Abstract:The shrouded blades are widely utilized in significant rotating equipment such as aircraft engines due to their functions of enhancing blade stiffness and reducing secondary losses. And the multi-channel high-precision online monitoring of blade tip clearance parameters is a crucial aspect in ensuring the safety and efficiency of engine operation during its runtime. To achieve this goal and inhibit the adverse effects of the axial displacement, a novel "I-shaped" core-pole capacitance sensor is designed in this paper. Based on the modified labyrinth teeth, a tip clearance measurement model is established using the parallel-plate capacitor principle. A 12-channel blade tip clearance measurement system based on I-shaped core-pole sensor was developed. Calibration within a 3 mm measurement range was conducted on a simulated labyrinth disc, and precision testing was performed under ±1 mm axial displacement. The results indicate that the measurement accuracy of the aforementioned blade tip clearance measurement system exceeds 45.4 μm. The final stage involved the synchronization testing platform for blade tip clearance of actual engine blades with shrouds, encompassing twelve measurement points across three levels. Results demonstrate that the proposed 12-channel blade tip clearance measurement system exhibits high reliability and repeatability, and meets the needs of tip clearance measurement of shrouded blades.

Abstract:Aiming at the problems of traditional asymmetric half-bridge LLC converter in low-voltage DC power supply applications, such as large secondary current stress, uneven output current and difficult efficiency improvement, a dual transformer LLC resonant converter topology with symmetrical resonator is proposed, and the working status and characteristics of the symmetric resonator are deeply analyzed. The topology dual transformer structure can divide the resonant current in real time. The input current of the converter is continuous during the switching period and the current ripple is greatly reduced. The output current can be evenly divided and the output current stress can be reduced by paralleling the secondary sides. In order to further reduce the converter loop current, improve the converter efficiency and improve the traditional converter parameter design method, the converter gain model is established and the excitation inductance optimization curve is analyzed. The initial cavity parameters are designed considering the extreme working conditions of the converter, and then the precise time domain model of the converter is established to optimize the parameters. Finally, a 600 W converter prototype is built, and the experimental results verify the effectiveness and feasibility of the proposed symmetrical resonator double transformer topology and parameter design method. The maximum efficiency is 95.9%.

Abstract:To address the issues of single feature extraction and low detection accuracy in current malicious URL detection models when handling URLs with complex structures and diverse character combinations, this paper proposes a malicious URL detection model based on multi-scale attention feature fusion. First, Character Embeddings and DistilBERT are employed to encode characters and words separately, capturing both character-level and word-level feature representations in URL strings. Next, an improved convolutional neural network (CNN) is used to extract multi-scale character structural features and word-level semantic features, while a bidirectional long short-term memory (BiLSTM) network is employed to further extract deep sequence features. Additionally, an innovative attention feature fusion (AFF) module is introduced to dynamically fuse multi-scale features at both the character and word levels, effectively reducing information redundancy and enhancing the extraction of long-range sequence features. Experimental results show that the proposed model outperforms other baseline models, with accuracy improvements ranging from 0.32% to 4.7% and F1 score improvements from 0.46% to 5.5%, achieving excellent detection performance on datasets such as ISCX-URL2016.

Abstract:In order to obtain the defect location of the continuous casting billet caused by the failure of the deburring machine, a defect location method based on binocular vision was proposed. Firstly, aiming at the poor matching effect of AD-Census algorithm in casting blank images, a noise detection method of window center pixels is proposed, which replaces noise pixels with neighborhood pixel information, and integrates multi-directional gradient cost into the calculation of cost function to improve the reliability of initial cost; Secondly, an adaptive window cost aggregation based on gradient threshold is designed to improve the matching effect of the algorithm in weak texture region and edge region; Finally, the parallax map of casting billet is converted by three-dimensional coordinate to complete the defect location of continuous casting billet. Experiments show that the binocular vision defect location method proposed in this paper has high parallax map matching accuracy, and the average location error of the casting blank defect depth is less than 1 mm, which can provide reliable location information for the subsequent defect processing device.

