Abstract:With the rapid development of intelligent sensing and new generation information technology, digital twin technology is leading the transformation of obstetrics. This article provides an overview of the development history, concepts, characteristics and related technologies of digital twin. At present, there is no universal understanding of the concept of digital twin among various institutions and scholars, and extensive discussions and research are needed in the future. It summarizes the application of digital twins in aerospace in the United States, and elaborates on the current application status of digital twin technology in the design, production, and operation stages of domestic aerospace. The current main problem is that there is still no effective connection between virtual and real entities, and so the direct guidance and optimization of physical entities by virtual models have not been truly achieved. It also summarizes the technical problems that still need to be continuously solved in the application of digital twins in aerospace, including sensing and data processing, high fidelity models, software platform construction, integration with new generation technologies such as artificial intelligence, etc. The current application of digital twin technology in aerospace is still in its early stages, and in the future, it is necessary to actively expand its application scope and scenarios in order to maximize the value and role of digital twin technology.

Abstract:Accurately predicting Alzheimer’s disease(AD) progression is crucial for timely treatment and intervention before advanced stage of AD. In this paper, a method on AD prediction based on joint decision-making utilizing dual priority prediction hierarchical model is proposed, which converts the three category prediction problem on AD, mild cognitive impairment(MCI) and normal cognitive(NC) into two levels of two category prediction problem. Firstly, the statistical features are extracted from the time series data of magnetic resonance imaging(MRI) and cognitive scores(CSs), which is obtained from individual historical followup, and the high-importance MRI volume statistical features are selected using the cumulative weighted embedded feature selection method. Then, both the NC priority prediction hierarchical model and the AD priority prediction hierarchical model are constructed. Using the selected high-importance MRI volume statistical features and CSs statistical features, these two hierarchical models are used to achieve AD/MCI/NC three category prediction. The NC and AD individuals are first predicted, and finally the MCI individuals are determined. The proposed AD prediction method is evaluated on the TADPOLE dataset. The accuracy(ACC) and macro average of F1 score of the proposed AD prediction method are 89.29% and 88.81%, respectively. The experimental results show that the proposed method is effective and better than conventional AD prediction method.

Abstract:To address the technical challenge of maintaining a constant bandwidth during the center frequency tuning process in traditional tunable filters, a new balanced bandpass filter with adjustable center frequency, constant bandwidth and high frequency selection performance is designed. Based on the electromagnetic coupling between the resonators, three coupling structures are formed through four tunable oscillators, and the coupling between the source and the load on the feed circuit is introduced, so that two transmission zeros are generated on each side of the filter passband. Using only the same DC bias voltage to load on the transformer diode, the center frequency of the filter is adjustable. Utilizing odd-even mode analysis theory and the target curve of coupling coefficient and external quality factor varying with the resonance frequency, the design objectives of constant bandwidth, low insertion loss and good out-of-band suppression are achieved. The simulated and measured features of the filter are in good agreement, with an adjustable central frequency range from 788 to 978 MHz, while sustaining a 3 dB bandwidth of (43±1.2 MHz) throughout this range. Compared to existing technologies, this design effectively reduces device size through optimized folding of resonator structures while maintaining constant bandwidth, providing a feasible solution for the realization of high-performance tunable filters.

Abstract:In order to address the issues of low trajectory planning efficiency and easy to fall into local solutions of unmanned agricultural vehicles in complex and narrow unstructured environments, an improved bidirectional A* algorithm combined with optimal control method is proposed in this paper. Firstly, the direction-guided search is introduced and the heuristic function is improved to accelerate the speed of bidirectional A* path planning in large-scale complex environments. Additionally, a path smoothing strategy is designed to reduce the number of inflection points and improve the quality of the reference path. Next, to address the challenge that the difficulty of handling obstacle avoidance constraints in optimal control problems increases significantly with the density of obstacles, safe driving corridors are constructed to reduce the impact of environmental complexity on computational efficiency. Finally, a penalty iteration framework is established based on the vehicle’s nonlinear kinematic model to solve optimization problems iteratively, thereby improving the success rate of trajectory planning and obtaining globally optimal or approximately optimal trajectories. In three different scale map simulations, the results show that compared with A* algorithm, the proposed improved bidirectional A* algorithm reduces the planning time and path length by 48.0% and 5.2%, and the path is smoother. In the unmanned agricultural vehicle trajectory planning, compared with Hybrid A* algorithm, variant 1 and variant 2, the proposed method reduces the cost of generated trajectory by 19.3%, 5.4% and 33.1%, respectively. The trajectory quality has obvious advantages, and provides an effective solution for practical application.

