Abstract:Research questions: In optical wireless communication systems, atmospheric turbulence can cause the transmission beam to expand, drift and light intensity fluctuation, which will seriously reduce the signal quality of the receiving end and reduce the performance of the communication system. Therefore, the study of methods to suppress atmospheric turbulence is the key to improve the performance of optical wireless communication systems. Method and process: Large-aperture receiving technology, diversity technology, partially coherent beam technology and adaptive optics can effectively suppress the atmospheric turbulence effect, which is an important means to improve the performance of optical wireless communication systems. Detailed detail the principle of suppress atmospheric turbulence and its means. These key technologies can improve the quality of the received signals and enhance the reliability of the communication system by changing the transmission or reception strategy, regulating the structure of the optical field, enlarging the receiving aperture, and compensating for wavefront distortion. Meanwhile, the effects of different parameter indicators on the system performance are also analyzed. The current status of domestic and international research on the relevant suppression techniques is discussed, and the improvement of different performance indexes of the system under the influence of atmospheric turbulence by the relevant techniques is showed. Conclusions: Finally, the challenges and problems in atmospheric turbulence suppression in the field of optical wireless communication are summarized, and the future development trend of the technology is outlooked, which can provide a reference for the future development in this field.

Abstract:The operation of power transformers involves significantly fewer fault data compared to normal data, resulting in a severe data imbalance issue. Additionally, the complex coupling relationships among the monitored variables make the modeling of condition assessment tasks challenging and lead to low evaluation accuracy. Aiming at the related problems, a power transformer health condition evaluation method based on double input residual graph convolutional network is proposed. First, SMOTE Tomek mixed sampling algorithm was used to pre-process the unbalanced data of the training data, which solved the problem of insufficient fault data and difficult classification. Then, a multi-metric fusion graph construction method is proposed to learn the correlation between variables from multiple variables and construct the graph structure data. Finally, a double input residual graph convolutional network(DI-ResGCN) based on the ChebyNet is proposed, feature extraction is carried out on the constructed graph structure data, and feature fusion is carried out through the selfattention mechanism to obtain the transformer health evaluation results. Experiments were carried out on a dataset of dissolved gases in oil and oil test collected by a real power transformer, and experimental results show that the proposed method has a state assessment accuracy of 94.7% and F1 score of 0.942, outperforming other deep learning methods and exhibiting the best evaluation performance.

Abstract:To improve the reliability of the integrated navigation system, an improved fault detection and information fusion method is proposed for the fault-tolerant mechanism of the integrated navigation system. Designed an improved sequential probability ratio test method, introduced a fading factor to improve the tracking speed of residual information at the current time, combined with Mahalanobis distance to determine the end time of the fault, and fully reset the judgment information at the appropriate time based on the judgment result; A self-adaptive normalization fusion algorithm based on federated filtering was designed to construct normalized detection values of fault detection statistics, which were used as weight coefficients for the measurement noise variance matrix. The corresponding sub filters were weighted and updated to change the weight allocation in the global fusion process. The results of the in vehicle experiment show that the improved fully reset sequential probability ratio test algorithm has improved the positive detection rate of soft fault detection by 96.43%, 25.00%, and 19.57% respectively compared to the traditional residual chi square test, fading sequential probability ratio, and fast reset sequential probability ratio methods. The adaptive normalization fusion algorithm used also improved the positioning accuracy by 44.70% and 35.60% compared to the traditional federated filtering method. Therefore, the two improved methods can greatly enhance the fault tolerance performance of the integrated navigation system and have high practical value.

