing the balance constraint during sharp turns to allow necessary leaning) without retraining. The model is a FiLM-Mixer denoising network trained with V-prediction diffusion. At deployment, the controller runs in real time using a single diffusion step with warm-started predictions. We validate the approach on a full-sized Thormang3 humanoid operating a Gogoro Viva scooter in simulation, and deploy it on physical hardware. It matches a PPO baseline tuned with six reward components on balance, survival, and heading tracking while producing smoother commands, all without the per-task reward-shaping step. Diffusion training introduces its own loss-weight hyperparameters; unlike reward weights, however, these are task-agnostic. They govern the denoising procedure rather than the desired behavior, and are therefore set once and reused unchanged across goals rather than re-tuned for each new task. Because the model learns to predict trajectories rather than to maximize a reward, its training signal depends only on observed states and actions, not on reward labels. Real hardware recordings can therefore be folded directly into the same loss, providing a route toward closing the sim-to-real gap that reward-based methods such as PPO structurally cannot use. Full article

To address the oversteering risk during aircraft ground towing with a towbarless towing vehicle, this study proposes a dual-LiDAR point-cloud detection and pose estimation method for aircraft rear-wheel targets. First, a complementary dual-LiDAR acquisition strategy is adopted to reduce rear-wheel point-cloud occlusion caused by the aircraft nose landing gear and towing mechanism. Second, considering the small size and distinctive local geometry of rear-wheel targets, vertical density enhanced encoding and a lightweight CNN-Transformer BEV backbone are introduced into the PointPillars framework. The vertical density enhanced encoding explicitly describes the normalized height-wise distribution of valid points within each pillar, thereby improving the representation of cylindrical wheel structures. The CNN-Transformer BEV backbone incorporates a window-based self-attention Transformer module into deep features to strengthen local contextual modeling in the BEV space. Based on the detected coordinates of the left and right rear wheels, the aircraft fuselage pose is then estimated in combination with the TLTV coordinate system. In three-seed experiments on the fixed validation split, the Full model achieves an mAP@0.5 of $0.8788\pm 0.0161$, which is 8.50 percentage points higher than the original PointPillars baseline. The model contains 4.1069 M parameters and runs at 38.0732 FPS. The towing-angle estimation error remains within the allowable engineering range. These results show that the task-specific adaptations improve rear-wheel detection while retaining a compact model and real-time processing capability. Full article

Practical training on embedded systems is a very good way to assess the skills of students in firmware development, real-time programming and hardware–software co-design. However, many low- and middle-income institutions still use closed, proprietary boards, such as the Altera DE series, which hide the physical behavior. To address this, a low-cost, fully reconfigurable Microcontroller Test Board (MTB) was designed and technically validated for application in undergraduate laboratories. The MTB is manufactured as a one-sided through-hole PCB via toner transfer and FeCl3 etching, with a PIC18F2550 in-circuit programmer and support for 28- and 40-pin targets including the PIC16F877A. The MTB integrates six functional blocks, providing five canonical input/output modalities on a single regulated 5 V/0.56 A rail: (1) an ICSP programmer; (2) a multiplexed seven-segment display; (3) an 8 × 8 LED matrix; (4) a matrix keypad with LCD calculator; and (5) a 10-bit ADC stage. All subsystems operate with a flicker-free refresh rate > 122 Hz. The total board power consumption is 6.72 W at 12 V DC, with the linear regulator dissipating 3.92 W under full load. Probe points allow for signal transparency. Models for regulator dissipation, ADC quantization, multiplex refresh duty cycle, LED current limiting, oscillator timing, matrix-scanning latency, LCD timing, keypad response and Fe³⁺ with Cu etch kinetics were first-principles. Bench measurements agree with model predictions to less than 3% error, and confirm worst-case design margins. The MTB provides analytical rigor with accessible hardware to enable measurable experiments in microcontroller labs. The platform is cost-effective, with a material cost of less than USD 18. Each signal is routed to probe points, providing significant cost reduction from commercial trainers, and a transparent hands-on laboratory resource for engineering curricula with limited resources. Full article