Abstract:Aiming at the problem that landslide displacement is highly nonlinear and complex, and it is difficult to use traditional optimization algorithms combined with artificial intelligence methods for more reasonable and accurate predictive modeling, a L-vy flight strategy chaotic sparrow optimization algorithm (CLSSA)-variable modal decomposition (VMD)-support vector regression (SVR) landslide displacement prediction model is proposed. Firstly, CLSSA is used to optimize the VMD decomposition parameters to decompose the landslide displacement time series, secondly, the CLSSA-SVR model is used to predict the VMD decomposition subsequence, and finally, the cumulative displacement prediction is derived by superimposing the subsequence prediction data. The model is validated by taking the Baishui River landslide as an example, and the experimental results show that the proposed method has an MAE of 2.24 mm, an RMSE of 3.37 mm, and an R2 of 0.995 in the final cumulative displacement prediction, and relative to the sparrow optimization algorithm-variable modal decomposition-support vector regression (SSA-VMD-SVR), the improved optimization algorithm increases the adaptive ability of VMD that improves the efficiency of landslide displacement prediction for each component.

Abstract:In order to comprehensively show the research progress of the estimation method of the residual power of Lithium-ion batteries, this paper reviewed the relevant papers and patents in the databases of Web of science, cnki, the patent library of the China National Intellectual Property Administration et al since 2013, and summarized the mainstream estimation methods of the residual power of Lithium-ion batteries. This article summarizes the estimation errors of commonly used direct estimation methods (ampere hour integration method, open circuit voltage method, and impedance characterization), methods based on equivalent circuit models, methods based on electrochemical models, and methods based on artificial intelligence neural networks for estimating the remaining battery capacity of Lithium-ion batteries. The results show that the maximum estimation error of ampere hour integration method can reach 15%; the maximum estimation error of the open circuit voltage method is 12.4%; the average estimation error of electrochemical impedance spectroscopy is less than 3.8%; the estimation error of kalman filtering method is less than 1%; the average error of particle swarm filtering method can be less than 1%; the average error of the method based on electrochemical model is less than 2%; the average error of neural network-based methods is less than 2%; the maximum error of the multi method mixing and multi parameter joint estimation method is less than 5%, and the average error is less than 2.5%. The results indicate that the kalman filter method has higher accuracy and is easier to implement compared to direct estimation methods and other model-based methods; the method based on neural networks can obtain more accurate results without analyzing the battery model; the mixed use of multiple methods and the use of multiple parameters to correct the estimated values have further improved the estimation accuracy. This article also compares and analyzes the estimation accuracy, advantages, difficulties, and applicable battery types of various methods for estimating remaining power in electric vehicles and implantable medical electronic devices. It clarifies the specific application plans of estimation methods and looks forward to the development direction of estimation methods in these two fields. This article can provide comprehensive and detailed information on the research status and development direction of Lithium-ion battery remaining capacity estimation methods for researchers and practitioners in related fields.

Abstract:To enhance the safety and efficiency of dynamic obstacle avoidance for unmanned surface vessels (USV), an obstacle avoidance method combining the velocity obstacle method and deep Q-network (DQN) is proposed. First, when calculating the traditional velocity obstacle′s relative collision region, the future movement information of obstacles is considered. This improvement addresses the issue of obstacle avoidance failure caused by ignoring the real-time position changes of obstacles in the traditional velocity obstacle method. Second, a collision risk coefficient is introduced into the DQN state space, prioritizing obstacles with the highest risk coefficient as avoidance targets, thereby reducing redundancy in state space information. Third, a reward function is redesigned based on the improved velocity obstacle method′s obstacle avoidance concept, determining the timing and steering angle for USV obstacle avoidance. This solves the sparse reward problem in traditional DQNs, enhancing their learning efficiency and convergence speed. Finally, to validate the performance of this method, simulation experiments were conducted comparing it with three mainstream obstacle avoidance methods. The results show that this method can provide suitable avoidance directions for USVs, making their navigation paths more economical and safer. Additionally, real-ship experiments confirmed the method′s practical engineering value.