Abstract:To improve the accuracy of fault detection in array antenna, an enhanced differential evolution-genetic algorithm (DE-GA) is proposed. This algorithm combines the advantages of genetic algorithm (GA) and differential evolution (DE) by employing a dual crossover strategy to help individuals escape local optima. An adaptive weighting mechanism further optimizes offspring selection, enhancing the algorithm’s sensitivity and adaptability to fault conditions. Applied to array antenna fault detection, the DE-GA algorithm models the array and optimizes its radiation pattern to match the known faulty pattern, allowing the faulty array’s amplitude to be estimated. Experiments show that compared with DE and GA, DE-GA reduces the fitness function value by 11.15% and 12.90%, the mean absolute error by 19.36% and 23.85%, the mean square error by 12.90% and 11.15%, and the maximum error by 12.30% and 13.18%. This demonstrates higher accuracy and improved approximation capabilities. Additionally, the algorithm maintains excellent stability with larger arrays, making it suitable for large-scale fault detection.

Abstract:In the actual industrial environment, the collection of gas turbine rotor fault data is challenging, leading to a scarcity of fault samples and an inability to meet the massive training requirements of fault models. Leveraging the advantages of DenseNet in image feature extraction and the Transformer structure in the visual field, an improved gas turbine rotor fault diagnosis method based on the DenseNet-ViT joint network was proposed. Firstly, the classification layer of DenseNet was abandoned, and only the feature extraction layer of DenseNet was utilized. Subsequently, the output layer of the modified DenseNet was connected to the input layer of the ViT model to form the joint network. Additionally, in response to the issue of lengthy training time for the fault model, transfer learning was employed to transfer the trained model’s weight parameters, which could expedite the training process and conserve computing resources. Simulated data of gas turbine rotor faults could be acquired through the gas turbine rotor simulation experimental platform constructed in the laboratory, and real fault data of different types of rotors in the actual environment were obtained on a certain type of gas turbine test bed. Utilizing both the simulated and real data for model testing could better verify the reliability of the proposed method. The experimental results indicate that the fault recognition accuracy rates reached 96.8% and 97.3% in the tests of two distinct rotor fault datasets, respectively, demonstrating that this method possesses a relatively high rotor fault recognition accuracy. In the subsequent comparative verification experiments, by comparing with CNN and VGG-16, etc, the fault classification accuracy of this model was also higher than those networks, thereby further validating the superiority and reliability of this model.

Abstract:To address the inconsistent bearing fault data distribution that leads to the difficulty of feature offset and distinctive feature extraction, a bearing fault diagnosis method based on time-frequency filter and offset attention neural network is proposed, which processes the fault signal from offline and online parts. In the offline part, a time-frequency filter is proposed to extract the distinctive features from time domain and frequency domain; A spatial sampling method considering both global and local features is proposed. In the online part, an offset attention neural network is proposed. Compared with self attention, offset attention is more conducive to the extraction of offset features, so as to reduce the impact caused by inconsistent data distribution. Experiments on the bearing datasets of Xi′an Jiaotong University (XJTU) and Case Western Reserve University (CWRU) have achieved 100% accuracy, which proves that the proposed method can efficiently extract the distinctive features of fault signals, and effectively suppress the influence of feature offset. The comparative experiment on the bearing dataset of CWRU proves the superiority of the proposed method. In addition, experiments are also carried out on the dataset of gas turbine main bearing collected in the industrial field, and the results show that the proposed method has practical significance.

Abstract:Aiming at the problem that the convolutional neural network extracts the features from the input signal through the local receptive field, and cannot effectively capture the global context information under variable load and noise environments, resulting in the low recognition accuracy of rolling bearing fault diagnosis, a rolling bearing fault diagnosis method based on multiscale adaptive depthwise separable convolution (MADSC) and spatial interaction double-stream Swin Transformer (SIDSwinT) is proposed. Firstly, one-dimensional vibration signals are converted into two-dimensional time-frequency maps using wavelet transform to retain the complete information. Next, MADSC is constructed to extract local feature information and capture the characteristic changes of rolling bearing vibration signals at different scales. After that, SIDSwinT is designed to extract the global feature information, and the proposed spatial interaction module (SIM) is utilized to adaptively adjust the feature weights, while the sampled information is weighted by the deformable attention to eliminate the distributional differences caused by fluctuations in working conditions. Finally, bidirectional long short-term memory (BiLSTM) is utilized to better understand the contextual information and to improve the diagnostic accuracy and stability. Two different datasets are used to verify the fault diagnosis performance of the proposed method, and the experimental results show that the accuracy of the proposed method is higher than 93.00% when the signal-to-noise ratio is -4, and the accuracy is higher than 92.00% under the condition of variable load, which verifies that the proposed method has a stronger anti-noise performance and generalization ability than the comparison methods.