Food waste represents a significant environmental and economic challenge, partly due to inefficient portion management during preparation, storage, and consumption. One strategy to mitigate this issue is to use single-dose (monodose) packaging, which allows precise dosing and encourages complete consumption, thereby minimizing leftovers. This work presents the design, construction, and experimental validation of an automated machine engineered for single-dose packaging of low-viscosity sauces. The system is capable of filling, sealing, and cutting off a continuous plastic film to produce hermetically sealed 60 mL pouches. The system contains a deposit equipped with an ultrasonic level sensor, a pneumatic dispenser mechanism, a PI-based temperature controller for the sealing and cutting processes, and a Human–Machine Interface (HMI) for system configuration and process monitoring. The general controller was implemented using a Programmable Logic Controller (PLC) and coordinated the stages of pouch filling, sealing, cutting, and counting. Experimental tests demonstrated stable and repeatable volumetric dosing. Mean dosing errors remained below ±1 mL for water and tomato-based sauce, whereas tomatillo-based and Chilean sauce exhibited higher deviations of +2.10 mL and +3.25 mL, respectively, reflecting the influence of viscosity and product heterogeneity. In addition, the system achieved hermetic sealing in more than 95% of samples and stable operation during continuous cycles. Additionally, the system showed thermal stability under PI control and high precision in the deposit level. Full article

With the increasing construction of pile foundations on mountain slopes, the asymmetric stress field induced by slopes significantly reduces the lateral soil resistance of piles. Most existing studies focus on piles located at slope crests, and there is an obvious gap in p-y curve calculation methods for piles embedded in clay slopes. In this study, model tests of single piles in clay slopes with slope angles of 0°, 15°, and 30° are carried out. The test results reveal that the increase in slope angle aggravates the horizontal displacement and bending moment of piles and moves the pile rotation point downward. Numerical model is adopted to extract p-y curves at various embedded depths and slope angles. It is found that rising slope angles continuously degrade the initial stiffness and ultimate soil resistance of p-y curves, and such degradation weakens with increasing embedment depth. The core innovation of this paper is establishment of a modified p-y curve model considering slope effects. A simplified analytical method for slope piles is further proposed based on the load transfer method. The prediction errors of this modified model are within the allowable engineering range, which can provide a reliable analytical tool for calculating the lateral bearing capacity of pile foundations in clay slopes. Full article

Submarine sand waves are widespread on shallow continental shelves. Their complex morphology and potential mobility create challenges for engineering surveys, navigation safety, and seabed stability assessment. Multibeam surveys provide accurate bathymetry but are costly and spatially limited, whereas satellite-based methods offer broader coverage but remain challenging in complex sand wave fields. Here, we propose a spatial-sequential 2DCNN–LSTM model for retrieving submarine sand wave bathymetry from Sentinel-2 surface reflectance imagery. The model represents each target point as a sequence of local multispectral image patches, allowing convolutional layers to extract two-dimensional textural features and LSTM layers to learn profile-scale rhythmic continuity associated with sand wave morphology. The model was trained using multibeam bathymetry and applied to a large extrapolation area of approximately 4000 km² on the Taiwan Banks. Evaluation on the large extrapolated area against in situ bathymetric data achieved a root mean square error (RMSE) of 3.78 m, a mean absolute error (MAE) of 2.99 m, and a mean relative error (MRE) of 9.1%. The results demonstrate that sand wave-induced optical textures can provide useful information for broad-scale bathymetric reconstruction, although model performance remains dependent on image texture visibility controlled by hydrodynamic, illumination, and atmospheric conditions. This framework offers a cost-effective approach for satellite-based monitoring of large submarine sand wave fields, providing a new perspective for engineering applications. Full article

Maintaining the geometric quality of railway tracks is essential for safe and efficient operations. This study examines the prediction of the Track Quality Index (TQI) using Multiple Linear Regression (MLR) and Random Forest (RF) models, based on inspection data from the mountainous Kolashin–Podgorica railway section. Data collected from 2017 to 2022 were used for model development, while 2024 data served as an independent validation set. The primary contribution is a high-resolution analysis of 20-m homogeneous track units across multiple years, combined with independent validation. The dataset includes precise measurements segmented into 20-m units, covering infrastructure, geometric, operational, and maintenance variables. Model variables were selected for their relevance to track quality and data completeness. Both models were trained on scaled input data and evaluated using the coefficient of determination (R²), mean absolute error (MAE), and root mean square error (RMSE). The RF model outperformed MLR, achieving a higher R² (0.69 versus 0.57) and an MAE approximately 15% lower. Furthermore, the RF model identified severe localized degradation trends and demonstrated robust screening capability, achieving 86.4% sensitivity in detecting high-risk track segments requiring urgent intervention. These findings highlight the effectiveness of ensemble machine learning methods in reducing prediction errors and enabling proactive, data-driven track maintenance in complex railway networks. This methodology allows railway engineers to identify segments most susceptible to rapid deterioration, supporting more precise scheduling of tamping, renewals, or other targeted interventions. Model outputs can assist managers in prioritizing maintenance activities, optimizing resource allocation, and minimizing unexpected failures, thereby enhancing safety and cost efficiency in daily operations. However, the study is limited by its focus on a single railway corridor and a relatively short observation period with few maintenance events, which may restrict the generalizability of the findings to other lines or maintenance regimes. Full article

