This paper investigates the challenges of event-triggered trajectory tracking control for multiple high-speed train nonlinear systems subject to model uncertainty. First, a neural network-based minimum learning parameter (MLP) method is proposed for multi-train nonlinear systems with model uncertainty and integrated into the controller design, eliminating conventional neural network weight selection. Second, to reduce the communication resource burden, a fixed-threshold event-triggered mechanism is developed. Finally, an event-triggered adaptive tracking control approach based on the MLP method is presented within the sliding mode control framework. Theoretical analysis demonstrates that all closed-loop control signals are uniformly ultimately bounded and that Zeno behaviour is avoided. Simulation results verify the effectiveness of the proposed control scheme.
In this paper, an innovative approach named the dynamic pole motion approach (DPMA) has been explained for analyzing equilibrium points for the stability of nonlinear dynamical systems. The stability of a nonlinear dynamical system can be analyzed in g-space using DPMA without linearization. In a linear system, an equilibrium point is uniquely located at the origin; however, in nonlinear systems, it may have several isolated equilibrium points. DPMA describes the motion of dynamic poles in g-space, where ‘g’ is represented in a three-dimensional g-space; [Formula: see text] (where [Formula: see text]). For a stable system, the characteristic equation's roots (poles) of the dynamical system must lie on the left-hand side of the g-space. This implies that the net feedback in the system is negative. One or more poles in the g-space's right-hand side will provide positive net feedback, causing instability. Stability of L.T.I. systems is shown in the s-plane. However, the ‘g’ operator and g-space can be used for both linear and nonlinear systems. The stability and classification of equilibrium points for nonlinear systems are also explained by using DPMA. There is less mathematics by using DPMA, which simplifies the analysis of equilibrium points for the stability of the nonlinear dynamical systems with constraints.
In this work, we study the tracking control problem for the longitudinal motion of aircraft in the presence of uncertainties and input nonlinearity. Prescribed performance control strategy is utilized to obtain the preset transient and steady-state performance, while the unknown input deadzone is dealt with according to a robust method. Backstepping method is employed to design the adaptive prescribed performance flight-path angle controller, with event-triggered control strategy used to reduce the control and computation expenditures. To alleviate the inherent problem of “explosion of complexity” in backstepping design, two tracking differentiators are constructed to approximate the time derivatives of virtual control laws. The stability of the controlled system is proved by using Lyapunov theory. The tracking error is always restricted within the preset bound during the control process, and ultimately converges to the neighbourhood of origin. Numerical simulation results are provided to demonstrate the efficiency of the proposed event-triggered prescribed performance flight-path angle control scheme.
Conventional cleaning and waste-collecting robots face significant challenges when operating in dynamic and unpredictable environments. Traditional obstacle-avoidance algorithms often fail to account for moving objects, while existing learning-based navigation techniques demand extensive datasets and high computational resources, limiting their applicability in low-cost service robots. To address these limitations, this study proposes an adaptive autonomous navigation framework that integrates Imitation Learning (IL) with the Deep Deterministic Policy Gradient (DDPG) algorithm. The hybrid learning model accelerates policy acquisition through expert demonstrations and continuously refines navigation behavior via reinforcement feedback, ensuring real-time adaptability and robust obstacle avoidance in dynamic environments. The system was developed using ROS2 and evaluated within a Gazebo simulation environment. Experimental results across multiple validation trials demonstrate a consistent 23.7% improvement in path efficiency, an 18.4% reduction in collision rate, and 27% faster convergence compared to baseline reinforcement learning approaches, including standard DDPG and its variants. The proposed neural control architecture successfully generalizes across both indoor and outdoor navigation tasks, maintaining stability under sensor noise and environmental variations. These outcomes validate the feasibility of deploying the proposed framework in practical cleaning and waste management applications, particularly for resource-constrained robotic systems, offering enhanced autonomy, responsiveness, and energy efficiency.
