study proposes a hybrid anomaly detection framework based on Independent Component Analysis (ICA) and Bayesian Long Short-Term Memory (LSTM) networks for early fault diagnosis in rolling element bearings. The model is trained exclusively on normal condition data from the Case Western Reserve University (CWRU) Bearing Dataset, enabling unsupervised deployment suitable for real-world industrial environments. Vibration signals are preprocessed using ICA to extract statistically independent features, followed by a Bayesian LSTM that captures temporal dynamics and provides uncertainty-quantified predictions. A dynamic thresholding mechanism based on Interquartile Range (IQR) autonomously distinguishes between normal and anomalous behavior without manual calibration. Experimental results demonstrate exceptional performance with 99.1% accuracy, perfect recall (100%) on fault conditions, and zero false negatives, ensuring no faults are missed. The composite anomaly score effectively tracks degradation progression, thereby rendering the system highly reliable and practical for predictive maintenance applications. This approach combines statistical signal processing with probabilistic deep learning to deliver a robust, explainable, and adaptive solution for bearing health monitoring.
Numerous investigations have explored automated control mechanisms for artificial hands; nonetheless, the performance of individual robotic fingers remains suboptimal. To enhance motion precision, this study proposes a model of hand movement control utilizing Artificial Intelligence (AI). Selecting suitable sensing components and applying specialized computational algorithms are key factors in optimizing the prosthetic control framework. A control architecture based on machine learning was developed, enabling the hand prosthesis prototype to perform movements automatically according to the classifications generated by the trained model. Experimental results demonstrate that increasing the neuron count in the learning model enhances predictive accuracy while minimizing loss values. A comparison between Artificial Neural Network (ANN) and Recurrent Neural Network-Long Short-Term Memory (RNN-LSTM) architectures revealed that the ANN configuration with a 32-neuron hidden layer provided the best results, achieving a loss below 0.1 and accuracy above 90%. Therefore, this ANN model was chosen as the main control algorithm for the prosthetic system. When implemented in the master controller, the model produced prediction accuracies exceeding 90% across all output classes, successfully activating the prosthetic hand according to ten predefined motion labels.
Bearing faults are a major cause of degradation in Brushless DC (BLDC) motors, making reliable detection of different bearing fault conditions essential for maintaining industrial equipment. This study proposes a diagnostic method that combines ReliefF feature selection with a Random Forest classifier to identify both single and compound bearing faults. Unlike earlier studies that applied this combination mainly to single-fault or binary classifications, the present work addresses a more realistic seven-class bearing fault problem. From the vibration signals, eighteen statistical features were calculated, and ReliefF was employed to identify the features that most effectively distinguish among the bearing fault categories. The resulting feature ranking improves separability, especially for compound-bearing faults that often exhibit overlapping spectral characteristics. With these selected features, the Random Forest model achieved strong diagnostic performance, demonstrating that the proposed framework offers an efficient and practical solution for identifying complex bearing fault conditions in BLDC motors.
Jumping robots adopt an efficient locomotion strategy to overcome obstacles in competitive and unstructured environments. However, many existing designs rely on complex mechanisms with multiple degrees of freedom and sophisticated control algorithms, which limits their practicality in time-constrained applications. In this study, a lightweight single-degree-of-freedom jumping robot actuated by a pneumatic cylinder combined with elastic elements was developed. The design of the system and the results of an experimental evaluation are also discussed. The robot uses rubber bands for passive energy storage to enhance cylinder rebound and enable repeatable vertical jumping without complex control. A simplified dynamic model was developed to analyze the relationship between pneumatic force, elastic restoring force, and jumping performance. Experimental tests were conducted by varying the number of rubber bands to adjust the effective spring constant. The results demonstrate that using four rubber bands provides an optimal balance between rebound force and gas efficiency, and the robot achieved a maximum vertical jump height of 270 mm with a rebound cycle time of approximately 0.8 s. Configurations with fewer bands produced insufficient rebound force, whereas higher stiffness resulted in excessive air consumption. The findings confirm that pneumatic-elastic actuation can deliver efficient, stable, and repeatable jumping performance with minimal control complexity. Given its lightweight structure, inexpensive components, and ease of assembly, the proposed jumping robot offers a practical and resource-efficient solution for competition-oriented tasks and potential deployments in disaster relief scenarios where conventional robotic systems are limited.
