Abstract:To balance the charging efficiency and service life, lithium-ion batteries usually require constant current followed by constant voltage charging mode. A switchable series hybrid topology WPT system for charging lithium-ion batteries has been introduced in this paper. The system can be switched from constant current mode to constant voltage mode without real-time feedback from secondary sides and complex control strategies. Firstly, based on the T model, a theoretical analysis of the hybrid topology has been carried out, which deduced that the hybrid topology WPT system outputs load-independent constant current and constant voltage at constant frequency and zero phase angle. Secondly, a model has been built in the power electronics simulation software to analyze the operating characteristics of the two topologies in constant voltage mode and system efficiency. Finally, a WPT experimental platform with a charging current of 3.5 A and a charging voltage of 65 V has been built. The experimental results showed that the current fluctuation is less than 5.9% in constant current mode and the voltage fluctuation is less than 8% in constant voltage mode over a wide range of loads.

Abstract:In recent years, knowledge representation learning has played a crucial role in intelligent recommendation, intelligent question answering, and intelligent retrieval, and has received widespread attention. Knowledge representation learning aims to vectorize semantic information and infer knowledge through mathematical formulas by means of low-dimensional embedding of entities and relationships. Among many knowledge representation learning models, TransE is considered to be the most promising model due to its fewer scoring function parameters, low computational complexity and high computational efficiency. However, TransE has some limitations in dealing with complex relationships other than one-to-one. In order to solve this problem and improve the quality of knowledge embedding, this paper proposes an improved knowledge representation model TransREF based on translation model. Firstly, the embedding of entities and relations is realized by means of relation matrix projection. Secondly, on the basis of the original vector, the relational neighborhood is added to enhance the learning ability of the model. During the training of the model and for entities with high semantic similarity, the replacement of the head entity and the tail entity is realized by the probability method, and then the high-quality negative example triples are generated, and the five-point random method is used to select the relationship neighborhood nodes. Finally, the relevant link prediction experiment is carried out on the subset WN18 of WordNet and the subset FB15K of Freebase, and then the triplet classification experiment is carried out on the three public datasets WN11, FB13 and FB15K. The results show that compared with TransE and TransH, TransREF has better performance improvement in MeanRank, Hits@10, and ACC indicators, which proves the effectiveness of TransREF.

Abstract:Axial field permanent magnet memory motor (AFPMMM) is a new type of motor. The accurate calculation of iron loss is of great significance to the performance analysis of the motor. In order to obtain accurate AFPMMM iron loss, firstly, the core multi ring delamination model is established and the magnetic field at the typical location of the core is analyzed. Secondly, based on the classical loss separation model, the column coordinate system is introduced to decompose the magnetic density to build the iron loss calculation model. This model not only considers the influence of higher harmonics, but also improves the iron loss calculation accuracy through the hysteresis loss correction coefficient and eddy current correction function. Finally, the prototype is tested, and the results of the improved model and several calculation methods are compared with the measured values. The research results show that the calculation accuracy of the improved iron loss model at each speed is significantly improved, with an average deviation of 3.45% compared to the experimental results, which is closer to the measured value.

Abstract:To segment target regions in retinal blood vessel images more accurately, a network based on improved W-Net is proposed. The network uses diamond structure fusion for semantic feature aggregation by stacking the parts containing the diamond structure layer by layer to form a U-shaped widening framework and introducing nested dense jump connections to form the final model. The fusion scheme improves the flexibility of feature map combination, and the designed jump connections reduce the semantic gap between feature maps, thus reducing the learning pressure on the optimizer and achieving better image segmentation performance. The effectiveness of the proposed network is verified using the DRIVE dataset. The dice similarity coefficient values, sensitivity, specificity, and accuracy obtained by BW-Net in the segmentation task are 76.86%, 73.66%, 99.12%, and 94.55%, respectively, which perform better than the output of most of the current state-of-the-art network frameworks, and the network parameters are reduced. The results demonstrate the improvement of this extended structure in the performance of retinal vascular image segmentation.