Abstract:A key factor in high-resolution synthetic aperture radar imaging is that the radar operates under ideal conditions. However, the motion trajectory of radar is usually not an ideal straight line or stable, so any small motion error within the synthetic aperture time can cause image blur or distortion. For small-scale imaging scenes, due to the susceptibility of GPS to signal interference and multipath effects, the traditional motion compensation method that combines GPS and INS data is not very effective. In this scenario, this study proposes a terahertz SAR imaging motion compensation method that only utilizes inertial navigation information. This method fully utilizes the velocity information provided by the inertial navigation system, establishes a radar motion trajectory model, and effectively estimates the echo phase error in the radar line of sight direction, thereby achieving focusing on terahertz SAR imaging targets. The experiment used a SAR system with a center frequency of 0.2 THz for motion compensation, and analyzed the strong scattering points of SAR images before and after compensation. Compared with the existing technology based on GPS and INS joint motion compensation methods, the motion compensation method proposed in this study respectively improved by 0.7 and 0.8 dB on PSLR and ISLR. In terms of imaging speed, the motion compensation method proposed in this study also improved by 0.2%. The experimental results showed that the focusing effect of this method was better for small-scale imaging scenes, further verifying the correctness and effectiveness of the motion compensation algorithm mentioned in this study.

Abstract:A detection method based on multi-scale inverse correction enhancement and lossless downsampling is proposed to improve the detection of hidden targets in millimeter wave images with low local signal-to-noise ratio. Firstly, a multi-scale reverse correction feature enhancement module was designed, which integrates the reverse correction operation on the Res2Net multi convolution kernel. This achieves the reverse correction of convolution calculation between large receptive field regions and related small receptive field regions, enabling finer-grained features across multiple scales. Secondly, utilizing non-step convolutional layers of SPD-Conv to achieve lossless downsampling and preserve more information. Finally, the K-means++ clustering algorithm generates new anchor boxes suitable for hidden object detection tasks. The experiment selected YOLOv5s, which has moderate performance in all aspects, as the basic framework, targeting two existing millimeter wave image datasets (array image dataset and line scan image dataset) mAP@0.5 reaching 96.21% and 97.97% respectively. Compared to the original YOLOv5s and other YOLO series, the performance has significantly improved. The experimental results show that this method can effectively improve the detection performance of deep models without significantly increasing the number of parameters and inference time.

Abstract:Aiming at the phenomena of leakage, false detection and low detection accuracy in complex scenes such as dense targets, small targets in distant view, etc., which occur in the helmet detection algorithm for two-wheeled vehicles, an improved RT-DETR helmet detection algorithm for two-wheeled vehicles is proposed on the basis of RT-DETR-r18. A dual cross-stage multi-scale feature fusion module (DcspBlock) is designed, and a multi-core initialization module (PKIBlock) is incorporated into the cross-stage module, which reduces the number of model parameters while effectively enhancing the network’s ability to capture targets of different scales in the near and far scenes; a small target detection module Decoderhead-p2 is introduced, which effectively enhances the model’s ability to detect small target detection accuracy; in order to alleviate the positive and negative sample imbalance and the inaccurate positioning of the bounding box in complex detection scenarios, the original model’s GIOU is replaced by the improved loss function MPD_Focaler-IOU, and the computation of the IOU is improved by setting the threshold parameter, so as to minimize the impact of the positive and negative sample imbalance on the model’s performance, and the computation of the minimum vertical distance is introduced to enable the bounding box to be finely localized. which makes the bounding box have better performance in fine localization. The experiments show that on the TSHW dataset, the improved RT-DETR improves the mAP@0.5 by 3.6% and reduces the number of parameters by 17.6% compared with the original model, while keeping a smaller computational volume, indicating that the improved model can effectively enhance the performance of two-wheeled vehicle helmet detection in complex scenes.