This study presents the design, optimization, and experimental validation of low-cost, 3D-printed robotic grippers utilizing compliant mechanisms. The development is based on a three-dimensional Topology Optimization Method (TOM) implemented with the Solid Isotropic Material with Penalization (SIMP) approach. The Method of Moving Asymptotes (MMA) was employed to solve the associated nonlinear optimization problems for synthesizing flexible fingers, while a quadratic approximation method was utilized for designing rigid support structures. Two distinct objective functions were formulated: the first aimed to maximize the output displacement of a monolithic Polylactic Acid (PLA) finger by exploiting material compliance, and the second focused on minimizing the weight of the support structures under stiffness constraints. The optimized designs were fabricated via additive manufacturing and assembled into two-, three-, and four-finger gripper configurations. Experimental evaluation on an ABB IRB-120 robotic arm demonstrated the grippers' exceptional adaptability in manipulating objects of varied geometries and surface textures. Furthermore, payload capacity tests revealed maximum loads of 1.5 kg, 2.2 kg, and 2.7 kg for the two-, three-, and four-finger grippers, respectively. The results confirm that the proposed methodology, centered on MMA-based 3D topology optimization, provides an effective framework for developing high-performance, low-cost compliant robotic grippers.
Americas is an organization that promotes the development of electric vehicles for competition. The present work is focused on the design of the America single-seater structure which was based mainly on the protection of the driver, space for various accessories, improve range and aerodynamics. The design was carried out in strict adherence to the Electrathon America 2021 manual. The 3D Computer Aided Design (CAD) modeling was performed using SolidWorks software. Aluminum and steel profiles with diameters of 1/2 in and 3/4 in with a thickness of 2.7 mm were used. The regulations specify that tests must be performed on the structure of the single-seater, mainly torsion and bending tests to ensure its correct operation with a minimum safety factor of 2; therefore, structural analyses were conducted using Ansys software. Additionally, a fluid analysis was performed. Strain gauges were used to measure structural deformations in order to validate the stress analyses carried out on the single-seater chassis. A functional single-seater vehicle was obtained, featuring a structure with a safety factor of 1.7, improvement aerodynamic, suspension and adequate steering, achieving a range of 38 km with a 1 kWh battery according to Electrathon guidelines, as well as a total mileage of 200 km without failures.
This study presents a three-stage optimisation framework designed for the dynamic performance enhancement of a shipboard horizontal canned-motor pump. The framework integrates high-fidelity Finite Element (FE) modelling, experimental modal validation, and topology-optimisation-based structural redesign, considering realistic installation constraints. A comprehensive whole-pump model, accounting for actual mounting conditions, is developed, and a strategy combining stiffening, lightweighting, and reinforcement is implemented to reduce structural mass while improving dynamic performance. The model's accuracy is validated through impact-hammer modal tests, with the first six natural frequencies predicted within 2% of experimental values. Modal analysis identifies a low-frequency global-high-frequency local vibration pattern, with deformation predominantly in the base transition, flange, and outlet regions. Following optimisation, the first natural frequency increases from 42.31 Hz to 76.86 Hz, and higher-order modes increase by approximately 18%, leading to more uniform mode shapes and reduced local vibration. This framework offers valuable insights for vibration mitigation and lightweight structural redesign of shipboard rotating machinery. However, internal fluid-structure interaction within the canned motor and ship-hull foundation flexibility are not incorporated in the current model, and these factors may influence low-order modal characteristics and the applicability of the results under real-world operating conditions. Future research will incorporate Fluid-Structure Interaction (FSI) modelling and shipboard experimental validation to further assess the robustness and engineering applicability of the proposed framework.