Abstract:In the railway system, the switch machine is a critical device to ensure the safe and smooth operation of trains. Fault diagnosis of the S700K switch machine is crucial for accident prevention and the maintenance of railway operations. To address the shortcomings of traditional diagnostic methods in terms of speed and accuracy, a diagnostic model integrating a deep residual shrinking network with a bidirectional long short-term memory network is proposed. First, the power curve of the switch machine is preprocessed. Next, DRSN is used to automatically learn features from the preprocessed data and compress the data length, improving the speed of diagnosis. Its attention mechanism and soft thresholding reduce the influence of noise features, and the DRSN structure helps to overcome network degradation and overfitting issues. Following that, the bidirectional structure of BiLSTM is utilized to capture complex relationships in the time-series data. Finally, a Softmax classifier is employed for fault classification. Simulation results show that the accuracy, precision, and recall rates of the DRSN-BiLSTM model all exceed 98.3%. Compared with models such as DRSN, deep neural network, and convolutional neural network, the diagnostic accuracy of this model is improved by at least 1.47%. Even when Gaussian white noise in the range of 15~40 dB is added, the accuracy remains above 92.7%, an improvement of at least 2% over other models. This model not only ensures the efficiency of the training process but also improves the accuracy of point machine fault diagnosis and demonstrates excellent robustness in noisy environments.

Abstract:To enhance the accuracy of photovoltaic (PV) array fault diagnosis, this study proposes a novel method that utilizes an improved golden jackal optimization (IGJO) algorithm to optimize a deep hybrid kernel extreme learning machine (DHKELM) for PV array fault diagnosis. Initially, a range of PV array faults are simulated using the MATLAB/Simulink platform. Based on a comprehensive analysis of fault characteristics, a 12-dimensional feature set is proposed for fault diagnosis. Subsequently, the golden jackal algorithm is improved by introducing lens imaging reverse learning strategy, cosine and sine algorithm strategy, and adaptive T-distribution perturbation strategy to enhance its convergence speed and global optimization capability. Additionally, IGJO is compared with other optimization algorithms using test functions. Furthermore, radial basis kernel functions and polynomial kernel functions are incorporated into the extreme learning machine and combined with an autoencoder to form the DHKELM model. Finally, IGJO is employed to optimize the initial parameters of the DHKELM model, resulting in the establishment of the IGJO-DHKELM PV array fault diagnosis model. Analysis of the results indicates that the proposed 12-dimensional feature set provides higher diagnostic accuracy compared to traditional 4-dimensional and 5-dimensional feature sets. Moreover, the IGJO-DHKELM-based fault diagnosis method demonstrates superior diagnostic accuracy compared to other fault diagnosis models.

Abstract:Hull corrosion is primarily caused by the deterioration of protective coatings. Given that inspections of hulls are both time-consuming and labor-intensive, thereby conducted infrequently, this study aims to propose a rapid detection and localization method for hull coating damage. Using the changes about underwater electric potential caused by the impressed current cathodic protection system and coating damage, the location of the damage can be identified by measuring the potential difference underwater at symmetrical positions on the hull. Underwater electric field transmission experiments have confirmed the feasibility of methods for ranging and locating fault positions based on electric field characteristic signals. Additionally, COMSOL Multiphysics simulation software was utilized to model the corrosion electric field generated by the cathodic protection current and the electrochemical corrosion processof metallic surfaces on the hull.By analysing the potential variation along the measurement segments on both sides of the hull, it is observed that the underwater potential difference is notably largest around the damaged coating area.Based on the transmission characteristics of the underwater electric field, longitudinal and transverse positioning of the damage points have been achieved, with average deviations of 0.2 m and 0.21 m. Moreover, a linear relationship between potential magnitude and damaged area was observed. This method offers high accuracy and is suitable for longitudinal positioning of multiple points with intervals greater than 3 m. It mitigates environmental factors’ interference with detection, enabling early and rapid detection of hull coating damage and improving corrosion management for vessels.