The PSO-LBFNN algorithm integrates Particle Swarm Optimization (PSO) with a Logistic Basis Function Neural Network (LBFNN), which uniquely employs a Logistic probability density function as its basis function. A key innovation lies in that only one hidden layer with 600 neurons is sufficient to achieve the desired accuracy. Furthermore, unlike conventional approaches that rely on search-based optimization methods such as Adam, the proposed PSO-LBFNN algorithm employs an iterative formula as the training mechanism to directly minimize the loss function. It can simultaneously solve the reliability equation and optimize implicit objectives, providing a theoretical foundation for reliability estimation and optimal control in engineering applications. To evaluate performance, the proposed algorithm is compared with classical methods such as Genetic Algorithm+ANN (GA-ANN) and Differential Evolution+ANN (DE-ANN). Results demonstrate that our Algorithm has a better performance than both GA-ANN and DE-ANN in convergence speed and solution accuracy. As for the optimal control, compared to the uncontrolled case, PSO-LBFNN enhances the reliability function by 63.8%. In addition, the mean first-passage time (MFPT) under PSO-LBFNN prolongs significantly from 24.4183 (uncontrolled) to 38.6786 (controlled), far exceeding GA-ANN and DE-ANN. At last, a Mann-Whitney U test confirms the statistical superiority of PSO-LBFNN with high confidence (p<0.01).
Cloud computing is revolutionizing the finance sector for scalable machine learning applications for use in fraud detection, risk assessment, and credit scoring, among others. However, still, there is still a need for balancing high accuracy with strict adherence to data privacy and regulatory compliance. This paper discusses a new framework called the Secure Federated Cloud for Financial Analytics, dubbed SFC-FA, which is specifically constructed for ML-driven fraud detection in decentralized financial systems. We propose a novel learning paradigm, Adaptive Secure Federated Learning (ASFL) - an advanced federated approach that integrates reinforcement learning (RL), differential privacy (DP), and homomorphic encryption (HE) for privacy-preserving and resource-efficient fraud detection in distributed financial environments. By utilizing real-time workload patterns, ASFL ensures efficient resource allocation with minimal latency and high computational performance. Experimental evaluation of the federated financial transaction dataset establishes that the proposed framework correctly detects fraud with an accuracy value of 97.5%, which is 9% more accurate than traditional federated ML models and reduces resource consumption by 12%. This work provides a transformational solution in terms of the deployment of robust, privacy-preserving ML applications in the financial sector, providing secure, accurate, and scalable fraud detection on the cloud.
Correct identification of anterior cruciate ligament (ACL) tears from MRI scans of the knee con tinues to be an immense challenge in the field of orthopaedic diagnostics. This study proposes hybrid technique that integrates Zernike Moments for robust rotation-invariant feature extrac-tion with a Support Vector Machine (SVM) classifier, whose hyperparameters, namely kernel type, regularization parameter, C, and kernel coefficient y, have been optimized thoroughly using the Amended Golden Search Optimization (AGSO) algorithm. The AGSO algorithm with chaos the ory initialization and opposition-based learning systematically identified the Linear kernel as the best-performing configuration on the MRNet dataset. The method was evaluated on a balanced test set of 638 images (319 ACL tears, 319 normal), demonstrating superior diagnostic accu racy with an accuracy of 92.25%, a precision of 85.23%, and a sensitivity of 98.25%. The results are shown to significantly outperform six most state-of-the-art methods, such as Deep Belief Networks, Inception-v3, and Random Forests, specifically regarding sensitivity and negative pre dictive values. The suggested SVM-Zernike-Moments-AGSO framework provides an elegant and interpretable non-invasive tool for automated ACL tear detection with the potential to influence and improve clinical decision-making.
Landslides threaten human life, infrastructure, and environmental stability, necessitating rapid response systems enabled by accurate and timely detection. This study carried out a comparative evaluation of the deep learning methods against the traditional machine learning algorithms in the detection of landslides using satellite remote sensing imagery. The proposed CNN-based architecture, comprising VGG and ResNet models, is compared with conventional ML algorithms, namely, Random Forest, Support Vector Machine, Decision Tree, K-Nearest Neighbors, and Logistic Regression. Experiments were carried out on two benchmark datasets Recent Landslide Database and Relict Landslide Database using False Positive Rate (FPR) and Matthews Correlation Coefficient as key metrics. Results reveal that the proposed ResNet gives the minimum FPR (0.0683% and 0.1296% for RLD and LLD, respectively) and maximum MCC values (0.7012 for RLD and 0.724 for LLD), thus besting all traditional models. VGG gives competitively very high MCC scores with stable accuracy. RF offers good accuracy but suffers from more false positives, while SVM and KNN offer the worst of classification results, especially on LLD. Additionally, an Average Relative Predictor Importance (ARPI) study is performed that distinguishes slope gradient and curvature as the most important features for landslide prediction.