Abstract:A Harris Hawk optimization algorithm combined with controllable response power is proposed to address the issue of low accuracy in seismic source localization in special and complex media such as shallow underground layers, which is improved through multiple strategies. Firstly, the initial population is optimized using the Logistic chaos model, while introducing nonlinear escape energy and adaptive adjustment of weight factors to improve the convergence accuracy and speed of the algorithm. Then, the energy information in the seismic source area is constructed using the SRP positioning model. Finally, the precise localization of the seismic source is achieved through simulation verification. The simulation results show that compared with traditional HHO, HUHHO, and MGTO algorithms, the MHHO algorithm has significant improvements in particle search range, convergence speed, and positioning accuracy. The ball probability error has also been reduced from 0.56 m to 0.23 m. Finally, experimental comparison and verification were conducted, and the experimental results showed that the algorithm proposed in this paper has higher positioning accuracy and good engineering application value in the field of shallow underground positioning.

Abstract:In order to improve the accuracy of geomagnetic vector measurements and effectively deal with the shortcomings of traditional compensation methods in dealing with the problem of time-varying magnetic interference model parameters, this paper proposes a time-varying interference oriented compensation method for geomagnetic vector measurements. The method can accurately obtain the time-varying magnetic interference parameters and compensate them during the measurement process by iteratively updating the previous parameters using Woodbury′s equation based on the latest measurement data, the previous compensation parameters and the compression matrix. The effectiveness of the proposed method is verified through simulations and experiments, especially when the magnetic interference characteristics vary greatly. The experimental results show that the total geomagnetic field intensity and the rms errors of the three components can be reduced from 4 596.3、1 310.0、5 768.9 and 3 245.7 nT to 123.34、35.93、142.12 and 100.54 nT, respectively, when using the traditional method, but it can be further reduced by our proposed method to 31.57、11.09、35.13 and 27.26 nT. This shows that compared with the traditional method, the present method not only successfully captures the dynamic changes of the time-varying magnetic interference parameters, but also significantly reduces the root-mean-square error of the total geomagnetic field strength and each component, and realises the high-precision compensation of the time-varying magnetic interference.

Abstract:Industrial 4.0 revolution has led to a deeper integration of industrialization and digitalization, resulting in industrial control systems (ICS) characterized by nonlinear and high-dimensional data. These complexities render traditional intrusion detection methods ineffective. In this study, we propose an intrusion detection model for ICS based on the coronavirus herd immunity optimizer (CHIO). The model leverages Fisher-Score and kernel principal component analysis (KPCA) for feature extraction, effectively reducing the complexity of the data. To enhance the search performance of the CHIO, adaptive mechanisms and differential evolution strategies are incorporated. The improved algorithm is then applied to a support vector machine (SVM) for parameter optimization. The performance of the model is validated using the natural gas pipeline dataset from the University of Mississippi. Experimental results demonstrate that the proposed model offers significant improvements in both detection accuracy and speed compared to traditional methods, achieving a detection rate of 97.1%.

Abstract:Aiming at the inherent instability and nonlinearity of power load, which makes it difficult to improve the accuracy of short-term power load prediction. In this paper, we propose a short-term power load prediction method based on the combination of EMD and LSTM. First, the original signal is decomposed into a series of eigenmode functions and a residual quantity using EMD. Subsequently, all the components are input into the LSTM. To further improve the accuracy of load forecasting and optimize the generalization ability of the model, an improved sparrow search algorithm is introduced to optimize LSTM hyperparameters for large component signals, and a table generative adversarial network is introduced to generate new data samples for raw load data, forming a short-term power load forecasting method based on table generative adversarial network and EMD-ISSA-LSTM. Finally, the load data of the ninth mathematical modeling competition for electricians and the load data of a prefecture and city in Hunan Province containing distributed power sources are used to validate the effect, and the results show that the mean absolute percentage error of the model under the two datasets is 2.37% and 2.26%, respectively. The validity of the method is verified.