Abstract:Aiming at the durability problem caused by the polarization reaction of UAV battery charging, and in order to improve the charging rate of UAV battery by the laser wireless energy transmission system, a P&O constant current pulse charging scheme was proposed. This scheme first uses P&O to track the maximum power point of GaAs thin film photocells, which can quickly and accurately track the current and voltage at the maximum power point, ensuring the energy supply stability of the drone during energy transmission; Subsequently, a constant rheological pulse charging scheme based on P&O was established by combining the output characteristics of P&O and unmanned aerial vehicle lithium batteries, which improved the polarization reaction of unmanned aerial vehicle lithium batteries, thereby enhancing their durability and overall charging rate. The simulation results show that the constant flow pulse charging scheme based on P&O can not only improve the durability of drone lithium batteries, but also reduce the charging time by 289 seconds compared to conventional constant current and constant voltage charging methods, and improve the charging efficiency by 9.45%.

Abstract:In order to meet the increasing bandwidth requirements of wireless communication, expand the impedance bandwidth and axial ratio bandwidth of circularly polarized microstrip antenna while maintaining the miniaturization of wireless equipment, this paper proposes a design of broadband circularly polarized microstrip antenna based on microstrip slot antenna. The overall size of the antenna is 30 mm×30 mm×1.6 mm, the top structure is an inverted S-shaped branch, and the eccentric side feeding mode is adopted. The impedance matching of the antenna is improved by loading a stub on the microstrip line, which expands the impedance bandwidth of the antenna, and the axial ratio bandwidth of the antenna is expanded by chamfering and trenching the ground plate. The measured results show that the -10 dB impedance bandwidth of the antenna is 62.1% (5.16~9.81 GHz), and the 3 dB axis specific bandwidth is 63.9% (4.43~8.59 GHz). The antenna has the characteristics of small size and wide band, which is suitable for the wireless communication system in the 5.8 GHz band and WiFi-6E band.

Abstract:In a high-speed satellite data transmission system, factors such as the Doppler effect will cause symbol timing offset between transmitter and receiver, it can be effectively corrected by the Gardner clock recovery algorithm. However, the existing implemental structure of clock recovery algorithm has disadvantages of high sampling rate and complexity, it is difficult to meet the requirements of higher symbol rate and timing offset tolerance because of limited sampling rate and hardware resources in the real-time system. Thus, this paper proposes a new parallel structure based on the Gardner clock recovery algorithm. By introducing add/delete state machine, sample adjusts module, and symbol extract module, it can realize fast symbol timing estimation when sampling at twice the symbol rate. Simulation and FPGA board-level tests show that this structure is suitable for multiple modulation formats of QPSK/8PSK/16APSK. It can tolerate timing frequency offset of up to ±400×10-6, and has a stable bit error rate in long-term tests. In addition, when implementing a real-time receiver system with 625 MBaud symbol rate, the parallel structure proposed in this paper saves about 37% of LUT resources and more than half of the Register and DSP resources compared with the traditional structure. It is practical and has great value in a real-time communication system with limited resources.

Abstract:To enhance the real-time performance and stability of the two-wheeled self-balancing control system, a mechanism combining quaternion and PID algorithms has been proposed for the control of the two-wheeled self-balancing vehicle. This mechanism is based on quaternion calculation to determine the attitude angles and uses the PID algorithm to achieve motion control of the balancing vehicle. The system measures the angular velocity and angular acceleration of the balancing vehicle using MPU-6050 and measures the motor rotation speed using encoders. These measurement data are used as feedback for the balancing vehicle control system, which enables the control of the vehicle and human-machine interaction through a mobile app. Under the control algorithm designed in this paper, the balanced vehicle can maintain good balance in three scenarios: self-balancing, self-balancing with high load and self-balancing with heavy load.In total, 20 000 sets of data of the pitch angle of the balanced vehicle in the three scenarios were collected, with variances of 0.013, 0.084 and 0.065 respectively. When compared to the Kalman filter and complementary filter algorithms in a stationary state, the pitch angle variance of the balanced car in the 5 000 sets of self-balancing data was 0.000 239, smaller than the other two algorithms. These experimental results show that the proposed control mechanism can achieve stable control of the balanced vehicle while satisfying the real-time requirements.

