Abstract:Due to the characteristics of early minor faults in industrial processes, such as small data amplitudes and strong feature coupling, the detection performance of traditional autoencoder models for these faults is poor. In response, a method for early fault detection in industrial processes based on adversarial quadratic autoencoders and ensemble learning is proposed. Initially, a quadratic neuron is introduced into the hidden layer of a conventional autoencoder model to enhance its expressive power. Subsequently, an adversarial quadratic autoencoder is introduced, incorporating a GAN network during training to enforce feature learning to adhere to specific probability distributions. Then, employing the concept of ensemble learning, normal operational data is sampled, and each subset is used to train an adversarial quadratic autoencoder. Subsequently, two matrices, SPE and T2 statistical quantities, are generated for each subset model. A fusion strategy utilizing singular value decomposition within a single-step sliding window is employed to utilize the maximum singular value within each window as a detection statistic. The proposed method is validated using a numerical example and the TE process, demonstrating its effectiveness in early detection of minor faults in industrial processes.

Abstract:There is an increasing demand for large bandwidth signal testing in the fields of new generation wireless communication, internet of things and virtual reality, etc. Signal analyzers are the primary tool for studying the spectral composition of signal. To enhance the large bandwidth signal testing capability of signal analyzers, a wideband signal acquisition system based on digital bandwidth interleaving is studied and designed; aiming at the problem of deterioration of the in-band frequency response caused by the gain imbalance of each channel of the acquisition system and stitching of frequency bands, a method of correcting the in-band frequency response error by designing a compensation filter based on multi-carrier testing and the least squares method is proposed. The actual experimental results show that the designed system can effectively improve the signal analyzer’s analysis bandwidth. The in-band amplitude flatness of the 1.8 GHz bandwidth acquisition system is ±0.97 dB, the phase linearity is ±0.11 rad, and the in-band frequency response error correction effect is better than the mainstream method of frequency sampling method.

Abstract:Islanded DC microgrid systems need to be configured with multiple DESUs to cope with different load demands and energy fluctuations. Considering DESUs with different capacities, the SOC balancing effect is seriously affected due to the impedance mismatch in the line; for this reason, a multi-storage SOC balancing control strategy considering different capacities is proposed. Firstly, the relationship between SOC and the droop coefficient is established by an exponential function containing an acceleration factor, which improves the sensitivity of the droop coefficient to the late change of SOC balancing and realizes the adaptive adjustment of the droop coefficient. Each DESU only exchanges state information with neighboring nodes through a sparse communication network and adopts a dynamic consistency algorithm to obtain the global average state information. A multi-objective controller is designed, capable of compensating bus voltage dips while achieving accurate distribution of load power, eliminating the effect of mismatched line resistance, reducing the control process, and alleviating the system communication burden. Finally, a simulation model is built based on MATLAB/Simulink software, and the effectiveness of the proposed control strategy is confirmed under various complex operating conditions.

Abstract:In the field of autonomous driving technology, the identification of pavement roughness directly influences subsequent driving decision-making processes. However, existing algorithms for pavement roughness recognition suffer from issues of low accuracy and slow recognition speed. Addressing this challenge, a Hidden Markov Model based pavement roughness recognition method is proposed, leveraging an improved multi-scale feature extraction network. Significant enhancements in both recognition accuracy and speed are achieved by an enhanced multi-scale convolutional neural network, which autonomously learns and extracts hierarchical features from raw data. Subsequently, t-SNE visualization is applied to the extracted features for improved understanding and analysis of feature distributions. Finally, a Hidden Markov Model is utilized for feature recognition. Experimental results demonstrate recognition accuracies of 99.6% for simulated data and 98.6% for real-world collected data, thereby proving effective for pavement roughness recognition.