Privacy-preserving machine learning is critical if you cannot centralize data or share it with third parties (ethical AI). We propose the first such end-to-end empirical analysis of privacy-utility trade-offs - spanning from model performance, fairness, and explainability across the machine learning pipeline with differential privacy (DP) applied at distinct steps. We analyze nine machine learning models using the User Privacy and Advertising Dataset (UPAD), which includes 50,000 synthetic user records with 29 privacy-aware features, under various & varepsilon; is an element of [0.1, 5.0] privacy budgets. Our results demonstrate that privacy-preserving mechanisms can produce classification accuracy on the order of 54-67%, presenting a direct challenge to the canonical postulate of the privacy-utility trade-off. Importantly, we show that privacy and fairness are orthogonal: both can be satisfied simultaneously for a given algorithm, but achieving strong privacy does not ensure fair outcomes too - on average, violations of demographic parity are 22.2%. We have 80% consensus on feature importance under DP: Interpretability analyses via LIME and SHAP 2.4. A mixed-effects model detects two clusters of performance, with the best possible accuracy similar to 67.35% obtainable when & varepsilon; = 0.133. Collectively, these results offer a conceptually grounded and empirically validated framework to go beyond the privacy-utility trade-off with important implications for regulatory compliance and responsible AI deployment.
Multilevel inverters find a significant role over two-level inverters in Adjustable Speed Traction Drives (ASTD) and Renewable Energy Sources (RES) applications, due to their inherent uniqueness in producing high-quality output voltage with considerable reduction in harmonic distortion, reduced Electromagnetic Interference (EMI), and lower Total Standing Voltage (TSV) of the switching devices. Despite their benefits, classical topologies require an increased number of switching components, gate drivers, and isolated DC sources proportional to an increase in voltage levels, which makes the system more complex and increases implementation cost. To overcome the above drawbacks, a new solution is proposed in the form of a novel MLI structure that arranges a set of four isolated DC sources in an “S” shape to minimize total power components and switches in the conduction path. The proposed configuration has been modelled in MATLAB/Simulink and realized using an experimental prototype. Simulation and experimental studies on 9-, 15-, and 31-level output configurations demonstrate that the proposed topologies achieve peak efficiencies of 96.8%, 97.6%, and 99.3%, respectively, under a fixed RL load with a power factor of 0.954. These results confirm the energy-efficient performance of the system under realistic conditions. The proposed structure demonstrates substantial improvements in output performance over traditional MLIs, bringing a new avenue in MLI topology.
The propulsion system of vertical take-off and landing vehicles relies heavily on batteries as the main power source. The batteries’ health and remaining capacity should be carefully monitored for the safe and healthy operation of the eVTOL vehicles. State-of-the-art data-driven machine learning algorithms are increasingly used for battery status estimation. This study proposes a machine learning-based estimation technique to monitor the state of health, remaining useful life and maximum operating temperature of an eVTOL vehicle's battery. In this context, random forest and 2nd-, 3rd- and 4th-degree polynomial regression algorithms are implemented on a mission-based, publicly generated dataset. The limited data availability condition considered for each mission dataset and the hyperparameters of machine learning algorithms are also optimized to enhance the estimation accuracy. The results indicate that the least weighted average error rates in estimating battery state of health, remaining useful life and maximum operating temperature are achieved by algorithms: 2nd-degree polynomial regression, 3rd-order polynomial regression and random forest. Finally, the proposed technique achieves lower error rates in battery state estimation than prior studies, despite the limited data availability during the learning phase of the selected data-driven algorithms.
In this paper, an event-triggered adaptive control law is proposed by backstepping for a class of nonlinear systems with unknown saturation input. It also considers uncertain constant parameters and unknown time-varying input gains in the system. The analysis of the relationship between the continuous input signal before triggering and the discrete signal after triggering, transformed by a saturated actuator, is key to the controller design. Based on this, a new smooth function is constructed to approximate asymmetric saturation. The shortcomings of the auxiliary signal method and fuzzy approximation method can be overcome by this approximation. The approximation error is analysed in conjunction with the uncertainty introduced by the unknown time-varying input gains. The Nussbaum function is introduced to compensate for the impact of this combined uncertainty. All signals remain bounded by the proposed control scheme in the closed-loop system and simulation results can also prove the effectiveness of this proposed controller.