Abstract:In order to solve the problems of long period and low efficiency in microwave antenna design, a multi-objective microstrip patch antenna automatic design and optimization method combined with machine learning was proposed. In this paper, by using genetic algorithm to optimize the initial weights and thresholds of the neural network model, the optimized GA-BP model was used to predicts the multiple groups sets of antenna parameters on the resonance point of |S11|, the effective area below -10 dB and the corresponding reward value. Given the electromagnetic response of the target antenna, the geometric parameters of the antenna can be also predicted by the GA-BP model. The results show that the determination coefficient R2 predicted by BP model is about 0.968, while the GA-BP model proposed in this paper is as high as 0.994, which is significantly better than the traditional BP neural network model.

Abstract:The linear combination of various models can effectively avoid the disadvantages of poor convergence and low reliability for photovoltaic power forecasting. Simplifying the linear relationship between a single model in a linear combinatorial model can simplify the calculation of the combinatorial model, but reduce the prediction accuracy. Aiming at this problem, a prediction model based on Sparrow Search Algorithm (SSA) was proposed to optimize Bidirectional Long Short-Term Memory (BiLSTM) nonlinear combination method. Firstly, the Kernel-based Fuzzy C-means (KFCM) and Variational Modal Decomposition (VMD) are used to preprocess the original data samples. Then, using the Elman and SSA-BiLSTM forecast after photovoltaic (PV) power of pretreatment; Finally, the nonlinear combination of the two single models is optimized by the sparrow search algorithm to establish a nonlinear combination prediction model for short-term photovoltaic power. A comparative calculation example is established based on the measured data of a photovoltaic power plant, and the results showed that the average RMSE and MAE values of the proposed combined model under different weather conditions are 0.689 kW and 0.540 kW, respectively, which are superior to other comparative models, verifying the effectiveness and superiority of the proposed combined model.

Abstract:Photovoltaic models are both nonlinear and multimodal, and traditional algorithms are prone to fall into local optimality and insufficient recognition accuracy when identifying their parameters. In this paper, an improved artificial ecosystem optimization (IAEO) algorithm is proposed to balance exploration and exploitation by introducing a nonlinear control parameter adjustment strategy to enhance the exploration capability of the algorithm by exploiting the ergodic and non-repetitive nature of chaos. Simulation experiments show that the parameter identification accuracy of the improved algorithm exceeds 99.9% on both single, dual and triple diode and PV module models, and the RMSE value is improved by 5.5% on average on the four models compared to the original algorithm, which has a strong advantage compared to the five advanced algorithms. The improved algorithm still maintains high accuracy and stability in different environments when tested under different lighting and temperature conditions using real manufacturer data for three types of PV modules: thin-film, mono-crystalline and multi-crystalline.

Abstract:In view of the influence of large deflection and root connecting flange deformation on the bending moment and measuring accuracy of deformation during the static testing of wind turbine blade, calculation formulas of blade deflection and bending moment after deformation are obtained based on the flexibility method and the piecewise rigidization method respectively. Moreover, the influence brought by flange deformation is modified and the simulation model of wind turbine blade under multi-point static loading is established. Then, the theoretical calculation results are compared with the FEM analysis data. The results show that the bending moment error of the section near the blade root can be reduced to 1.07% considering large deflection when designing the loading scheme. After correcting the blade deflection caused by flange deformation, the calculation error of blade deflection at 5.7 m is reduced by 0.1%, so as to ensure the test accuracy of full-size blade under static loading.