This study proposes a sequential Hybrid GP-ANN architecture based on the concept of autonomous feature engineering for multi-output performance prediction of single-phase induction motors, serving as a high-fidelity surrogate model ready for digital twin integration. The required high-resolution motor dataset (88,375 samples) was generated using a parametric sweep approach with Ansys RMxprt. In the proposed architecture, Genetic Programming (GP) is not positioned as a final predictor but as an analytical filter that discovers hidden physical relationships in raw input data and converts them into super features, thereby structurally eliminating the polynomial explosion risk and scale sensitivity that arise when raw data are modelled directly. The nonlinear physical relationships discovered autonomously by GP are added to the network input matrix as mathematical vectors, breaking the black-box structure of standard artificial neural networks and transforming it into a physics-inspired grey-box model. The nonlinear terms discovered by GP are fed into the network after independent Z-score normalisation to preserve gradient stability. To comprehensively evaluate this framework, its predictive performance is benchmarked against industry-standard machine learning algorithms, including Random Forest (RF), Support Vector Regression (SVR), and Extreme Gradient Boosting (XGBoost). Results on variation-based unseen test data show that the hybrid model achieves competitive prediction accuracy relative to XGBoost—matching or exceeding it for Output Torque and Input Current, and remaining broadly comparable for Output Power, though XGBoost achieves substantially lower RMSE for Efficiency and Power Factor—while additionally offering transparent mathematical traceability, reducing error rates (RMSE) by 42.1% to 66.1% per parameter compared to the standard ANN. The model achieves an $R^2$ score of 0.9879 for the power factor parameter, where conventional approaches struggle. To rigorously validate the interpretability of this grey-box architecture, a global Permutation Feature Importance (PFI) analysis was conducted, showing that the GP-derived super features collectively account for 61.46% of the decision logic, outweighing the combined contribution of the raw inputs (38.54%). Furthermore, the autonomous feature engineering layer is found to reduce the learning burden on the ANN, allowing it to converge to a substantially lower error floor within the same fixed epoch budget. With an online inference time below 0.02 ms, the proposed architecture offers a robust methodological infrastructure for sustainable motor digital twins through a balanced trade-off between prediction accuracy and computational efficiency. Full article

This study addresses the inverse design of carbon fibre-reinforced polymer laminated plates with prescribed natural frequencies. The problem is difficult because stacking sequences are discrete, the design space is large, and multiple layups may produce nearly identical frequency spectra. This study does not seek to introduce a new tandem-network architecture. Rather, it adapts the established tandem inverse-design strategy to the discrete and non-unique vibration design of carbon fibre-reinforced polymer laminated plates. In the proposed framework, a trainable inverse network is coupled to a pre-trained forward frequency surrogate, allowing the inverse model to be optimised through frequency reconstruction instead of direct ply-angle supervision. A dataset of 50,000 symmetric CFRP laminates is generated using Classical Laminate Theory and a Rayleigh–Ritz vibration solver, covering four boundary conditions and a range of plate geometries. The forward model achieves R² values above 0.99 and mean absolute percentage errors below 3% for the first five natural frequencies. Compared with a genetic algorithm, the proposed inverse model provides stacking sequences about 7000 times faster while producing multiple feasible designs for each target. Permutation sensitivity analysis shows that plate geometry has the strongest influence on the frequency response, followed by boundary condition and ply orientation. Four engineering cases confirm the method’s usefulness for vibration isolation, frequency-gap control, and multi-mode frequency prescription. The principal contribution is the integration of multi-boundary-condition vibration modelling, discrete stacking-sequence inverse design, response-based treatment of non-uniqueness, speed/diversity benchmarking, and sensitivity-based physical interpretation within a single composite-laminate design framework. Full article