In currently-used target tracking algorithms, such as Mean Shift (MS), Kalman Filter (KF), Particle Filter (PF), and Convolutional Neural Network (CNN)-based trackers, the external environment significantly affects feature extraction accuracy and the success rate of target tracking. To address these limitations, we propose a hybrid algorithm combining a Stacked Denoising Autoencoder (SDAE) and Particle Filter (PF), which enhances feature extraction, reduces noise, and adapts to dynamic environmental conditions. Two functions employ a uniform update weight across frames to maintain consistent feature representation. Experimental results demonstrate that the proposed SDAE - PF algorithm achieves a tracking accuracy of 94.3%, a 12.5% improvement in robustness under environmental disturbances, and a 15.2% reduction in feature extraction errors compared to conventional methods. These results confirm the effectiveness of the proposed method for real-time, high-precision target tracking in complex scenarios.
The nonlinear loads create power-quality (PQ) issues in the electrical distribution system. Various types of power electronics-based devices are being used to minimize the PQ problems in the distribution system. These devices inject currents and voltages into the grids. Mostly, a phase-locked loop (PLL) is utilized for the estimation of synchronous data like voltage magnitude and phase angle. However, the occurrence of DC offsets and harmonics in grid voltage creates fundamental frequency oscillations. To overcome these issues, an advanced second order generalized integrator (SOGI) based PLL is usually used. Hence, SOGI-based advanced controllers are being developed and used for generating reference current to inject the triggering pulses to the distribution static compensator (DSTATCOM). This work provides a comparative analysis of various SOGI-based advanced controllers to reduce the harmonics and DC offsets problems with DSTATCOM in the distribution system. The performance of the controllers is analyzed and compared.
Sports analytics and prediction play an increasingly important role in modern sports by providing insights that enhance performance, strategy, and decision-making. Bowling is a precision sport in which players roll a ball along a lane to knock down pins. The sport comprises multiple disciplines practiced internationally, with 9-pin bowling being among the most widely played. In competitive 9-pin bowling, team competition is a prominent format. Team matches consist of parallel individual duels. The final match outcome is determined by both the win-loss outcomes of individual duels and the team’s aggregate number of knocked-down pins. This paper presents a two-level predictive framework for 9-pin bowling: one model predicts individual player performance, and another predicts team match outcomes based on those individual predictions. The models were evaluated using match data from the second half of the 2022/2023 season of the Croatian Men's Bowling Super League. Individual performance was predicted with an average deviation of less than 4%, while the winning team was correctly predicted in over 75% of matches. Beyond these strong results, the proposed models are, to the best of our knowledge, the first developed specifically for predicting outcomes in 9-pin bowling.
Chest diseases pose a significant health challenges and requires earlier diagnosis. Chest X-ray (CXRs) images are globally utilized due to their accessibility and cost-effectiveness, but physical intervention is time-consuming and error-prone. Recent improvements in Artificial Intelligence (AI) have enabled faster and more accurate automated disease classification. To address the challenges of limited data and longer training time, this research employs three stage deep learning pipeline for chest disease classification:(1)Bounding box guided U-Net based segmentation to isolate disease affected region, reducing irrelevant background noise and focusing on the clinically significant area (2) Conditional Generative Adversarial Networks (cGAN) and conventional geometric transformation based augmentation technique to address the issue of limited training data and improve the model generalization, (3) attention-based classification mechanism by incorporating transformer-based models which effectively captures long range dependencies within the affected region. This method has been evaluated on two Datasets comprising 4800 images, which were generated from 880 images selected from the NIH Chest X-ray 14 data source across 8 classes through traditional and cGAN-based augmentation techniques. The U-Net + cGAN + DeiT achieves 93.23% accuracy, 93.53% precision, 93.23% recall, 93.28% F1-Score, and requires 98.08sec training time, illustrating its effectiveness in chest diseases.