Abstract:Deep learning has been widely used in environment perception of autonomous driving field，Its application on multi-object traffic sign (such as shield, overlap, incomplete, and small goals) detection on urban roads is current research emphasis. This paper proposed an improved detection method based on YOLOv3. Firstly, it introduced the depthwise separable convolution layer into the backbone network of YOLOv3 to optimize the parameters and reduce the quantity of calculation in convolution neural network. Secondly, the spatial channel attention mechanism (CBAM) was introduced after the residual module in the backbone network to, aiming to Enhance the ability of the network to extract weak feature information, and improve the detection accuracy of small target traffic signs. Finally, the optimized intersection ratio function of IOU combining with the CIOU function, can improve the target detection accuracy of the candidate box screening. The experiments were conducted with CSTSDB open source traffic sign dataset and partial self-built dataset. The experimental results show that the improved YOLOv3 network improves the accuracy rate has been improved by about 7% than the original YOLOv3 detection algorithm, and has a lower leakage detection rate, faster speed, which has some practical significance.

Abstract:Aiming at the problem that the jerk value is discontinuous in the traditional S-curve model, which affects the accuracy and service life of the equipment, a new trajectory planning and multi-axis synchronization algorithm are proposed. Firstly, an infinitely differentiable 15-segment jerk model is constructed based on the piecewise trigonometric function, then the time-optimal point-to-point trajectory satisfying the kinematic constraints is interpolated, and finally, the parametric expression of point-to-point multi-axis synchronization algorithm with minimum velocity is derived. The method is simple to calculate and can obtain analytical solutions, and at the same time, it can ensure that the generated curve is smooth. The simulation show that the proposed method can generate continuous smooth trajectories while increasing the total planning time by less than 5% compared with the conventional S-curve. Therefore, the correctness and effectiveness of the proposed method are verified.

Abstract:The L-N bus could be widely used in building electrical intelligent monitoring system as the characteristics of strong anti-interference, low application cost and autonomy. Aiming at the CPU overhead problem caused by the current software implementation of L-N bus communication protocol, a configurable L-N bus controller is designed to achieve the hardening of communication protocol. Through the in-depth analysis of L-N bus communication protocol, the feasibility of the protocol hardening is confirmed. On this basis, we established the hardware architecture, each component module and its relationship of the protocol controller. Then, we described function of each module in the behavior layer, simulated its logic function and implemented the prototype verification on the FPGA. Finally, we simulate the application scenario of the controller by the construction of the physical layer test platform and test its functions, including communication test, configurable function test and reliability test. The results show that the L-N bus controller could well implement the protocol specification, and shows good reliability. It can improve the CPU efficiency by 46% in the test environment.

Abstract:Aiming at the problem that rolling bearings are difficult to identify faults in noisy environments, a rolling bearing fault judgment method combining attention mechanism and Inception-ResNet is proposed. Firstly, a method combining grayscale image and pseudo-color processing is proposed to convert one-dimensional vibration signal into three-dimensional RGB image; then combined with Inception module and residual network, the network is expanded in both width and depth, and the network is improved. Finally, combined with the CBAM attention mechanism, the channel attention module and the spatial attention module are integrated to enhance the more important features of the input features and suppress unnecessary noise features, thereby effectively improving the diagnostic accuracy. In this paper, the bearing data set of Case Western Reserve University is used for verification, and several mainstream deep learning methods are selected for comparative experiments. The test results show that this method has a good diagnostic accuracy rate, the average accuracy rate is as high as 99.32%. The analysis experiment is carried out under the noise state, and the results show that the method still has a good accuracy rate under the noise state, which verifies the robustness of this method.

Abstract:To address the problem that small-scale receptive fields for wearable activity recognition tasks make it difficult to extract long range associations and that large-scale receptive fields lead to feature compression reducing the network′s resolution for signal features. In this paper, we propose a multi-scale channel attention mechanism based human activity recognition method. Firstly, temporal features and sensor channel features are extracted from multiple receptive fields, so that high semantic features and low semantic features are extracted at the same time to ensure high resolution of features. Secondly, cross channel association is established between multi-scale feature maps to obtain the interaction between low semantic features and high semantic features. multi-scale channel attention mechanism can fully integrate multi-scale features and correlation information of multiple feature maps, enhancing the recognition ability of weak signals and violent signals. The comparative experiments on the UCIHAR, DSADS, PAMAP2 and UniMib-SHAR datasets show that the classification accuracy of our method is improved by 0.43%, 0.75%, 2.90% and 0.83% respectively compared with the state of the art methods.