Abstract:The time-dependent features and context information of speech signals are crucial in speech enhancement tasks. Aiming at the problem that codec networks insufficiently capture these features, resulting in poor enhancement performance, an asymmetric residual dilatation convolutional and gated codec network (RD-EGN) is constructed. The network comprised three parts: the encoder, intermediate layer and decoder. The encoder designed a causal convolution layer structure to model the temporal feature, capture the features of different layers in the speech sequence and maintain the speech signal’s causality. The intermediate layer incorporated a residual dilated convolutional network (RDCN), which integrated dilated convolution, residual connections, and cascaded expansion blocks to endow the network with a larger receptive field. It facilitated cross-layer information transfer and extracted long-term dependency features in speech. The RDCN is combined with the long short-term memory network to capture broader context information. The decoder introduced a gating mechanism to adjust the gating degree of information flow dynamically, obtain richer global features and reconstruct enhanced speech. Ablation and performance comparison experiments were conducted on the TIMIT,UrbanSound8k,VoiceBank,and NOISE92 datasets. The results show that, RD-EGN has fewer training parameters and higher scores in SSNR and subjective evaluation metrics (CSIG, CBAK, and COVL) than CRN, AECNN and DDAEC. In objective evaluation metrics, the PESQ is increased by 2.5% to 7.1%,and the STOI is increased by1% to 5.3%. RD-EGN demonstrates outstanding enhancement performance and generalization ability.

Abstract:To address the issues of inaccurate target localization, missed detections, and false detections caused by large scale differences, diverse categories, and uneven distribution of targets in remote sensing images, an improved YOLOv8n remote sensing image target detection algorithm is proposed. Firstly, the SC_C2F module is constructed as the feature extraction module of the backbone network. By introducing spatial channel reconstruction convolution into the Bottlececk structure, the feature extraction ability of different scale channels and spaces is enhanced; Secondly, design an ESPPM module to replace the original pyramid pooling module, introduce an adaptive average pooling layer and a large separable kernel residual attention mechanism, enrich contextual information, and improve the model’s multi-scale feature aggregation ability; Again, by combining GSConv lightweight convolution with VoVGSCSP structure, the Slim PAN structure is introduced into the neck network to reduce model computation while maintaining detection accuracy; Finally, a rotation box with added parameter representation is introduced as the angle coordinate regression, and an RBCL loss function is designed to calculate the rotation box loss, making the detection process more in line with the target shape and improving the detection accuracy for small and dense targets. The improved YOLOV8n algorithm will be tested on the DOTA dataset and compared to the original algorithm mAP@0.5 Increase by 5.1% and reduce computational load by 0.4 GFLOPs.

Abstract:With the continuous increase in the transmission rate of high-speed data transmission lines, how to use the S parameters familiar to microwave circuit engineers to characterize the time-domain eye diagrams familiar to digital circuit engineers has become a problem that measurement engineers must solve in the context of impedance matching and bandwidth issues for high-speed digital transmission lines. This paper designs a measurement and characterization method for high-speed data transmission lines, which uses a multi-port vector network analyzer to measure and characterize the transmission characteristics of high-speed data transmission lines, establishing a direct correspondence between the S parameters familiar to microwave circuit engineers and the eye diagrams familiar to digital circuit engineers. By using a multi-port vector network analyzer to measure the frequency-domain S parameters of multi-channel data transmission lines, including the reflection S parameters, transmission S parameters, and crosstalk S parameters of the data transmission lines, the frequency-domain modeling of multi-channel data transmission lines is achieved. Through the inverse Fourier transform of the frequency-domain reflection S parameters, transmission S parameters, and crosstalk S parameters of multi-channel data transmission lines, the time-domain reflection, transmission, and crosstalk impulse responses of multi-channel data transmission lines are obtained. Assuming that an ideal error-free digital signal is input to a nonideal multi-channel data transmission line, the output digital signal of the multi-channel data transmission line can be obtained through the convolution of the input digital signal and the impulse response, thereby obtaining the eye diagram familiar to digital circuit engineers and achieving the time-domain modeling of multi-channel data transmission lines.Taking the frequency domain and time domain performance characteristic measurement and characterization of HDMI data transmission lines as the application scenario, a test fixture for HDMI data transmission lines was fabricated. The time domain reflection, transmission and crosstalk eye diagrams of HDMI data transmission lines were measured, and the experimental results were presented. From the measured time domain eye diagrams, it can be seen that as the transmission rate increases, the transmission quality of HDMI data transmission lines deteriorates. When the output rate is 5 Gb/s, the time domain eye diagram is relatively clear, and the digital transmission line can complete the transmission task well. When the digital transmission rate is 10 Gb/s, the time domain eye diagram is already blurred and unclear, and cannot complete the digital signal transmission. This verifies the effectiveness and accuracy of the data transmission line measurement and characterization method. The traditional eye diagram measurement of digital storage oscilloscopes and sampling oscilloscopes requires separate measurements of the input and output eye diagrams, followed by manual comparison to determine the signal transmission quality of the transmission line. The measurement and characterization method proposed in this paper directly measures the reflection, transmission, and crosstalk eye diagrams of digital transmission lines, directly reflecting the signal transmission quality of the transmission line. This solves the measurement and characterization difficulties of high-speed digital transmission lines and is of great significance for the measurement and characterization of digital transmission in fields such as high-speed digital communication, computing power networks, and mobile communication.