The objective of this study is to investigate and predict wear behavior in contacting surfaces through an integrated approach that combines Continuum Damage Mechanics (CDM) and Artificial Neural Networks (ANN). Pin on disk wear experiments were performed under dry conditions using three engineering materials ST37, C45E4, and Al7075 with variations in load, sliding speed, and hardness systematically designed using a full factorial Design of Experiments (DOE). The CDM model quantified material degradation and estimated wear coefficients, which were then used as training data for a feed-forward back-propagation ANN. Both models were validated against independent experimental data. Results indicate that the ANN model achieved high prediction accuracy (average error <5%), outperforming the CDM model (average error <= 12%). Scanning Electron Microscopy (SEM) revealed adhesive wear as dominant in the steels, while Al7075 exhibited reduced wear due to higher hardness. The interaction effects showed that load and sliding speed have significant influences on wear, whereas hardness plays a secondary role. The findings establish a robust framework for wear prediction, process optimization, and potential real-time monitoring in engineering applications, demonstrating the effective integration of physics-based and data-driven modeling in predictive tribology.
This study investigated the influence of machining parameters and performed a comparative multi-objective optimization of surface roughness (Ra) and Material Removal Rate (MRR) in the milling of 7075 aluminum alloy using Multi-Objective Particle Swarm Optimization (MOPSO), Non-dominated Sorting Genetic Algorithm II (NSGA-II), Strength Pareto Evolutionary Algorithm 2 (SPEA2), and Multi-Objective Ant Colony Optimization (MOACO). Three machining parameters, including spindle speed (S), feed rate (f), and depth of cut (d), were considered. Analysis of Variance (ANOVA) results showed that S and f significantly affect Ra (p < 0.05), contributing 20.06% and 79.68%, respectively, while f and dare the most significant factors influencing MRR (p < 0.05), accounting for 45.01% and 40.11% of the total contribution. A predictive model for Ra developed using the Group Method of Data Handling (GMDH) demonstrated high predictive performance, with R2 values of 0.9990 and 0.9962 for the training and validation datasets, respectively. Comparative analysis indicated that NSGA-II produced the most stable solutions, whereas SPEA2 and MOACO exhibited less balanced performance, and MOPSO achieved rapid convergence with relatively dispersed solutions. Experimental validation of Ra and analytical verification of MRR confirmed the reliability of the proposed framework, with mean deviations of 6.5% and 0.37%, respectively. Unlike prior investigations that examined individual algorithms or lacked integrated experimental assessment, this study presents a systematic cross-algorithm evaluation under identical machining conditions. The proposed framework integrates statistical contribution analysis, predictive modeling, and experimental validation, thereby establishing a robust and practically applicable approach for multi-objective milling optimization.
In this research, we discuss how we extract features related to natural frequency of Nonlinear Composite Plates (NCP) using a combination of Wavelet Transform (WT) and Artificial Intelligence (AI) algorithms. The nonlinearity is represented in the plate geometry and Boundary Conditions (BCs). Our findings, which build on previous works by the authors, indicate that WT can effectively reflect the natural frequency features of NCP. However, we also noted that this approach can be quite complex, involving numerous calculations and iterations, which means it might not be ideal for quickly and accurately extracting natural frequency. To tackle this issue, we've developed a new AI model designed to learn from and test these results by extracting key natural frequency features that reveal crucial information about the NCP behavior. The AI model is based on a Recurrent Neural Network with Long Short-Term Memory (RNN-LSTM) blocks since the natural frequency datasets have a time-dependent and memory-dependent behavior. Our results strongly suggest that the proposed technique is a promising approach, especially for complex structures in varying environmental conditions.