This study introduces the ’Black-Backed Jackal Optimization’ (BBJO), a nature-inspired meta-heuristic algorithm designed for complex, non-linear, and high-dimensional search spaces. The fundamental mathematical model of BBJO relies on the opportunistic hunting behavior and survivability strategies of the black-backed jackal (Lupulella mesomelas). We use non-linear energy decrease and adaptive Lévy flight to maintain the equilibrium of the search. This allows the algorithm to scan large areas first, then zoom in with a high degree of precision once it has identified a suitable location. This configuration prevents the algorithm from getting stuck on a suboptimal local solution, which is a frequent danger during searches in complex spaces. BBJO has been validated against 23 standard benchmark functions, demonstrating significantly greater accuracy than Particle Swarm Optimization (PSO) on complex and large-scale search spaces. On fixed-size domains ($F_{21}$–$F_{23}$), the BBJO algorithm achieved a 100% success rate with zero standard deviation, surpassing the Grey Wolf Optimizer (GWO) and Differential Evolution (DE), which frequently suffered from structural stagnation. Visual convergence study shows that BBJO efficiently identifies optimal search regions early in the iteration budget, saving time compared to traditional linear decay models. BBJO optimizes fuzzy inference systems (FISs) for two practical applications: autonomous car speed control and industrial furnace regulation. Experimental results indicate that BBJO significantly decreased cumulative penalties and improved steady-state error reduction compared to baseline configurations and established meta-heuristic methods. The results show that BBJO is a reliable and useful technique for engineering optimization. Full article

A graph attention network-enhanced multi-agent proximal policy optimization (GAT-MAPPO) framework is proposed for cooperative guidance in adversarial engagement scenarios. A dynamic heterogeneous interaction graph is formulated over interceptors and targets at every decision epoch. Through a multi-head graph attention encoder, relational features capturing both inter-interceptor cooperation and target threat dynamics are adaptively aggregated. These graph-enriched observations are processed by a Centralized-Training, Decentralized-Execution (CTDE) MAPPO architecture, guided by a hierarchical reward function that mandates miss distance minimization, simultaneity of arrival consensus, multi-directional encirclement, and smooth control effort. Furthermore, the integration of a three-stage curriculum learning strategy allows for robust cooperative policy derivation across transitions from rectilinear to highly adaptive evasion patterns, eliminating the need for explicit rule engineering. Extensive Monte Carlo simulations confirm GAT-MAPPO’s superior performance: achieving $>95\%$ interception success rate in 4-vs.-4 scenarios and reducing mean simultaneity error by $41.4\%$ compared to the MAPPO baseline. Comprehensive ablation and sensitivity studies validate the critical roles played by graph attention encoding, reward hierarchy design, and progressive curriculum staging. Full article

Shared bikes are frequently parked in disorder, resulting in low efficiency of manual collection and transfer and heavy workload for maintenance staff. Random parking across various areas forces shared bikes to occupy sidewalks and fire exits, damaging urban landscapes and disrupting traffic order. To tackle these industrial pain points, this paper develops an integrated intelligent robot system equipped with functions of multi-pose grasping, automatic transfer and fixed-point delivery of shared bikes, which can effectively address the drawbacks of low efficiency and high labor costs in traditional manual maintenance. This paper focuses on the completion of the robot’s overall mechanical structure design, stiffness–precision collaborative optimization model construction, finite-element static simulation verification, 1:7 scaled prototype development and performance testing. Firstly, the overall layout design of the multi-posture adaptive floating clamping mechanism, transfer-bearing frame, and Mecanum wheel omnidirectional mobile chassis is completed, and the structural parameters and assembly benchmarks of the core components are clarified. Secondly, a stiffness–precision coupling optimization model is established, and the static analysis under extreme load conditions is carried out through Abaqus finite-element software, which verifies the rationality of 45# carbon steel material selection and the safety of structural strength. Subsequently, a 1:7 scaled principle prototype is developed, and repetitive grabbing and transfer tests are carried out to verify the system operation feasibility, stability and grabbing accuracy. Finally, the statistical analysis of the test data and the horizontal comparison of similar schemes are completed. The test and simulation results show that the maximum stress of the system under extreme working conditions is 131.21 MPa, which is far lower than the allowable stress of 355 MPa of 45# steel, and the safety factor reaches 2.71. The maximum total deformation is 4.0552 mm, which is concentrated at the end of the front-end clamping mechanism, and is within the allowable stiffness deviation range of the transfer system. The average value of the single clamping positioning error of the scaled prototype is 0.476 mm, with a 95% confidence interval of 0.457–0.495 mm, which is converted to a positioning error of ≤3.4 mm for the full-scale prototype, which is far better than similar industry solutions. The average time of a single complete grabbing and transfer operation is 12.38 s, which is more than 45% higher than the traditional manual mode. The structural design, grabbing accuracy and operation stability of the robot designed in this paper all meet the requirements of actual working conditions of urban sidewalks, which can effectively reduce the intensity of manual labor and improve the operation and maintenance efficiency of shared bicycles. It has strong engineering application value and can provide reference for the design and manufacturing of intelligent collection and transfer systems for shared two-wheelers. Full article