Abstract:Aiming at the problems of high subjectivity and low efficiency of the current traditional diagnosis of malocclusion, the article designs GraphTeeth, an automated diagnosis method of malocclusion based on graph neural network. GraphTeeth combines the architectural advantages of graph neural network, which is able to efficiently capture the topological information of the teeth and their surrounding structures. By modeling the position, morphology, and interrelationships of teeth as graph structures, finer local and global features are learned using a message passing mechanism between nodes. In the experimental phase, a large dataset containing various types of malocclusion cases was used to train and test GraphTeeth.The experimental results show that GraphTeeth significantly outperforms existing target detection methods in the key performance metrics. On the mAP metric, GraphTeeth achieves 43.45%, which is a significant improvement over traditional target detection algorithms such as Mask R-CNN at 32.26%, EfficientDet at 38.73%, and DETR at 25.05%. In addition, for specific types of malocclusions-such as fixed orthodontic appliance fitting-GraphTeeth achieves an accuracy of 91.28%, while the recognition rate for healthy teeth reaches 83.91%. The results suggest that GraphTeeth is able to provide faster, more accurate and objective diagnosis, providing reliable support for orthodontic treatment.

Abstract:Aiming at the problems of large size, poor port isolation and weak anti-interference ability of multi-band multiple-input multiple-output (MIMO) antennas, a triband MIMO antenna which can be applied to 5G n78/n79 and WiFi 6E bands is designed. The antenna is composed of a slotted diamond patch and a trapezoidal floor, and combines the innovative design of a semi-circular structure and a symmetrical inverted L-shaped branch. This design not only realizes the required triple-frequency characteristics, but also effectively controls the size of the antenna to adapt to more compact application requirements. The antenna is fed by coplanar waveguide (CPW), which has the advantage of easy integration with other microwave circuits. By placing the unit antenna orthogonally and without isolating branches, the port isolation of the MIMO antenna in the required frequency band is greater than 25 dB. The simulated antenna is processed and tested. The measured results show that when the return loss is less than-10 dB, the impedance bandwidth of the antenna is 3.28~3.67, 4.63~5.01 and 5.67~7.65 GHz, which is suitable for 5G n78/n79, WiFi 6E band. The maximum measured gain can reach 4.7 dB, the envelope correlation coefficient (ECC) is less than 0.001, and the diversity gain (DG) is greater than 9.999 9. The diversity performance is good. The measured results are highly consistent with the simulation results, which verifies the effectiveness and accuracy of the design. Considering the size advantage, good isolation and excellent diversity performance of the antenna, the triple-band MIMO antenna proposed in this paper shows great application prospects in 5G and WiFi 6E communication systems, which can meet the growing demand of future wireless communication and promote the development and innovation of wireless technology.

Abstract:Remote sensing images of ships is characterized by small target sizes, complex backgrounds, and significant attitude changes. Traditional ship detection algorithms focus on improving detection accuracy while neglecting model size and real-time performance, thereby limiting their practical application on resourceconstrained devices. A lightweight rotated fusion detection network RFDNet adapted to remote sensing ship images is proposed to address the above problems. Considering that the remote sensing ship images are taken at a long distance, resulting in small target sizes and rich background information in the images, ACFNet is designed to improve the detection accuracy of small ship targets by fully extracting local feature information and global spatial sensing information. To avoid accuracy degradation when detecting ship targets with different attitudes, a rotating bounding box loss function is introduced, which utilizes the orientation information of rotating targets for obtaining a more accurate bounding box regression loss, thereby improving the detection performance of ship targets rotating in any direction; for the problem of increasing parameter counts brought about by increasing the accuracy of the model, a lightweight convolution is introduced into the feature fusion part, which combines the convolution, the depth separable convolution, and the channel blending to reduce the number of parameters in the model. Through comparative and ablation experiments, it has been demonstrated that RFDNet achieved mAPs of 97.63% and 81.63% on the HRSC2016 and DOTA datasets, respectively, while reducing the model parameters to 2.99×106. This not only effectively improved detection accuracy but also realized a lightweight model design, providing a new insight for the application of remote sensing ship detection algorithms to resource-constrained devices.