Abstract:To address the problem of motorists using cell phones and smoking behaviors during driving threatening traffic safety, this paper proposes an improved YOLOv7-based network model . Firstly, the MobileNetv3 backbone network is used instead of the original YOLOv7 backbone network to reduce the number of model parameters and computation and improve the processing speed of the model. The depth separable convolution and sub-pixel convolution are used to build an improved feature pyramid branch and fuse it with the output feature layer of the original feature pyramid to enrich the feature information and enhance the feature extraction effect. Finally, the feature enhancement module is finally used to enhance the fused feature layer to improve the attention of both the feature layer channels and regions. The experimental results show that the mean average precision of the improved network model is 95.33%, and the detection speed is 75.31 frames per second. Compared with the original YOLOv7 network, the mean average precision is increased by 6.84%, and the detection speed is increased by 17.25 frames per second. It has higher detection accuracy on the basis of satisfying real-time detection and can realize real-time and accurate detection of drivers′ use of cell phones and smoking behavior.

Abstract:In response to the problems of traditional road crack detection methods, such as time-consuming, laborintensive, high cost, and subjectivity, a YOLOv5-based road crack detection model, named YOLOv5-Crack, is proposed. Firstly, a coordinate attention mechanism is introduced in the backbone and optimized as a CA-plus structure to enhance the crack feature focus. Secondly, a novel feature fusion network ESPP is proposed to reduce some computational costs while improving the feature fusion capability. Then, the heavy Ghost-Shuffle convolution is used in the neck network to replace the traditional convolution, which can keep the channel semantic information as much as possible while reducing computational costs. Finally, the SIoU loss function is introduced to improve the regression accuracy. To validate the effectiveness of the improved YOLOv5-Crack model, comparative experiments are conducted on the GRDDC 2020 dataset, and the results show that the F1 scores are 58.43% and 58.21%, respectively, which are 4.05% and 3.93% higher than those of the original YOLOv5 model, and the computational cost is reduced by 7.8 GFLOPs, with an FPS of 37.9, effectively addressing the shortcomings of road crack detection. Furthermore, compared with mainstream object detection algorithms, the proposed YOLOv5-Crack model has superior performance in road crack detection.

Abstract:Rapid 3D measurement has been widely used in industrial testing and reverse engineering. In this paper, a 3D measurement method based on binary fringe phase solution is proposed. Firstly, by projecting a binary fringe image, the image carrying the phase information of the object is divided into two fringe masks by utilizing the fringe order distribution characteristics, and the white pixels are marked by steps using the connected domain, so as to obtain the fringe order. Secondly, due to the interference factors such as environmental noise, hopping error occurs during unwrapping phase. Therefore, a method is proposed to compensate the phase by integrating the half-periodic displacement order, and the complementary order of the half-periodic dislocation is obtained by using the binary fringe of unwrapping phase. Finally, the phase unwrapped are verified experimentally according to fringe order and complementary order. Experimental results show that the proposed method solves the burr problem in the phase unwrapping process. Taking the flat plate as the measurement object, the root-mean-square error measured by the proposed method is 0.198 0 mm, and only one image is needed to complete the phase unwrapping. The proposed method has good robustness and effectiveness, and can be applied to the field of rapid measurement.

Abstract:Aiming at the problem that the existing 3D reconstruction methods for pipe depend on target′s structure and texture characteristics, this paper proposed a bridge-type point cloud matching method based on cooperative objective and monocular vision to realize 3D reconstruction of pipe. The method does not depend on any structure and texture characteristics of the pipe, and has faster matching speed, and is not affected by the scale of the point cloud. At the same time, a high-precision cooperative objective based on concentric circle is designed, which is used to calculate the high-precision matching pose of the point cloud. The accuracy of the pose can reach 0.02 mm, which is 53% higher than that of the elliptical cooperative objective. Simultaneously, the robust recognition algorithm of the cooperative objective and the ranking algorithm for feature points with rotation invariance are realized. Finally, using ROS robot, a 3D reconstruction system of pipe without any features was built, and in actual experiments, the overall 3D reconstruction of the pipe with a matching accuracy of 0.02 mm was completed.