Abstract:In view of the potential impact of secondary path mutation on the convergence speed and system stability of the algorithm, an adaptive variable step size FXLMS algorithm is proposed. Firstly, the formula of the upper bound of the step size is derived, and the key parameters are established based on the relationship between the optimal convergence step size and the upper bound of the step size. By comparing the optimal step size ratio before and after the mutation of the secondary path, the adaptive adjustment of the step size was realized. Secondly, in order to compare the performance of the proposed algorithm, the classical FXLMS algorithm and various variable step size algorithms in terms of convergence speed and steadystate error, it is found that the new algorithm converges after 200 iterations, and the mean square error remains around -85 dB, which is better than the classical FXLMS and other variable step size algorithms. Finally, the control effect of the new algorithm and the classical FXLMS algorithm is analyzed by using the secondary path data of real measurements. The results show that after the mutation of the secondary path, the mean square error of the new algorithm is stable at -47 dB after 5 s, while the classical FXLMS algorithm will make the system unstable. It is proved that the new algorithm can take into account the convergence speed and steady-state error well, and has good adaptability.

Abstract:In the complex industrial processes, the process data have the characteristics of imbalance and are incomplete due to the difficult-to-measure key variables, leading to the performance degradation of soft sensors. In order to deal with this problem, a novel noise injection supervised enhanced autoencoder virtual sample generation method is proposed. Firstly, in order to strengthen the mapping relationship between input and output and ensure the integrity of feature information, this method adds an enhancement layer to the encoding part of the autoencoder, and introduces label information for supervised constraint training in the decoding part. In order to increase the diversity of virtual samples, Gaussian noise is added to the features extracted from the hidden layer of the supervised enhanced autoencoder. Combine the generated virtual samples with the original small samples to enhance the performance of the soft sensing model. Unlike traditional virtual sample generation methods, the proposed NISEAE-VSG model can simultaneously generate useful input-output virtual samples. To verify the effectiveness of the proposed method, simulations were conducted using datasets of thermal power generation and polyethylene processes. The simulation results show that the proposed method generates virtual samples that are superior to other virtual sample generation methods and can effectively improve the accuracy of soft sensing modeling.

Abstract:During the filtering process of explosion-induced vibration signals in shallow underground layers, the fixed step size of the D-LMS algorithm is not flexible enough for time-varying signal processing, which can easily lead to the amplification of gradient noise. Moreover, it relies on prior information about the effective signal or noise as the desired signal, which is usually unknown in shallow underground vibration testing. To address these issues and meet the needs of shallow underground explosion-induced vibration detection, an improved D-LMS filtering algorithm was developed based on normalization principles. This improved algorithm was compared with traditional algorithms in terms of convergence speed and filtering accuracy through simulations. The results showed that this improved algorithm achieved approximately 2.3 dB higher filtering accuracy and doubled convergence speed compared to the D-LMS algorithm in adaptive denoising of vibration testing. It was deployed on a ZYNQ programmable logic device, where modules for delay, step size, coefficient update, filtering, and error calculation were designed and encapsulated into an IP core. This core was integrated into the acquisition system for field tests of shallow underground vibrations. Experimental results demonstrated that the filtered signals were significantly better than unfiltered ones, confirming the effectiveness of the adaptive filtering module. This achieved real-time on-chip adaptive denoising of vibration signals, providing crucial support for the reconstruction of shallow underground vibration fields.

Abstract:Aiming at the problems of high delay and high power consumption of the multiplier part in RISC-V processors, this paper proposes an improved multiplier optimisation design based on symbol extension on the basis of the booth2 algorithm, which reduces the execution cycle of multiplication instructions in the processor and supports the operation of signed/unsigned numbers at the same time. The improved CSA32 compressor and the choice to alternate the Wallace tree structure with a 3.2 compressor and a 4.2 compressor improves the compression efficiency of the partial product, and also reduces the critical path delay and improves the speed of the multiplier operation. The coding verification and functional simulation of the multiplier are carried out using verification tools such as NC-verilog, and the comprehensive analysis is carried out using. Design complier at SIMC 180 nm process, and the results show that the multiplier designed in this paper reduces the multiplication instruction execution cycle by 88.2% compared with PicoRV32, the area and power consumption are better than those of the same type of multiplier.