In the modern era, Robotics has become an essential part of modern life, enhancing efficiency and precision in manufacturing and supporting diagnosis, prediction, and surgery in medicine. This paper presents a real-time teleoperation framework that maps human upper-body motion, captured by a single RGB-D camera, to a dual-arm upper humanoid robot designed with low-cost servos. The system employs MediaPipe-based pose estimation and a torso-anchored coordinate transformation to achieve operator-centric retargeting that is robust to variations in camera placement and subject geometry. To suppress tremor and sensor noise, a constant-velocity Kalman filter combined with an adaptive dead-zone is applied to wrist trajectories, ensuring smooth motion while maintaining responsiveness. A min-max scaling function with saturation enforces safe workspace mapping, while joint commands are computed using damped-least-squares inverse kinematics with joint-limit and self-collision checks. The execution layer incorporates incremental speed-aware stepping to emulate continuous trajectories on servo actuators. Experimental results demonstrate accurate static pose reproduction, robust dynamic path following, and zero joint-limit violations, achieving an average wrist-tracking Root Mean Square Error (RMSE) of 12.4 mm and median end-to-end latency of 86 ms. The platform is reproducible, cost-effective, and adaptable for applications in education, rehabilitation, and human-robot collaboration.
This paper presents a method for identifying the hazardous configuration of a KUKA KR3 R540 Six-Degree-of-Freedom (6-DOF) industrial robot. A kinematic model is first established using the standard Denavit-Hartenberg (D-H) method, and a multi-body dynamic model is subsequently constructed by incorporating the Newton-Euler formulation. To identify the pose that induces the maximum joint torque, a search strategy based on the Particle Swarm Optimization (PSO) algorithm is proposed, implemented through co-simulation between Adams and MATLAB. The optimization objective is defined as maximizing the sum of the absolute driving torques of joints J2 to J6, with the corresponding joint angles serving as decision variables. The PSO algorithm, driven by MATLAB, generates candidate poses, while Adams performs high-precision dynamic computations. This framework enables an automated search across the high-dimensional joint space to iteratively locate the global optimum. Simulation results demonstrate that the proposed method effectively identifies the global most hazardous configuration, corresponding to a fully extended manipulator pose with J2 approximate to-90.81 degrees and J3 approximate to 82.57 degrees. The maximum total joint torque in this configuration is approximately 182.47 N.m. These results provide crucial load boundary conditions for structural strength verification and lightweight design, while also offering valuable insights for the selection of key components in structural optimization.
Policy transfer is an efficient approach for developing specific robots. Its effectiveness depends on high-quality imitation datasets and a stable learning process. However, substantial differences in geometry and dynamics between source and target robots pose challenges. Purely kinematics-driven mapping methods and manual parameter tuning often fail to maintain kinematic-dynamic consistency. In this study, we transfer control policies from the quadruped robot Unitree Go1 to our self-developed heavy wheel-legged robot Tiangou. We propose a Consistency-Aware Retargeting (CAR) method. This extends conventional inverse kinematics by adding dynamic consistency constraints. Using motion data from Go1's Model Predictive Controller (MPC), CAR generates a reference dataset for Tiangou. We then integrate Bayesian Optimization (BO) into the imitation learning framework. This enables autonomous tuning of policy model structures and optimization hyperparameters. Experiments show that CAR reduces foot-end position errors, mitigates joint angular velocity fluctuations, and decreases foot-end slippage. Moreover, Bayesian optimization improves sample efficiency and training stability. These contributions establish a practical foundation for policy transfer across heterogeneous robotic platforms.
Resistance Spot Welding (RSW) is widely used in automotive and manufacturing industries for joining metallic sheets; however, welding dissimilar metals such as mild steel and stainless steel remains challenging due to their differing thermal and metallurgical properties. This study investigates the influence of preheating the stainless steel component at varying temperatures (100 degrees C, 120 degrees C, 150 degrees C, and 180 degrees C) on the mechanical and microstructural characteristics of dissimilar RSW joints with mild steel. Tensile testing, hardness profiling, and Scanning Electron Microscopy/Energy Dispersive Spectroscopy (SEM/EDS) analyses were conducted to evaluate joint performance. The results revealed that preheating significantly affects residual stress distribution, ductility, and joint homogeneity. Without preheating, joints exhibited high tensile strength (127.46 MPa) but low ductility due to rapid cooling and martensitic formation. Optimal conditions were achieved at 180 degrees C preheat, yielding the highest tensile strength (156.32 MPa), improved ductility (strain = 1.80% Gauge Length), and the lowest standard deviation (6.95 MPa), indicating enhanced process stability. Hardness analysis confirmed a balanced gradient across the weld, heat-affected zone, and base metal, while SEM observations identified reduced microcracks and improved microstructural uniformity at higher preheat levels. Overall, a preheat temperature of 180 degrees C effectively minimizes thermal gradients and residual stresses, improving weld integrity and consistency. These findings provide practical insights for optimizing preheating parameters in industrial RSW applications, particularly in the fabrication of automotive body structures and other dissimilar steel assemblies.