Considering the increasing demand for environmentally friendly and economically viable transportation options, this in-depth analysis of fault-tolerant control (FTC) in the context of electric vehicles (EVs) covers all the latest developments and applications in the field. Maintaining vehicle stability and handling component failures with minimal or acceptable loss of performance are the primary goals of FTC systems in EVs. Thus, the FTC's crucial role in improving EV dependability is examined in this study. An explanation of the control strategies employed in EVs is followed by a description of the several types of FTC, including active, passive, and hybrid systems. The study aims to objectively evaluate advancements in tracking accuracy and robust performance by thoroughly reviewing FTC systems for modern EVs. This discusses various strategies as well as the challenges of integrating them into EV subsystems. Real-time deployment, validation against coupled failures, and the incorporation of learning-based FTC into safety-critical EV systems are some suggestions for future research. This study will help identify research gaps and topics that require additional investigation in order to advance the discipline.
This paper presents a comparative study between intelligent and conventional control strategies for accurate 3D trajectory tracking with joint limit avoidance in spatial robotic manipulators. The proposed intelligent approach integrates a Kohonen Self-Organizing Map (KSOM) neural network with a Weighted Least Norm (WLN) scheme referred to as KSOM-WLN to effectively address redundancy resolution while significantly reducing the computational load typically associated with Jacobian pseudo-inverse calculations. For comparative evaluation, three conventional methods are developed: PID-Controlled Closed-Loop Inverse Kinematics with Task Priority (PID-CLIK-TP), PID-CLIK with Weighted Least Norm (PID-CLIK-WLN) and Adaptive (gain-scheduled) PID-CLIK. A 5-degree-of-freedom spatial robotic manipulator is modeled in MATLAB to assess the control performance across seven 3D trajectory types: straight-line, circular, elliptical, triangular, rectangular, Lissajous, and spring-shaped. The simulation results confirm that the KSOM-WLN method consistently outperforms conventional approaches, achieving lower root mean square error (RMSE) and higher correlation coefficient (CC) values across all trajectory types. The KSOM-WLN method computational efficiency, requiring approximately 0.01 seconds per trajectory point, significantly faster than the 0.34 seconds observed for PID-CLIK methods. Experimental validation confirms that the KSOM-WLN method ensures smooth, efficient, and highly accurate 3D trajectory tracking.
This study considers the problem of how to guarantee the stability of systems whose behaviour is influenced by time delays and fuzzy parameters. We focus on fuzzy delay differential equations (FDDEs) and a framework based on Lyapunov-Krasovskii functionals. We derive delay - and fuzziness-dependent linear matrix inequalities that are easy to check with optimization software. These inequalities constitute conditions under which the zero solution of an FDDE is asymptotically stable at an exponential or $ p $ p-exponential rate. To show that the theory is more than a mathematical exercise, we test it on two examples. A temperature controlled chemical reactor with a thermal lag settles once criteria are imposed. In a second case, a market price model subject to constant information delay and fuzzy demand uncertainty drifts back to equilibrium in a p-exponential fashion that methods fail to capture. framework enlarges the certified stability region and demands less computational effort than Lyapunov approaches. These results suggest that engineers and economists can use the criteria as a design tool when delays and fuzzy parameters are unavoidable. Because the method relies only on standard optimization routines, it is readily extendable to adaptive, stochastic, or higher-dimensional settings, opening new avenues for robust controller synthesis and policy design.
Banana leaf diseases such as Cordana, Sigatoka, and Pestalotiopsis significantly reduce crop yield and quality, necessitating accurate and early detection for effective management. This study proposes ResViT HybridNet, a novel deep learning framework that integrates ResNet50 for spatial feature extraction and a Vision Transformer (ViT) for global context modelling, bridged by a Hybrid Pool Block (HPB) to preserve spatial locality. Using a dataset of 2537 images across four classes, the model achieved an overall accuracy of 99.21%, precision of 0.9843, recall of 0.9921, and F1-score of 0.9881, outperforming conventional CNN and transformer-based models. Extensive ablation and statistical significance analyses confirm the complementary synergy between CNN and transformer components. These results demonstrate that ResViT HybridNet provides a robust and accurate solution for automatic banana leaf disease identification, offering strong potential for deployment in real-world agricultural disease monitoring and crop management systems.