Abstract:For the measurement of seawater low conductivity environments such as salty tide, this paper proposes a single-coil measurement model based on magnetic resonance and eddy current technology. An equivalent coupled circuit model of seawater eddy current applicable to single-coil magnetic resonance is established, and the specific influence of eddy current loss on the circuit parameters is theoretically analyzed. The influence of series and parallel resonance on seawater conductivity measurement is investigated by comparing theory and simulation, and it is proved that the variation of the imaginary part of the voltage of the parallel load is more stable than that of the series resonance. The single-coil seawater conductivity measurement system was built for laboratory tests, and the test results show that the loop resonance has an obvious enhancement effect on seawater eddy currents, the load voltage at resonance varies linearly with the growth of conductivity, and the parallel resonance method is better than the series resonance method in the measurement. The magnetic resonance single-coil model has the feasibility of seawater conductivity measurement and has a good development prospect in the direction of low conductivity, non-contact sensors.

Abstract:The speed and acceleration planning of the traditional 3-5-3 polynomial interpolation trajectory planning algorithm is too conservative, which is far from the motion limit conditions of the manipulator, and does not give full play to its motion performance, which leads to the increase of the time for the manipulator to complete the task. To solve the above problems, this paper proposes a 3-5-3 polynomial interpolation trajectory planning algorithm based on improved PSO-GSA algorithm. Firstly, the adaptive inertia weight and dynamic learning factor are introduced to improve the PSO-GSA algorithm, and then the improved algorithm is used to optimize the time of 3-5-3 polynomial interpolation algorithm. In the optimization process, the particle group with overspeed used a different fitness function than the one without overspeed, which led the particle group to move closer to the direction of the joint speed decrease, and accelerated the convergence speed of the algorithm. The simulation results show that the improved PSO-GSA algorithm has faster convergence speed, higher search accuracy and is not easy to fall into local optimum than the original algorithm and some similar algorithms. The running time of the 3-5-3 polynomial interpolation trajectory planning method is reduced by 22.9% compared with that before optimization. The obtained trajectory meets the speed limit and is smooth and stable, and the operation is safer and more efficient.

Abstract:This paper introduces the importance of surface wave attenuation rate of radar absorbing materials and its measuring method. Using the radar equation, this paper analyzes the influencing factors by the measurement of radar scattering cross section under near-field conditions. The antenna beam width and spacing between transmitting antenna and receiving antenna are important error sources of radar cross section measurement. A scheme is proposed to reduce the measurement error by increasing the 1 dB beam width of the direction diagram and reducing the distance between the transmitting antenna and the receiving antenna. The 8~12 GHz conical horn antenna is designed and manufactured. The antenna has good circular symmetry, and the 1 dB beam width reaches 22°. A near-field RCS measurement system is set up，validation experiments of the system is conducted with metal cube. Compared to theoretical values，the difference of the RCS curve peak values are less than 0.5 dB, and the difference of the RCS curve average values are less than 0.2 dB. It can been seen that the system has a relatively high measurement accuracy. Then the surface wave attenuation rate experiments are conducted with the near-field RCS measurement system, the difference of the measurement result between near-field RCS measurement system and compact range is 0.38 dB, so the measurement method of this paper can accurately measure surface wave attenuation rate under near-field conditions.

Abstract:In order to solve the problem that the least square projection twin support vector clustering (LSPTSVC) algorithm fails to make full use of the potential information among sample neighborhoods and is not practical, this paper proposes an efficient weighted least square projection twin support vector clustering algorithm with neighborhood information. Firstly, the algorithm introduces the concept of relative density to fully extract local similarity information between data points of the same class. Then, the algorithm calculates the relative weight of the point. Finally, in order to better reflect the geometric structure of similar samples, the algorithm calculates the weighted average value of the data points by using the relative weight. These experimental results verify the effectiveness of the algorithm. The results show that the proposed algorithm achieves better clustering accuracy than the existing methods under the similar computational complexity and good clustering performance in the practical application of medical datasets in the real world.