This paper proposes a New Adaptive Neural Fuzzy Sliding Mode Controller (NANFSMC) for regulating a Coupled Tank System (CTS), with unknown nonlinear dynamics in experimental environments. The CTS exhibits strong nonlinearities and uncertainties arising from sensor noise, parameter variations, variations in output valve characteristics, and significant time delays. The proposed control architecture integrates two synergistic components. The first component is an adaptive control system that utilizes a Radial Basis Function Neural Network (RBFNN) to approximate the adaptive control law, featuring an adaptive updating mechanism to compensate for RBFNN approximation errors. The second component is a Sliding Mode Control (SMC) system, whose parameters are updated in real-time via a fuzzy inference mechanism to enhance robustness. Both control laws are derived within the framework of Lyapunov stability theory, ensuring closed-loop stability under all operating conditions. The proposed controller possesses a simple structure, resulting in low computational load and requiring only a few tuning parameters. Although the RBFNN weights are initialized to 0, the integration with the adaptive fuzzy mechanism allows fast convergence and rapid stabilization. Furthermore, this study presents the first experimental validation of a Takagi-Sugeno (TS)-fuzzy-based adaptive tuning of the SMC robustness gain on a real CTS under external disturbances. The proposed method achieves improvements of up to 22.9% and 14.2% in the Integral of Absolute Error (IAE), Mean Absolute Error (MAE), and Integral of Time-weighted Absolute Error (ITAE) indices compared to the Adaptive Neural SMC (ANSMC) and Proportional Integral Derivative (PID) controllers, respectively.
fire suppression activities subject rescue personnel to severe thermal conditions, hazardous fumes, and blast risks, creating extremely perilous environments for human operators. Rapid urban development has amplified fire emergency occurrences, necessitating the deployment of advanced autonomous firefighting platforms. This study presents an innovative tracked firefighting robot designed to navigate complex terrain and autonomously detect and approach fire sources. The system integrates a You Only Look Once version 8 (YOLOv8)-based deep learning model for real-time fire detection and employs depth imaging to calculate angular deviation and distance to the fire. These measurements are transmitted to a Programmable Logic Controller (PLC)-based control unit via a Modbus RS485 interface for responsive control. To enable autonomous navigation, the proposed robot combines an enhanced Bug-2 pathfinding algorithm with LiDAR-based environmental mapping and Hector Simultaneous Localization and Mapping (SLAM) for real-time localization and mapping. The core innovation lies in the integration of YOLOv8-based fire detection with deviation-angle-optimized Bug-2 navigation and a PLC-Robot Operating System (ROS) control architecture, enabling precise fire localization and obstacle avoidance in dynamic environments. Experimental validation confirms the effectiveness of the proposed firefighting robot in identifying fire sources and navigating around obstacles, demonstrating its potential as a reliable solution for autonomous firefighting in hazardous scenarios.
significant concerns have arisen regarding the application of biologically inspired robots in rehabilitation for individuals with movement disabilities. These types of robots must ensure a high level of safety, which is typically achieved through more flexible construction. Pneumatic Artificial Muscles (PAMs), driven by compressed air, exhibit performance similar to biological muscles. Consequently, PAMs are considered strong candidates for actuators in rehabilitation robots. This paper investigates a control algorithm for a multi-fingered robot actuated by PAMs for grasping and manipulating circular objects. A dynamic model of the general robot-object system was formulated using the Lagrange method, combined with the natural force-length-velocity relationship of contracting muscles. Based on this model, control algorithms were proposed to achieve stable grasping and dexterous manipulation of the object by the multi-fingered robot. The asymptotic convergence of the closed-loop system was analyzed using Lyapunov's principle and the extended LaSalle invariance theorem. Simulation results further validated the effectiveness of the proposed control algorithms.