Abstract:Electromagnetic pulse eddy current detection technology is widely used in metal pipeline defect detection because its transient response signal has rich time domain characteristics. However, the rapid attenuation phenomenon of pulse electromagnetic field in space makes the transient response signal susceptible to noise interference. In order to improve the signal-to-noise ratio of the transient response signal in pulsed eddy current testing and realize the pipeline defect detection, this paper optimizes the parameters of the detection coil and proposes a method for metal pipeline wall thickness corrosion defect detection. Firstly, the influence of excitation coil on transient response signal was analyzed by finite element simulation software, and the parameters of excitation coil were optimized. Then, the detection effect of the detection coil on metal pipe defects before and after optimization was studied, the amplitude of the detection signal is increased by 4 to 5 times after optimization. The relationship between the defect detection sensitivity of the transient response signal and the detection signal time was also analyzed, and the defect detection method was given. Finally, through the established pulse eddy current metal pipeline defect detection experimental platform, the defect detection method was verified on the metal pipeline with annular defects. The experimental results show that the defect detection curve obtained by the proposed method has preferable signal-to-noise ratio and defect detection sensitivity at the previous time of the transient response signal, and can be used to evaluate the wall thickness thinning of metal pipes.

Abstract:Due to the limited storage and computing resources of underwater detection instruments, using high-parameter models will occupy more storage space and computing power. To solve this problem, a target detection model based on YOLOv8n, called YOLOv8n-MAL, was proposed, which achieves lightweight overall while maintaining detection accuracy. Firstly, a lightweight multi-scale convolutional block attention module (MSCBAM) is proposed, which can extract features of different scales from input feature graphs through convolution kernel and pooling operations of different sizes. It can improve the robustness of the model in complex scenarios and ensure that the model can maintain high detection accuracy in the face of different types of inputs. Then, a new neck model multi-scale feature fusion network (MSFFN) is designed. Compared with the original neck model, the multi-scale fusion capability of the model is enhanced, and the interaction of features at different levels is strengthened, so that the high-level and low-level features can be more fully combined. This cross-level fusion can make use of the feature representation ability of the network more effectively, avoid the loss or weakening of information in the process of transmission, and improve the detection effect of the model. Secondly, a lightweight multi-scale convolution module lightweight multi-scale convolution module (LMSCM) is proposed, and the module and some convolution modules are integrated into the C2F module to form PC2F-LMS. By introducing a more efficient convolution structure. Finally, the original network loss function is optimized using WIoU. The experimental results show that the average accuracy mAP@0.5 of the improved algorithm on the URPC dataset is increased by 1.4%, and the number of parameters is reduced by 38.6% compared with the YOLOv8n algorithm, which provides an effective reference value for underwater target detection.

Abstract:The surface defects of cotton fabric directly determine the quality of the fabric. To address the problems of false detection and missed detection due to the significant scale variations and weak small defects in cotton fabric defect detection tasks, YOLO-CFD, a cotton fabric defect detection network based on YOLOv8s is proposed. Firstly, in order to better adapt to the scale changes of defects, a new module named BRASPPF is designed based on the Bi-Level Routing Attention mechanism; Secondly, in order to improve the feature extraction and localization ability of weak small targets, space to depth convolution blocks replaces partial convolution, and a small target detection layer is added in the neck feature fusion stage; Finally, in order to reduce the sensitivity of IoU to position shift, the NWIoU loss function is designed as the bounding box regression loss function. The experimental results show that the YOLO-CFD network model can achieve mAP@0.5 of 87.2% on the self-made cotton defect dataset, an increase of 16.5%, and the speed can meet the real-time detection requirements of industry. In addition, in the visualization experiment, the YOLO-CFD network model demonstrated a more comprehensive multi-scale feature extraction capability, which can detect weak small defect targets such as knots, splice and stains with only 12 pixels, and more accurately focus on slender global defect features such as broken end and holes. Compared to other mainstream object detection algorithms, the proposed algorithm has higher defect detection performance and can provide effective exploration for cotton fabric defect detection.