Abstract:The recognition of space precession cone by using parameter characteristics can effectively distinguish true and decoy targets. In order to obtain accurate parameter characteristics, the current research generally focuses on the high resolution range profile sequences of the target output by wideband radar, but there are also limitations of high radar bandwidth requirements and no practical engineering applications. To solve this problem, a method based on narrow band radar is proposed to extract high resolution features and complete parameter estimation. Firstly, a target motion model is established to analyze the modulation characteristics of target echo in narrow band observation, and the pseudo-range resolution profile is defined and extracted using the scattering centers′ micro-Doppler and phase information, then the micromotion and structure parameters are estimated using the pseudo-range resolution profile, and the correct estimation result is selected by a two-level pick-out structure. Finally, a simulation experiment based on electromagnetic calculation data is carried out. The implementation results show that the estimation accuracy of the proposed algorithm is better than 97% under the condition of high signal-to-noise ratio, which is 5% higher than that of the traditional method, and the average parameter estimation accuracy is still better than 80% under the condition of 5 dB signal-to-noise ratio, and the application threshold is 9 dB higher than that of the traditional algorithm, indicating the effectiveness and robustness of the proposed algorithm.

Abstract:Addressing the constraints of utilizing a single signal for diagnosing various faults in high-voltage circuit breakers. In this paper,a multi-fault diagnosis method for high-voltage circuit breakers with joint characteristics of electric vibration is proposed.Firstly, extracting key time nodes of current waveforms and corresponding amplitudes of coil current signals during high-voltage circuit breaker closing operation by peak-valley algorithm to construct electrical features;and perform a variational modal decomposition(VMD) of the vibration signal, calculating multiscale dispersion entropy values for different modal components to construct mechanical characteristics.Next,he electrical and mechanical feature vectors were subjected to principal component analysis with dimensionality reduction, generating joint features of electric vibration based on the obtained variance contribution, effectively solving the feature vector redundancy problem; finally, the combined characteristics of signals under different faults are input into fuzzy cluster analysis, successfully identified the specific fault type in the high-voltage circuit breaker.According to the experimental findings,the proposed method demonstrates superior accuracy in fault diagnosis compared to single-signal approaches.It classifies effectively,validated in different diagnostic models with 98.6%.It successfully enables the diagnosis of faults in high-voltage circuit breakers.

Abstract:Aiming at the shortcomings of silicon photodiode type radiation sensor used in rotating shading solar radiation monitoring system, such as narrow spectrum, low precision and poor long-term stability, a method of dynamic calibration of silicon photodiode array using thermonuclear array is proposed, and a lightweight, low-cost and high-precision rotating shading solar radiation monitoring device is designed. The detection system uses static process and dynamic process to detect in real time alternately. It is divided into static acquisition of light intensity and dynamic acquisition of light intensity changes of thermonuclear array and silicon photodiode array in the rotating process of the light shield. By calculating the solar altitude angle and subsequent data processing, the value of total solar radiation, direct radiation and scattered radiation can be obtained. The error analysis figure shows that the root mean square error of the total radiation is 1.859 W/m2, the direct radiation and the scattered radiation are 2.922 W/m2 and 2.770 W/m2, respectively. Compared with the single silicon photodiode array, the root mean square error is reduced by 8.908 W/m2 and 5.454 W/m2, respectively. It is proved that the measurement method has high detection accuracy.