This study presents the design and development of a fuzzy control system for a hybrid Three-Wheeled Omnidirectional Mobile Robot (3WOMR). As a holonomic robot, it can perform simultaneous translational and rotational motions. The proposed fuzzy controller enables efficient obstacle avoidance in environments with static and dynamic obstacles while minimizing structural vibration using rubber wheels for damping. An RPLIDAR sensor is employed to detect obstacle distances in three 45 degrees-spaced sectors. The controller uses three input parameters distance from the head (DH), left (DL), and right (DR) to determine two output variables representing the angular velocities of the left and right wheels. The system operates through seventeen cognitive states and twenty-seven fuzzy rules implemented in Python using fuzzy logic libraries. Experimental tests were performed in a 3 & times;4 m environment, both with and without obstacles. The robot successfully navigated the area, avoiding collisions and maintaining stability. MATLAB 2023b simulations confirmed the system's reliability and performance. The proposed fuzzy controller demonstrated improved accuracy and efficiency compared to existing methods, providing effective control of the robot's linear and angular motion for safe navigation in real-world conditions.
High-precision mechanical assemblies require accurate error propagation models to predict error accumulation in the final assembly and ensure optimal performance and reliability. Conventional linear models exhibit limited accuracy in predicting geometric errors. To address these limitations, this study proposes a modified connective assembly model based on second-order nonlinear error propagation using homogeneous transformation matrices. The model is implemented in Python and validated against existing linear and fully nonlinear assembly models to evaluate predictive accuracy and computational efficiency. At an angular orientation error of 1.0 degrees, the linear model exhibits Z-direction errors of 0.40 mm, 0.60 mm, and 0.80 mm for 4, 6, and 8 component assemblies, respectively, whereas the fully nonlinear model predicts 0.10 mm, -0.57 mm, and-2.18 mm. The developed model reduces these discrepancies to 0.23 mm, -0.25 mm, and-1.59 mm, achieving improved predictive accuracy of 56.67%, 72.65%, and 80.20%, respectively, over the linear model. Similarly, under a geometric run-out tolerance of 1.0 mm, the linear model predicts an error of-1.60 mm, -2.40 mm, and-3.20 mm, compared to-1.70 mm, -2.79 mm, and-4.18 mm for the fully nonlinear model in the Z-direction. The proposed model narrows these gaps to-1.66 mm, -2.68 mm, and-3.99 mm, delivering predictive accuracy of 60.00%, 71.79%, and 80.61%, respectively. Moreover, Monte Carlo simulation results on 4-component assembly confirm that the proposed model replicates the statistical characteristics of the fully nonlinear model while reducing execution time from 9.34 s to 4.82 s, achieving a 48.39% reduction in execution time.
Path planning remains a critical research area in mobile robotics, yet current approaches often suffer from suboptimal path quality, limited sampling efficiency, and inadequate adaptability across diverse operational scenarios. To address these issues, this paper proposes an improved algorithm combining Artificial Potential Field (APF) and Restricted Path Time (RRT*) approaches. This algorithm employs an optimization model that combines dynamic sampling with potential field guidance, constructing a two-stage dynamic sampling mechanism. During sampling, candidate nodes with Gaussian noise are generated along the resultant force direction. Finally, path cost comparison and parent node reselection are performed within the dynamic optimization radius to ensure asymptotic optimality of the path. Experimental results show that in complex maps, path length is reduced by 33.41% and 26.64%, respectively, and planning time is reduced by 21.36% and 86.32%, respectively; in narrow passages, path length is reduced by 49.6% and 49.8%, respectively. The results confirm the effectiveness of the two-stage dynamic sampling mechanism, which not only preserves the probabilistic completeness of the RRT* algorithm but also adaptively adjusts the sampling strategy, improving both planning length and time.