Abstract:In order to achieve accurate monitoring of abnormal wind turbine blades, a method combining complementary ensemble empirical mode decomposition with the sound features of wind turbine blades was proposed. Firstly, the voiceprint data of four kinds of fan blades under abnormal working conditions and normal operating conditions are collected and pre-processed for noise reduction, frame division and window addition. Through experimental comparison, the complementary ensemble empirical mode decomposition algorithm is selected for secondary noise reduction of voiceprint data. Secondly, the modal decomposition of frame signals after secondary noise reduction is carried out to extract modal components. The effective modal components were selected by calculating the Pearson correlation coefficient of the modal components, and the characteristics of mel frequency cepstrum coefficient, linear prediction cepstrum coefficient, gammatone cepstrum coefficient, short-time energy and short-time mean zero crossing rate were extracted for each layer of modal components. Finally, based on these feature combinations, support vector machine, naive Bayes and neural network are used as fault classification models to identify voicing data. The research results show that the neural network model based on the combination of the above five vowels features and the parameter optimization can achieve the accurate recognition of blade anomalies, with the recognition accuracy of 97.5%. The model has a good recognition effect on early abnormal fan blades, and has good generalization performance.

Abstract:In order to improve the accuracy and classification speed of fault diagnosis of valve cooling equipment in converter station, a fusion feature algorithm based on Fisher ratio criterion and a fault classification model based on particle swarm optimization least squares support vector machine are proposed. Firstly, the static parameters and dynamic first-order difference parameters of Mel cepstrum coefficient and inverse Mel cepstrum coefficient are extracted as fault feature quantities respectively, and all the high and low frequency information of valve cooling equipment fault is obtained. Then, Fisher ratio criterion is used to fuse the fault features of valve cooling equipment twice, so as to reduce the repeated data and interference signal caused by direct superposition signal. The 1×13 dimensional Fisher ratio data is selected as the fusion feature of the noise signal of the valve cooling equipment. Secondly, in order to improve the accuracy and classification speed of LSSVM algorithm fault identification, the PSO algorithm is used to optimize the kernel function bandwidth and penalty factor of LSSVM algorithm, and the optimal solution of the two parameters is obtained, and the LSSVM valve cooling equipment fault classification model is established. Finally, the main pump between the valve cooling equipment is taken as an example, and different feature fusion algorithms and fault identification methods are used for comparative analysis. The results of the example verify that the proposed method can quickly and accurately identify the fault signals of the valve cooling equipment at different frequencies, and the accuracy of fault identification can reach 96.67%.

Abstract:When using a Doppler microwave sensor to measure the flow of granular fertilizer, the vibration generated by the operation of the fertilizer applicator and various external disturbances can cause the collected signal to be distorted. This article first explores the optimal parameters for wavelet analysis and Kalman filtering algorithms. By comparing the denoising effects of the two algorithms, a denoising algorithm based on the combination of empirical mode decomposition and sample entropy combined with wavelet is proposed. Taking Stanley 15-15-15 granular fertilizer as the experimental object, the detection system such as Doppler microwave sensor is deployed on the fertilizer applicator to collect the mass flow signal of granular fertilizer for algorithm effect experimental verification.The results indicate that, compared to the original signal, the average signal-to-noise ratio of the Kalman filtering algorithm improved by 3.548 dB after optimizing the gain coefficient. After optimizing the wavelet denoising parameters, the average SNR of the wavelet analysis algorithm increased by 7.184 dB. When combining the optimized wavelet analysis with the denoising algorithm of integrated empirical mode decomposition and sample entropy, the average SNR of the denoised signal increased by 7.899 dB, while the average root mean square error decreased by 0.184, this algorithm demonstrates significant advantages in denoising the mass flow rate signals of granular fertilizers.