Abstract:Aiming at the problem of the high failure rate of electric vehicle charging piles, a fault prediction method of electric vehicle charging piles based on cooperative game strategy and dung beetle optimization algorithm-bidirectional long-term and short-term memory network-attention mechanism (DBO-BiLSTM-Attention) is proposed. Firstly, abnormal values are processed through parameter statistical distribution, missing values are processed through mean imputation, and the processed data is normalized. Secondly, from different perspectives, subjective evaluation methods such as analytic hierarchy process, objective evaluation method CRITIC weighting method, and machine learning algorithm random forest are selected to calculate feature weights in sequence. Cooperative game strategy is used to combine the above feature weights to obtain new feature weights, and the parameter feature matrix is enlarged. Then, the beetle optimization algorithm and attention mechanism were introduced separately to build the DBO BiLSTM Attention model. Under simulation experiments, the accuracy and F1 coefficient of the training and testing sets of this model were 0.89, 0.89, 0.90, and 0.90, respectively. Finally, relevant comparative experiments were conducted, and the results showed that compared to the model without feature amplification, the accuracy and F1 coefficient of the test set were improved by 5% and 6%, respectively; Compared with the model that does not adopt cooperative game strategy, the accuracy and F1 coefficient of the test set have been improved by 2% and 3% respectively, verifying the effectiveness and rationality of the proposed model.

Abstract:The proposed system is committed to solving the full-view monitoring problem faced by port portal cranes in large-scale operations. Considering that traditional monitoring systems often fail to adequately cover the complexities of port environments, there are significant shortcomings in addressing the comprehensive monitoring needs of portal cranes and bulk cargo yards. To achieve this goal, the proposed system utilizes a self-developed visual radar fusion system in conjunction with Unity3D and three-dimensional real-time color point clouds. It introduces a digital twin mapping method for bulk cargo yards and portal crane operations within the port area, driven by real-time laser point clouds and operational data from the portal crane. This method accurately simulates and presents the real-time conditions of the port’s bulk cargo yard and the operational status of the portal crane by collecting and processing laser point cloud data and crane operational information in real-time. Additionally, the system achieves a comprehensive three-dimensional visualization of the bulk cargo yard in the port area. This capability allows users to dynamically explore and switch real-time perspectives. The system also includes a robust equipment information management function capable of real-time monitoring the operational status of each portal crane. It supports real-time monitoring of both single-machine and multi-machine operations, thereby enhancing the efficiency and safety of port operations. The experimental results indicate that the system’s overall efficiency is improved by approximately 13.88% compared to the 3D monitoring system that is directly connected to the database. The developed system exhibits excellent real-time performance, accurately reproducing unmanned operations in large-scale port bulk cargo yards, thereby laying a solid foundation for digital port management.

Abstract:The design of regenerative braking strategies requires a comprehensive consideration of multiple factors, among which vehicle driving conditions, driver characteristics and the road surface on which the vehicle is traveling have a significant impact on the regenerative braking process. In order to formulate regenerative braking strategies for electric vehicles that are adaptable to various driving conditions, improve the vehicle braking energy recovery rate and maintain braking stability, a regenerative braking strategy that comprehensively considers the influences of driving cycles, drivers and road information is proposed. Firstly, a simulation driving platform is set up to conduct driver-in-the-loop experiments and collect driving data from different drivers, thereby extracting feature parameters of driving conditions and driving styles. Then, a support vector machine (SVM) is used to train the models for identifying driving conditions and driving styles. Secondly, a road image dataset is established and a semantic segmentation network is used for road image preprocessing to remove the complex background information of the image and thereby improve the recognition efficiency. Then, a lightweight conuolutional neural network, MobileNet V3, is adopted to train the road recognition model. Finally, the regenerative braking strategy base on this is formulated. The front and rear braking force distribution is optimized considering the road adhesion conditions, and a regenerative braking force correction method that takes driving cycles, driver and road information as weight factors is put forward. The simulation results show that the proposed regenerative braking strategy can take into account different driving cycles, drivers and road conditions, and further improve the vehicle energy recovery rate and braking stability.