In this paper, a machine learning model to predict the Young’s modulus of polylactic acid specimens manufactured by Fused Deposition Modeling is proposed, based on a stacked ensemble architecture. The model uses as input parameters the fill degree, printing speed, filling pattern, yield strength, and tensile strength, along with additional features obtained through feature engineering. The proposed approach integrates nine base models with a linear meta-model, allowing it to capture both linear and nonlinear relationships between the variables. The results obtained on the test dataset show strong predictive performance, with a Mean Squared Error with a value of 7.31 together with a Coefficient of Determination R² with a value of 0.99, which is noticeably better than the performance of the individual models. To validate the model, a separate group of specimens was tested, and the difference between the measured and predicted Young’s modulus was about 1% on average. The model was also implemented in a desktop application with a graphical interface, in which the calculation can be run directly, thus allowing a rapid estimation of Young’s modulus. In this way, the need for laborious experimental testing is reduced with the help of AI-based approaches in additive manufacturing. Full article

In recent decades, numerical methods have become indispensable tools for solving complex problems in science and engineering. Among these, the finite element method (FEM) is widely recognized as a powerful computational approach. However, traditional FEM has significant limitations when modeling discontinuities such as cracks, holes, or material interfaces, particularly in functionally graded materials (FGMs). To address these challenges, this study proposes an advanced framework that integrates the Extended Finite Element Method (XFEM) with Isogeometric Analysis (XIGA), referred to as XFEM–XIGA, to simulate stress concentration factors (SCFs) around circular holes in both isotropic and FGM plates. The proposed methodology employs the level-set method to represent discontinuous boundaries and incorporates appropriate enrichment functions into the displacement field, allowing accurate modeling of stress concentrations without the need for remeshing. MATLAB codes were developed to implement this integration, providing a flexible computational platform for practical engineering applications. The performance of the proposed XFEM–XIGA approach was evaluated using several benchmark problems, including isotropic plates with circular holes near material boundaries and FGMs subjected to uniaxial tensile loading. The results obtained from the XFEM–XIGA model were compared with analytical solutions, standard FEM results, and available experimental data. For isotropic plates, the XFEM–XIGA model achieved a stress concentration error of 1.71%. For FGM plates with cracks or circular holes, the error was 2.55% compared with exact solutions. These findings demonstrate the robustness and accuracy of the integrated method in handling complex geometries and heterogeneous material properties. Overall, this study shows that the combination of XFEM and XIGA offers an efficient and reliable tool for analyzing stress concentration factors in FGM structures. The proposed approach provides improved modeling capabilities for industrial components where stress concentrations at material boundaries are critical to structural integrity and performance. Full article

Fracture orientation data in structural geology are commonly affected by non-negligible angular uncertainty, which can significantly impact the reliability of classification and interpretation of deformation patterns. In this work, we address the problem of discriminating between two groups of directional observations. To account for measurement uncertainty inherent in field data, we adopt a deconvolution-based circular kernel discriminant rule specifically designed for noisy angular observations. This approach explicitly incorporates the measurement-error mechanism into the estimation process, allowing for more robust classification in the presence of observational noise. The methodology is applied to measurements arising in structural geology, where the discrimination of fracture orientations is relevant to the interpretation of deformation patterns and to applications in rock engineering. Specifically, we consider two datasets from Ordovician turbidites, involving different types of orientation data. The first dataset consists of $L_0^1$ axes, representing linear features described by Plunge–Azimuth coordinates, while the second dataset concerns axial-plane cleavage surfaces, expressed in terms of Dip and Dip direction. We assess the performance of the estimator under varying levels of angular uncertainty and alternative error distributions, with a focus on its ability to correctly separate the two geological groups. Results show that explicitly modeling measurement error leads to improved discrimination accuracy and more reliable identification of structural patterns compared to standard methods that neglect noise. Full article