Abstract:Pneumonia, a common respiratory infection worldwide, often leads to various complications, making its precise classification a critical issue in clinical diagnosis and treatment. This study addresses the need for accurate classification of respiratory infections and pneumonia by developing an effective diagnostic method based on the Raman spectroscopy of respiratory mucus. Initially, respiratory mucus samples from normal individuals, patients with common pneumonia, and those with concomitant plastic bronchitis were collected. Through Raman spectroscopic analysis, molecular features and chemical changes related to mucin glycosylation and fibrosis in each group were accurately identified, detailing the components and molecular bond alterations associated with the disease. Subsequently, combining principal component analysis and partial least squares discriminant analysis, a classification model capable of distinguishing between different types of pneumonia was constructed. Experimental results demonstrated high accuracy of the model in classifying pneumonia, with an overall classification accuracy reaching 99.08%, and specifically, 100% and 97.4% accuracy in distinguishing common pneumonia and plastic bronchitis, respectively. The study not only confirms the potential of Raman spectroscopy in the diagnosis of infectious diseases but also provides a reference for the broader application of molecular spectroscopic techniques in infectious disease diagnostics.

Abstract:For flux weakening control of the internal permanent magnet synchronous motors, when the degree of flux weakening is deeper, the motor parameter perturbation and external disturbances will cause the voltage loop output, torque and current pulsation to increase, and the speed convergence is too slow. A speed-voltage loop feedback super-twisting non-singular fast terminal sliding mode controller (FST-NFTSMC) is proposed for deep flux weakening control. To reduce the dependence of flux weakening control on the system model, the voltage-loop hyperlocal model is constructed according to the mathematical model of the built-in permanent magnet synchronous motor during parameter perturbation. And it is combined with the speed loop hyperlocal model to establish the speed-voltage loop hyperlocal model. Based on this hyperlocal model, the speed-voltage loop FST-NFTSMC is designed by combining the feedback super-twisting algorithm and the non-singular fast terminal switching function. At the same time, an improved sliding mode disturbance observer is built to estimate the unknown part of the system and compensate for the estimated value feedforward to FST-NFTSMC, which further improves the robustness and control accuracy of the system. Simulation and experiment show that compared with the traditional PI control, the convergence speed of the proposed method in the no flux weakening region, shallow flux weakening region, and deep flux weakening region is improved by 66%, 40.6%, and 28.6% respectively. It has better stability and fewer pulsations of the torque and current, proving that the method in the flux weakening control is effective in suppressing the output jitter after the voltage loop is perturbed as well as improving the speed response.

Abstract:This study addresses the long delay issue in underwater wireless sensor networks (UWSNs) caused by the spatio-temporal complexity and dynamics of the underwater environment, which significantly impacts the information propagation between mobile sensor nodes and consequently leads to large node localization errors. To this end, a novel underwater mobile node localization algorithm based on CNN-LSTM sound speed prediction is proposed. Initially, the sound speed dataset is partitioned using the K-fold cross-validation method. Subsequently, a hybrid CNN-LSTM model is constructed and trained, leveraging the feature extraction capability of CNN and the sequence modeling strength of LSTM. This model efficiently captures both spatial and temporal information from the sound speed dataset, thereby enhancing the prediction accuracy. During the localization process, the predicted sound speed values from the CNN-LSTM model are employed for time difference of arrival (TDOA) ranging, and the ranging values are refined accordingly. Finally, the refined ranging values are utilized to adaptively select the optimal ranging and localization method for unknown nodes under varying node densities, based on the number of reference nodes, thereby achieving precise localization of underwater mobile nodes. Experimental results demonstrate that, compared to existing localization algorithms such as SLMP, DMP, NDSMP, and BLSM, the proposed MCLS localization algorithm reduces the mean localization error by 46.96%, 39.93%, 27.64%, and 15.24%, respectively, under the same beacon node conditions, significantly improving the localization accuracy and stability of underwater mobile nodes.

