The stability of autonomous liquid transporter unmanned aerial vehicles (UAVs) is crucial for precision in agricultural field spraying. This paper proposes an adaptive robust control scheme designed to stabilise agricultural quadrotor UAVs affected by variable mass and liquid sloshing dynamics during pesticide application. A comprehensive nonlinear mathematical model is developed to explicitly represent UAV-liquid interactions, accounting for both liquid sloshing and time-dependent mass variations. An adaptive PI-based sliding-mode controller is proposed, featuring a neural network identifier to estimate unknown sloshing forces and a nonlinear disturbance observer to mitigate external disturbances. Theoretical analysis through designed Lyapunov functions confirms the closed-loop system's robustness, stability, and bounded estimation errors, even despite varying liquid mass conditions. Simulation experiments validate the effectiveness of the proposed method, achieving approximately 41.4% and 53.8% in trajectory tracking accuracy, stabilisation, and disturbance rejection along the x-and y-axes, respectively, compared to the conventional proportional integral derivative (PID) counterpart. Overall, this research enhances UAV control methodologies by effectivelymanaging sloshing-induced dynamics, thereby facilitating improved quadrotor system designs and making a substantial contribution to agricultural automation.
The health status of apple leaves directly affects the quality and yield of apples, with foliar diseases posing a serious threat to overall production. Therefore, there is an urgent need for a rapid and accurate method for the classification and identification of apple leaf diseases to enable early and automated detection. This study utilises the Plant Pathology 2020 dataset and proposes a novel classification framework that integrates advanced vision transformers (ViTs) with convolutional neural networks (CNNs), enhanced by adaptive data augmentation strategies to improve classification performance. To address the inherent issue of class imbalance within the dataset, the synthetic minority oversampling technique (SMOTE) is employed to effectively augment the minority class samples. The proposed framework adopts a multi-stage training process to enhance feature representation and model robustness, enabling the effective capture of both local lesion patterns and global leaf characteristics. Experimental results demonstrate the stability and high performance of the proposed method, achieving a classification accuracy of 98.96%. Compared with traditional CNN architectures such as ResNet-50, DenseNet-121, GoogLeNet, InceptionV3, and VGG16, the proposed method exhibits significant improvements in both accuracy and generalisation across various augmented datasets. The findings indicate that the framework offers a feasible, reliable, and scalable solution for plant disease detection in the context of precision agriculture.
Elevator group control systems (EGCSs) are designed to efficiently manage three or more elevators for passenger transport. Most EGCS utilise hall call assignment methods to allocate elevators in response to passenger requests. Traditional EGCS algorithms typically exhibit an average waiting time (AWT) ranging from 20 to 60 s, a maximum waiting time (MaxAWT) exceeding 60 s, and a long-term waiting rate for passengers (LWP) greater than 10%. To enhance efficiency, a hall call assignment method based on the fuzzy EGCS (FEGCS) is proposed. The input variables for the evaluation index parameters of the FEGCS fuzzy inference are employed for elevator dispatching. A fuzzy inference mechanism is established for each evaluation factor, and the optimal elevator dispatching scheme is selected based on weighted criteria. Through MATLAB simulation evaluation, it is demonstrated that, compared to particle swarm optimisation (PSO), backpropagation neural network (BP), and fuzzy control algorithm (FCA), the combination of FCA and BP can reduce AWT by 7.5% to 24%, MaxAWT by 8% to 37.4%, and LWP by 16.7% to 100%.
In the context of processing titanium alloy thin-walled parts, tool runout and workpiece deformation are commonly observed due to their low rigidity. Therefore, it is crucial to comprehensively assess the combined impact of these factors on surface quality. This study proposes a surface topography model that incorporates tool runout and workpiece deformation. Initially, the influence of workpiece material and milling force is disregarded, and the movement path of the tool tip is analysed by modifying key parameters such as spindle speed, feed speed, and tool diameter. By applying the principle of reflection, the resulting geometric shape and texture of the machined surface are determined. Subsequently, a surface topography model is developed by integrating tool runout and workpiece deformation. Simulation results are utilised to investigate the effects of different milling parameters on surface roughness. To validate the accuracy of the established surface topography model, various milling parameters are selected based on a machining stability prediction model, targeting stable milling and chatter milling. When flutter occurs, it not only amplifies variations in surface topography but also exacerbates the vibration of the thin-walled workpiece, leading to the emergence of sawtooth stripes on the surface. Consequently, flutter adversely affects surface quality. The workpiece's surface roughness is measured, and a comparison and analysis are conducted between the predicted surface roughness values obtained through simulation and those measured experimentally. Finally, a genetic algorithm is employed to optimise machining parameters while considering efficiency, resulting in the attainment of optimal milling parameters and achieving desirable surface quality without chatter. The selected parameters adhere to the constraint of maximum force and mitigate the occurrence of chatter. In summary, the optimised processing parameters effectively meet the practical requirements of the machining process, providing valuable insights for practical machining applications.