Abstract:To address the problem that the existing spatiotemporal downscaling deep learning methods are not enough to learn the spatiotemporal characteristics of radar precipitation images, a spatiotemporal separation based 3D deep learning model is proposed. The model takes Unet3d as the core architecture. A hybrid spatio-temporal separation convolution unit is designed to enhance the extraction of local spatio-temporal features of precipitation images, and a three-dimensional Swin Transformer is used to compensate for the loss of spatio-temporal features of precipitation images caused by traditional Unet3d encoder downsampling, so as to improve the effect of spatio-temporal downscaling forecast. The model was tested and evaluated through the open data set provided by METEO FRANCE. The results show that the designed hybrid spatio-temporal separation unit has a better ability to extract local spatio-temporal features, and the spatio-temporal separation based method can improve the spatio-temporal downscaling forecasting effect. Specifically, the 3DUST model proposed in this paper increased SSIM and PSNR evaluation indexes by 5.2% and 6.7%, respectively, and reduced the number of parameters by 3.2% compared with the comparison model.

Abstract:The efficient detection of the underwater biological resources in a complex natural environment is of great significance to China fishery. Aiming at the problems such as low recognition ability and serious feature loss of underwater biological resources in complex low-light environment, a lightweight underwater biological detection algorithm is proposed in this paper. First of all, aiming at the problems of large color deviation and low resolution of underwater images, a Dark channel-contrast limiting-optical attenuation algorithm is proposed to enrich the feature information of underwater images. Thereafter, GhostNet module and C3CA module are used to improve the fusion capability of feature extraction network. Finally, the loss function is improved to reduce the total loss freedom. The experimental results show that the mean average precision of the algorithm reaches 86.22%, which is 0.48% higher than that of the original YOLOv5L. Moreover, the volume of the proposed algorithm model is only 20.4 MB, which is about 89.31% less than that of the original model, and the detection speed of the proposed model is increased by 56.56%. The experimental results show that the improved algorithm achieves good results in underwater images and provides a guarantee for the real-time detection of underwater biological resources.

Abstract:Addressing the challenges of ore recognition in the mining industry and the high cost of recognition equipment, we propose an improved Unet ore image segmentation algorithm. Firstly, we modify EfficientNetV2 as the backbone network of the model to extract ore features. Secondly, we introduce the MBconv module as the decoder, enhancing the feature extraction capability. We then replace the SE attention module with the CA attention module to retain more spatial position information. Finally, we substitute the commonly used Batch Normalization (BN) layer with the Filter Response Normalization (FRN) layer to prevent model performance from being affected by batch size. Experimental results based on FeM and Cu datasets demonstrate that our proposed model achieves a PA of 96.58% and 95.39%, an MIoU of 92.8% and 90.43%, and an F1 score of 95.15% and 93.47%. Compared to Unet, the Efficient_Unet model parameters are reduced by 60.4%, and the inference speed is improved by 19.23%, reaching 21.7 frames per second. Our proposed model outperforms existing classical segmentation models in terms of accuracy and speed, exhibiting strong generalization performance.

Abstract:Most denoising networks only perform well in the task of synthetic noise denoising, and only extract features from a single scale, which can not better reconstruct a clean image. To solve the above problems, this paper proposes a blind denoising algorithm for real noisy images based on multi-scale feature fusion. The horizontal network structure of the algorithm uses adaptive dense residual blocks to extract rich features of the same scale, and selectively enhance features with large amount of information. The vertical network structure uses pyramid layer and encode-decode to further obtain different receptive fields to realize multi-scale feature extraction, The full fusion of the features of the same horizontal scale and the features of different vertical scales is more conducive to noise removal and retain the edge details of the image. The proposed network is evaluated on the real noise test set (DND and SIDD). The peak signal-to-noise ratio (PSNR) is 39.62 and 39.49 respectively, and the structural similarity (SSIM) is 0.956 and 0.954 respectively. The experimental results show that the network proposed in this paper has achieved better performance.