Abstract:The photometric loss has been playing an important role in the training of video-based unsupervised monocular depth estimation models. However, it generally has large errors in special regions such as weak texture regions and edge regions, which leads to strong instability in the supervision signal of the training network. To solve the problem, a more robust unsupervised monocular depth estimation method is proposed. The method first combines the dual-branch encoder and the channel attention module to improve the extraction ability of the single-frame depth network for depth features. Then, the single-frame depth network results are used to guide the multi-frame depth estimation to improve the accuracy of depth estimation. On the basis, a new photometric loss function is designed. By calculating the photometric loss on the image gradient, the unreasonable supervision caused by local brightness changes is eliminated. At the same time, the difference between successive pixels is used to define the blurry pixels. Finally, the false supervision caused by the blurred pixels on the target frame and the reconstructed target frame is excluded based on the binary mask. In the test results of the KITTI dataset, multiple indicators such as the average relative error, the square relative error and the root mean square error have improved. The average relative error and the squared relative error are reduced to 0.075 and 0.548 respectively. The experimental result shows that the proposed method further improves the performance of existing models compared with other advanced methods.

Abstract:For the difficulty in extracting deep features of batch process variables, as well as low quality prediction accuracy caused by the temporal, nonlinear, and dynamic characteristics of variables, this article proposes a quality prediction model for batch processes based on convolutional neural networks spatial and temporal attention with double long short term memory networks (CNN-STA-DLSTM). Firstly, the three-dimensional data of the batch process are expanded into a two-dimensional matrix along the direction of the variables, and the two-dimensional data are normalized by the Max-Min method. Then, the partial least squares (PLS) method is used to reduce the dimension of the original data, and the variables with strong correlation with the quality variables are retained. The convolutional neural network (CNN) is used to mine the potential features of the process data and improve the attention of the quality-related feature information. Secondly, the temporal attention mechanism and the spatial attention mechanism are introduced to construct the encoder-decoder structure network of the double-layer LSTM, and the attention mechanism is used to adaptively learn the relevant historical information of the time step, so as to improve the long-term memory ability of the model and strengthen the spatio-temporal correlation between the process variables and the quality variables. Then, the random-grid search method is used to optimize the hyperparameters of the prediction model, and the prediction model is constructed. Finally, the penicillin fermentation simulation platform and the hot strip rolling production process data are used for experimental verification. The results show that the proposed model has more accurate prediction effect.

Abstract:Aiming at the problems of large torque ripple in Switched Reluctance Motor and traditional optimization design that only starts from the motor without considering the drive control strategy, a system-level multi-objective optimization design strategy for SRM considering model predictive torque control is proposed by simultaneously considering the motor structure parameters and control parameters. Firstly, the structural parameters of SRM were designed according to the design requirements and MPTC was adopted as the control method to determine the initial values and variation ranges of the motor structure and control parameters; Secondly, an SRM design model considering MPTC was established, and the relationship between structural parameters and prediction models was determined through magnetic circuit analysis. The optimization process of the motor was determined with torque ripple, average torque current ratio, and copper loss as optimization objectives. Sensitivity analysis of structural and control parameters was conducted through orthogonal experiments, and decision variables were selected based on the analysis results. Taguchi algorithm was used for multi-objective optimization of decision variables; Finally, in order to verify the effectiveness of the method, simulation verification was conducted, and a prototype was trial produced based on the optimization results. The experimental results showed that compared with the conventional design, the optimization results reduced the peak motor phase current by 33%, increased the average torque ampere ratio by 33.3%, and reduced torque ripple by 26.3%. The rationality and effectiveness of the optimization method were verified through experiments.