To address the limitations of traditional detection methods in complex battlefield environments such as haze and night vision, a simulation and recognition framework optimised for smoke interference is proposed. A mathematical model of atmospheric transmission under hazy night conditions is established, and the imaging advantages of linear-array near-infrared push-scanning systems are analysed. Using the Vega real-time simulation platform, a synthetic battlefield scene is used to simulate smoke effects through particle systems, generating target imaging data under different occlusion scenarios. To improve the accuracy of target recognition, an anti-smoke interference module based on the SDNet dehazing algorithm is integrated before the YOLOV5s detection network. The experimental results show that the proposed MSD-YOLOV5s serial architecture can increase the recognition confidence by 3% under smoke, with a mean average precision (mAP) of 0.908 and a detection speed of 130 FPS. These results demonstrate the robustness and real-time performance of the model in harsh environments, providing a practical solution for infrared detection under smoke interference.
During the actual operation of rotating machinery, the external environment and internal structure will change the working conditions. The time-varying speed conditions will lead to the deviation, skew and amplitude change of the vibration signal characteristics of rolling bearing. The fault characteristics under this condition are difficult to fully extract and diagnose. To solve this problem, a diagnosis model based on multi-scale convolution bidirectional long short-term memory neural network is constructed. An intelligent diagnosis method for rolling bearing fault under time-varying speed is proposed. This method combines the more efficient Nadam optimisation method with two independent networks for parallel optimisation training to accurately extract fault features. The influence of Nadam optimisation algorithm on the training process and diagnosis results is analysed. The results of test data diagnosis and visual analysis show that the method can effectively achieve fault diagnosis of rolling bearings under time-varying speed conditions. The comparative analysis shows that the accuracy and robustness of the diagnosis are superior to other methods under the two time-varying speed conditions of speed increase and speed decrease.
As bipedal robot application scenarios expand, traditional planning methods face limitations in balancing efficiency and biomimetic characteristics. This paper proposes a bilayer Bezier curve optimisation method for gait trajectory planning of bipedal robots on unstructured terrain. Unlike existing single-layer Bezier curve and traditional polynomial interpolation methods, this approach innovatively designs a cooperative optimisation architecture for motion state layer and swing phase space layer, achieving an organic integration of macro-level trajectory continuity and micro-level biomechanical characteristics. Through adaptive selection of Bezier curve order and foot trajectory optimisation based on human biomechanical features, the method achieves optimal balance between computational efficiency and trajectory smoothness while realising natural gait patterns. Experimental results show excellent performance across three typical testing scenarios: response time reduced by approximately 50% on ideal level pavement; mean square error decreased by 94.9% compared to traditional methods on cosine wave pavement; and in random rough pavement, center of mass stability improved by 44.1% with mean square error reduced by 74.4%.
In computer vision applications, convolutional neural networks (CNNs) have demonstrated significant effectiveness and achieved remarkable success, largely because of the abundance of well-labelled datasets. However, acquiring high-quality labels for polarimetric synthetic aperture radar (PolSAR) images is both time-consuming and expensive. Over the past few years, semi-supervised learning has gained attention as an effective approach to reduce dependency on labelled samples by utilising a combination of both labelled and unlabelled data. This paper proposes a novel semi-supervised method that utilises a hybrid CNN architecture with cross-pseudo supervision (CPS). This method transforms the PolSAR coherency matrix into two forms and trains two distinct CNN models to handle these different inputs,which addresses two critical challenges in semi-supervised PolSAR classification: (1) preserving phase coherence in complex-valued data to resolve ambiguities in terrains with overlapping magnitude responses and (2) mitigating pseudo-label noise propagation in label-scarce scenarios. This dual innovation enables robust classification in label-scarce PolSAR applications. Experimental results obtained from two different datasets validate the effectiveness of the suggested approach. The overall accuracies of two datasets are 99.19% and 96.01% with using 0.8% and 0.08% training samples per class.
To promote the development and application of intelligent robots, robots. Combining predictive control and whole-body control, the motion posture of the quadruped robot is controlled and adjusted. The adaptive adjustment algorithm and visual technology are used to improve the motion control of the climbing robot. According to the simulation results, the error rate of the quadruped robot's ground walking was 7.14% during the adaptive walking process from and climbing planning, the average pitch angle of the robot walking on stairs was 0.36 rad, and the error rate of walking on slopes was 5.42%, meeting the requirements for stable motion of quadruped operation of robots.