Abstract:Aiming at the contradiction between the large number of parameters, high computational complexity and realtime requirements of defect detection models in industrial embedded scenarios, a CCS-YOLO lightweight defect detection network is proposed to be constructed by CSPPC module, CCFM module and SA_Detect fusion module. Its lightweight performance is verified by designing ablation experiments and comparative experiments. In order to enhance the feature extraction and expression capabilities when processing complex visual tasks and combine partial convolution operations to optimize the performance and efficiency of the model, the CSPPC module is used. The CCFM module is used to fuse features of different scales to improve the model’s adaptability to scale changes and the ability to detect small-scale objects. The SA_Detect module that fuses shared convolutions is used to further reduce the number of model parameters and achieve model lightweight, which effectively improves feature expression, target positioning and classification performance. The experimental results show that compared with YOLOv8n, the model size, computational complexity and weight parameters of the CCS-YOLO model are reduced by 56.7%, 51.9% and 54.0% respectively, with a significant lightweight effect.The detection speed is maintained above 34 fps when deployed on the RK3568 embedded platform, and the real-time performance is verified, which is practical and efficient. It can be seen that the application cost-effectiveness of the system has been improved, effectively overcoming the shortcomings caused by a slight decrease in accuracy. The constructed defect detection network CCS-YOLO can solve the problem of resource constraints in industrial embedded scenarios and realize a feasible solution for low-computing power devices to achieve high performance, real-time and lightweight, which has important engineering value.

Abstract:In complex environments characterized by multiple obstacles, the traditional A* algorithm in path planning presents the problem of redundant turning nodes. This not only increases path length and complexity but also hinders the smooth navigation of the AGV. To address these challenges, this study introduces an improved A* algorithm predicated on the tensile mechanism of a flexible rope, aimed at diminishing path nodes and augmenting trajectory smoothness. First, the mechanism of flexible rope stretching was analyzed, and critical nodes were extracted from the paths generated by the A* algorithm. Subsequently, the degeneration of non-obstacle force points was executed to minimize redundant steering nodes, followed by the sequential stretching of paths between force points, thereby streamlining the trajectory and enhancing smoothness. Ultimately, the refined A* algorithm underwent simulation experiments and was applied to AGVs for autonomous navigation path planning experiments. The simulation outcomes demonstrated that the A* algorithm, refined with the flexible rope stretching mechanism, achieved a 59.2% reduction in turning angles, a 54.2% decrease in the number of turning points, and an 11% reduction in path length, significantly simplifying and smoothing the trajectory. In the AGV navigation experiments, the optimized A* algorithm, when compared to the traditional A* algorithm, registered a 16% decrease in average angular velocity and a 33% reduction in driving turning angles, with average travel trajectory length and time reduced by 2.4% and 4%, respectively. Additionally, the average travel trajectory length and time spent are reduced by 2.4% and 4%, respectively. The experiments results show that the AGV experiences smaller node transformations and posture adjustments while following the paths planned by the improved A* algorithm, leading to smoother and more efficient movement.

Abstract:To address the issues of uneven ORB feature point extraction, slow matching speed, and low matching accuracy in traditional image processing, this paper proposes an improved image processing algorithm for ORB feature extraction and matching. First, an enhanced quadtree algorithm is employed to achieve uniform extraction of ORB feature points, maximizing the utilization of information across the entire image. Second, brute-force matching combined with GMS screening is implemented to preliminarily filter feature matches, thereby enhancing matching accuracy. Finally, a dynamic Bayesian network is utilized to select the optimal matching model and best-matched pairs, further improving matching accuracy while reducing screening time. Experimental results demonstrate that compared with traditional algorithms, the proposed method significantly improves feature distribution uniformity. The average time required for feature extraction and matching is reduced compared to other improved algorithms, while achieving higher matching accuracy—specifically showing a 49.1% improvement over conventional ORB algorithms. The overall performance surpasses both traditional ORB and other existing improved algorithms. It is demonstrated that the image processing based on improve ORB feature extraction and matching can effectively achieve simultaneous improvements in both feature matching accuracy and feature matching speed.