Abstract:Aiming at the problems of small effective area, fixed position and cumbersome arrangement of the traditional velocimetry device in the field of area-intercept velocimetry, based on the principle of electromagnetic induction, a new type of electromagnetic induction sensing unit is proposed and verified for the accurate measurement of the initial velocity of the projectile. Compared with the traditional magnetic induction coil, this structure adopts an induction coil wrapped with a permanent magnet, so that the projectile does not need to be magnetised to generate an induced electromotive force, which improves the sensitivity and measurement accuracy of the velocimetry target. In addition, the sensing unit is independently arranged coaxially with the trajectory, which effectively solves the problem of the relative position between the direction of the ballistic trajectory and the stable position of the test device, increases the effective area of the magnetic induction, and strengthens the portability of the measurement device, which can make it Flexible use in a variety of projectile velocity measurement occasions. The solution uses COMSOL software to model the sensing unit, and conducts detailed simulation analysis of the permanent magnet model and the dynamic process of the projectile passing through the magnetic field under different conditions. Based on the simulation data to create a coil sensing unit, and the simulation results of a number of experimental verification, test results show that the sensing unit sensing voltage increases with the speed of the projectile, and the two are linear within a certain range, consistent with the results obtained from the simulation. This study not only provides theoretical basis and data support for the optimisation of electromagnetic induction velocity target, but also outlines an effective solution for the measurement of in-bore and out-of-bore ballistic muzzle velocity of electromagnetic artillery and other high-speed launch systems.

Abstract:Leveraging extensive ECG data, intelligent ECG recognition represents a pivotal research focus aimed at supporting physicians in conducting thorough data analysis and diagnosis, thereby enhancing efficiency and mitigating medical resource consumption. In order to solve the problem of feature loss and limited performance of single image and single deep learning algorithm in ECG intelligent recognition, a two-input improved VIT recognition method for ECG rainbow code is proposed. Firstly, a mathematical model is proposed to predict the standard period of ECG, and the potential features of ECG are mined by pumping method to generate ECG rainbow code. Then, a dual input feature extraction module is constructed with convolutional neural network to extract local features of multiple ECG images for fusion to achieve multi-dimensional ECG feature representation and fusion. A VIT coding module is used to pay global attention to fusion features to realize ECG recognition based on multi-feature images as input. The ECG recognition method in MIT-BIH database is used for experiments, and the average accuracy of the proposed ECG recognition method is 99.41%, and the accuracy of the N-type ECG collected in the field is 100%. The experimental results show that the proposed image transformation method can effectively visualize ECG features, and the effect is better than the traditional method. The proposed recognition method can realize ECG recognition effectively and has better performance than other similar methods.

Abstract:In the transformation and upgrade of special steel enterprises into “lighthouse factories”, real-time tracking of steel pipe materials is a core component. Due to the diversity of materials and the complexity of the production line, proximity sensors fail to meet the reliability requirements of material detection. Therefore, according to the existing environment and requirements of the workshop, a material tracking camera system is built, and the image data set composed of some characteristics of materials and production lines is collected. Based on video analysis, a steel pipe target detection algorithm for real-time material tracking in special steel workshops is introduced. The algorithm is based on the PPYOLOE network. Firstly, the CSPRepResNet backbone in PPYOLOE is replaced with the lightweight HGNetV2 backbone, which enhances feature extraction capabilities while reducing the number of parameters. Secondly, HG-Block and SPPELAN are integrated into the Neck, further reducing the parameters and improving speed. Finally, in the upsampling stage, the Dysample dynamic upsampling operator is employed to enhance the fusion of multi-scale features, thus improving detection accuracy. Experimental results show that compared with the original PPYOLOE algorithm, the improved algorithm enhances detection accuracy by 1.6%, reaching 80.5%, and increases detection speed by 16%, reaching 56.4 FPS, while GFlops and parameters are reduced by 35% and 33%, respectively. The improved algorithm effectively boosts both detection accuracy and speed,and through on-site deployment, it meets the real-time tracking requirements of steel pipe materials.