Abstract:In order to address the issue of the neural radiation field rendering results being overly smooth when sparse viewpoint input conditions are present, resulting in a lack of detail, a network model based on an information attention suppression module and a two-stage loss function has been proposed. The first step is to propose an information attention suppression module, which uses a feature vector normalization module to filter outliers in the weights between layers of MLP. It also uses a residual network to cascade global and local information and employs channel attention to differentiate fused information based on its degree of importance. This process improves the accuracy of the sampling points′ feature vectors. To address the issue of low perceptual accuracy resulting from overly smooth rendering, a two-stage loss function is proposed. This function partitions the training phase into two stages. In the initial coarse stage, training is guided by RGB and depth loss. Subsequently, in the fine stage, perceptual loss and TV loss are incorporated. This approach enables the utilisation of high-level image features, thereby enhancing the image perception ability via gradual optimization. This paper′s algorithm is compared with other classical methods, and on the LLFF dataset, the quantitative results demonstrate that the overall performance reaches its optimal value, which is 1.9% superior to the performance of the sub-optimal algorithm. Furthermore, on the DTU dataset, the qualitative results indicate that the reconstruction′s completeness and detail level, as observed in Scan37, Scan55, and Scan63, are notably enhanced.

Abstract:Aiming to address the false detections caused by background reconstruction errors in the self-supervised leaky cable snap defect detection algorithm, this paper proposes a two-stage leaky drop snap defect reconstruction method based on region of interest constraints. For the classified snap region images localized by the target detection algorithm, a segmentation network is first employed to differentiate between snap and cable leakage regions. Subsequently, the corresponding masks are utilized to guide a stacked adversarial generative network to reconstruct the snap regions, ensuring high-quality reconstruction of the defect areas while maintaining background consistency. Additionally, the generative network is optimized to place greater emphasis on reconstructing the regions of interest by integrating deep residual blocks and refining the loss function. Ultimately, the trained network is deployed for the reconstruction of snap images, determining the presence of defects based on the similarity scores of the images before and after reconstruction. Quantitative results on the leaky cable snap dataset demonstrate that the proposed algorithm achieves a defect recognition accuracy of 92.3% and a recall rate of 93.4%, surpassing the performance of other self-supervised snap reconstruction methods. Visualization results further indicate a reduction in background reconstruction errors in the proposed method.

Abstract:Aiming at the problems of low accuracy, high missed detection rate and insufficient real-time performance of railway track defect detection, this paper proposes a rail defect detection algorithm based on YOLO-FCA. First, the backbone network of YOLOv7 was replaced with the lightweight network of FasterNet, and the attention module of CloAttention was added to reduce the number of parameters and calculation load while improving the accuracy of defect detection. Secondly, a multi-scale adaptive feature fusion network (MS-ASFF) is proposed to obtain high-level semantic information and retain low-level detailed features to enhance the accuracy and robustness of model detection. Finally, the network pruning is carried out without affecting the accuracy, which makes the model more lightweight and greatly improves the detection speed of the model. Experiments on public data sets show that compared with the original YOLOv7 model, the mAP of YOLO-FCA is increased by 4.1%, reaching 80.7%, and the detection speed is increased by 38.5%, reaching 212.5 FPS. The experimental results show that YOLO-FCA can locate and detect rail defects efficiently and accurately.

Abstract:To address the challenges of target detection in foggy conditions in real-world scenarios, this paper proposes an improved foggy target detection method based on YOLOv8s. The design includes a front-end module, Edge-Dehaze, which employs joint training of dehazing and detection networks and uses the Sobel operator to enhance edge information in foggy images, thereby improving detection performance in foggy environments. The proposed hybrid attention feature fusion module (HAFM) utilizes parallel attention mechanisms to enhance information interaction and fusion between feature maps, increasing the model′s focus on critical features. Additionally, a lightweight shared attention convolutional detection (LSACD) head is designed, which reduces the parameter count of the detection head through shared convolutions and incorporates the SEAM attention mechanism in the shared layer to alleviate occlusion issues in foggy target detection. Experimental results on the RTTS dataset demonstrate that the improved YOLOv8s network achieves a 1.8% increase in mAP50 and a 1.7% increase in mAP50-95 compared to the original YOLOv8s network, with comparable parameter counts, thereby validating the high accuracy and practicality of the proposed method in foggy target detection.