Abstract:Liquid concentration detection is widely used in food, environment, biomedical and other fields. Based on the microwave method, a resonant coaxial probe liquid concentration detection device was studied, designed and realized. Firstly, ethanol-water solutions of different concentrations are selected as test samples, and the effects of coupling gap, probe immersion depth and probe conductor material on detection sensitivity are simulated and analyzed. To verify the feasibility of liquid concentration detection using the designed probe, ethanol-water solutions/glucose-water solutions with a volume concentration of 0%~75%/0%~50% are measured. The experimental results show that the probe is capable of accurately measuring the liquid concentration and the detection sensitivity in different concentration ranges can be optimized by adjusting the coupling gap. In addition, in this study, the quantitative inversion model of solution concentration is constructed by combining three electromagnetic parameters, namely, resonance frequency, S11 amplitude minima and quality factor. Compared with traditional method that only use resonance frequency as indicator, relative errors of liquid concentration of ethanol-water solutions/glucose-water solutions are suppressed from 5.79%/3.34% to 2.19%/1.36%, respectively. The probe is also able to effectively differentiate a variety of transparent liquids, such as ethanol solutions, glucose solution, salt water, tap water and deionized water, etc., showing good recognition ability and data reproducibility and thus shows a wide range of potential applications.

Abstract:Aiming at the problem that vibration signals collected in gear fault diagnosis are often accompanied by noise interference and fault features are difficult to extract, based on Fourier-Bessel series expansion (FBSE). A noise reduction method of gear vibration signal, which combines FBSE and energy scale space empirical wavelet transform (ESEWT), is proposed. Firstly, the frequency spectrum of the acquired gear vibration signal is obtained by using FBSE technology to replace the traditional Fourier spectrum. Then, the obtained FBSE frequency spectrum is adaptive segmented and screened by using the energy scale space partitioning method to accurately locate the boundary points of the effective frequency band. Then the signal components are obtained by constructing wavelet filter banks and reconstructed to reduce noise and redundant information interference. Then, in order to capture more comprehensive feature information, the processed signal is transformed by generalized S-transform to obtain time-frequency graph, and 2D convolutional neural network is input for fault diagnosis to verify the feasibility of the algorithm. Through experiments on Simulink simulation signals and actual acquisition signals, the results show that compared with the original EWT, EMD and other methods, FBSE-ESEWT has better noise reduction effect, the signal-to-noise ratio is increased by 13.96 dB, and the diagnosis accuracy is up to 98.03%.

Abstract:Due to the harsh working environment and complex working conditions, the hydraulic pump is often in the working state of variable load, severely challenging its condition monitoring and fault diagnosis. However, the existing model-based and data-driven methods have some limitations in fault diagnosis, so a fault diagnosis method based on the fusion of mechanism model and data-driven is proposed. First, the virtual prototype model of the hydraulic pump is built, and the faults under different loads are simulated to obtain the simulation pressure signal. Then, the hydraulic pump is tested for fault, and the experimental pressure signals of load and fault state corresponding to the simulation signal are collected. Following that, the variance of simulation and experimental data is calculated according to the proposed variance weight fusion method, and the optimal weight calculated by variance is used to fuse the simulation and experimental data. Finally, the fusion data is input into the deep convolutional neural networks with wide first-layer kernels for fault diagnosis under single and mixed loads. Experimental results show that this method can significantly improve the accuracy of diagnosis, and the accuracy is 2.42% and 12.92% higher than that of single model-driven and data-driven diagnosis methods in the case of mixed load, which verifies the effectiveness and superiority of this method.

Abstract:Photovoltaic (PV) arrays often operate in complex and harsh environments, making them susceptible to various types and degrees of faults. To enhance the accuracy of fault diagnosis in such challenging conditions, this study proposes a novel fault diagnosis model based on feature extraction and an improved pelican optimization algorithm (IPOA) optimized support vector machine (SVM). Firstly, 15 typical fault states are simulated on the MATLAB/Simulink platform, from which a 12-dimensional fault feature vector is constructed. Kernel principal component analysis (KPCA) is then applied for feature fusion and extraction to improve feature representation capabilities. Secondly, to address the limitations of traditional pelican optimization algorithms in balancing global search and local exploitation, enhancements are introduced, including the Tent chaotic map, inertia weight, nonlinear convergence factors, and an adaptive t-distribution mutation strategy, all of which significantly improve the algorithm’s optimization performance. Finally, the IPOA is used to optimize the penalty factor C and kernel parameter γ of the SVM model, establishing the IPOA-SVM PV array fault diagnosis model, which is then validated through both simulation and experimental tests. The results show that, compared to the traditional 6-dimensional feature set, the proposed 12-dimensional feature set achieves higher diagnostic accuracy. The improved model demonstrates fault diagnosis classification accuracies of 98.55% and 97.93% for simulation and experimental data, respectively, significantly outperforming other comparison models and demonstrating higher accuracy in PV array fault diagnosis.