Abstract:Current research on sensor array layout in magnetic positioning systems primarily focuses on quantity and spacing. In related research, the sensor array layout is typically evenly distributed, with limited investigation into the impact of spatial design on system positioning accuracy. Addressing the non-uniform distribution of sensor arrays in magnetic localization systems, this paper proposes an optimization method combining genetic algorithms with finite element simulations. This method determines the optimal sensor layout based on specific trajectories of magnetic targets. Firstly, a simulation model was established for numerical simulation of the magnetic positioning process, and the sensor array layout corresponding to the motion trajectory of each target was optimized by genetic algorithms. Secondly, based on the simulation optimization, an experimental platform for magnetic positioning systems with adjustable sensor installation positions was designed and constructed. Finally, comparative experiments were conducted on five specific magnetic target trajectories using both uniformly distributed and optimized non-uniformly distributed sensor layouts. For example, under trajectory five, the average positioning error of the optimized layout is reduced by 14.3% compared to the pre-optimization layout, and the average orientation error is reduced by 16.3%. The results indicate that uniformly distributed sensor array is not the optimal layout, and optimizing sensor array layout can effectively improve system localization and orientation accuracy.

Abstract:Aiming at the problem of low measurement accuracy due to the existence of random error in microelectromechanical system (MEMS) gyroscope, a compensation method based on chaotic particle swarm algorithm (CPSO) optimized back propagation (BP) neural network is proposed to deal with the random error. Firstly, The MEMS gyroscope data are collected, the reconstruction parameters are determined and the phase space is reconstructed using the C-C method, and the chaotic properties are analyzed and verified based on the Lyapunov exponent. Then, the reconstructed data are used as the training samples for the BP neural network model. The BP neural network model is trained, and the weights and thresholds of BP neural network are optimized by using the CPSO algorithm, then the optimized model for error compensation is obtained. Finally, ADXRS624 is used to validate the compensation effect of the optimized model in static experiment, and the compensation results are compared with BP model and particle swarm optimization (PSO) model. Experimental analysis results show that the mean and standard deviation of the gyroscope output errors are -5.76×10-4（°）/s and 5.19×10-4（°）/s, which are decreased by 68.6% and 98.4% compared with the BP model, and 52.1% and 93.5% compared with the particle swarm optimization model, respectively. By comparing the error coefficients after compensation for each method using Allan variance identification, the quantization noise, angle random walk and zero bias instability after being compensated by CPSO-BP method are reduced to 0.000 59 μrad, 0.001 51 （(°)·h-1/2） and 2.82 （(°)·h-1）, respectively. The new method has obvious effect in suppressing the random error and can improve the measurement accuracy of MEMS gyroscope.

Abstract:Aiming at the problems that the current inspection and monitoring methods cannot accurately identify the fire area below 10 m2 square meters in the high-concentration smoke environment, and it cannot accurately identify the mountain fire in the low-lying area, the fire location technology of the transmission line in the high-concentration smoke environment was studied based on the integrated inertial navigation multi-data fusion. This technology combines and fuses multi-source monitoring data such as satellite remote sensing and inertial navigation data of transmission lines in mountainous areas by using federated filter. An adaptive threshold detection algorithm was designed for ignition points based on context judgment and absolute threshold method, and implement the extraction of ignition point information for transmission lines in mountainous areas. A multi band optoelectronic composite detection target recognition method was designed combining PSO algorithm and BP neural network to achieve the recognition of fire point targets on transmission lines in mountainous areas under high concentration smoke environment. Location of transmission lines in mountainous areas fire point location method based on laser ranging, fire point location of transmission lines in mountainous areas in high smog environment is realized. The experimental test results indicate that the design technology is effective for fire points (10 m2 above the fire area), fire points in plain areas (10 m2 The fire area below), and the fire point in low-lying areas (10 m2 above the fire area). The positioning accuracy of the above three situations is higher than 99.5%.