Abstract:Aiming at the low detection accuracy caused by similar defect features, small defect target and large difference in defect scale in SOP chip defect detection, this paper proposes a defect detection method based on improved YOLOv8. The problem of information loss in the process of convolution pooling is solved by using SPD-Conv module. And introducing the SimAM attention mechanism, the model can learn the information in the 3D channel and improve the model′s perception of defect features. At the same time, BiFPN was used to replace the original feature extraction network, and multi-scale feature fusion was used to enable the model to better distinguish the defects with similar features and large-scale differences. Finally, a small target detection header is added to transmit more low-order feature information to the high-dimensional detection network to improve the detection effect of small target defects. Experimental data show that compared with the original model mAP@0.5/% increased by 5.4%, mAP@ 0.95/% increased by 4.3%, recall rate increased by 3%, has significant advantages compared with other models. In the generalization experiment, the mAP@0.5 of the improved algorithm is also improved by 2.7% compared with the original model, and a relevant system is designed to verify the effectiveness of the algorithm.

Abstract:Aiming at the problem that PV arrays show multipeak mutation under uneven light conditions, which leads to the imbalance of the traditional maximum power point tracking, an MPPT control based on the Improved Parrot Algorithm is proposed. Firstly, the Halton sequence is considered for population initialization to make the diversity change significant; secondly, the tangent flight mechanism in the Tangent Search Algorithm is selected to reduce the dynamic jumps and overcome the problems of precocity and local polarity; and later, the secondary update is performed by the parrot somersaulting foraging strategy to reduce the adaptive range and accelerate the convergence. Comparing the traditional parrot algorithm, the golden jackal algorithm and the gray wolf optimization algorithm, the test results show that the improved parrot algorithm′s MPPT control tracking efficiency is 98.23%, 97.26%, 96.91% and 96.81%, and the tracking time is 0.077 s, 0.112 s, 0.127 s, and 0.156 s, respectively, and the tracking accuracy and rate are significantly higher than the remaining three algorithms.

Abstract:In view of the problem of the limited vector action direction and amplitude of the control system in the prediction current control strategy, an optimized three-vector model prediction current control strategy is proposed. First, a virtual vector is introduced to extend the voltage vector control set to provide more flexible control choices for the system. Secondly, the reference voltage vector is obtained according to the difference-free control principle, and the specific position of the reference voltage vector in the sector is determined. Taking the voltage vector as the basic element of the value function, it analyzes the difference between the reference voltage vector and the effective voltage vector. The optimal effective voltage vector is obtained according to the designed screening rules, which simplifies the voltage vector screening process. The experimental results show that, compared with the traditional model predictive current control, the control strategy proposed in this paper significantly reduces the current and torque ripple, and presents a smoother speed curve and smaller overshoot in the speed of the motor. the dynamic response performance and control ability of the motor are improved.

Abstract:A lightweight detection algorithm based on improving YOLOv8 is proposed to address the issues of high complexity, false alarms, and missed detections in current PCB surface defect detection methods. Due to some redundancy in the feature maps of the YOLOv8 backbone network after downsampling, a lightweight multi-scale mixed convolution (MSMC) is designed. This is combined with the C2f module to enhance the capability of extracting features at different scales. Additionally, an improved Bidirectional Feature Pyramid Network (BiFPN) structure is designed in the neck network, using two cross-layer connections to obtain richer semantic information. The C2f-Faster module is employed to reduce computational complexity during the feature fusion process. Moreover, the introduction of the CA attention mechanism and the WIoUv2 loss function strengthens the ability to locate small defects on PCBs. The experimental results show that the improved algorithm compared to YOLOv8n improves the detection accuracy by 2.2% on the PCB dataset, while the number of model parameters and the computation volume are reduced by 36.7% and 18.5% to 1.9 M and 6.6 G. The final model size is only 3.8 MB, providing a new approach for mobile terminal device deployment.

Abstract:Aiming at problems such as the dung beetle optimization algorithm′s poor global exploration ability and its tendency to fall into local optimization, this paper proposes a hybrid algorithm based on the positive cosine algorithm and the dung beetle optimization algorithm called the SCDBO algorithm. The hybrid algorithm adopts the positive cosine search algorithm instead of the search mechanism of the rolling dung beetle in the dung beetle algorithm, which balances the global search and local exploitation ability of the algorithm. In addition, while introducing the t-distribution perturbation to update the dung beetle population with a certain probability during each iteration, the Levy-Corsi variation operator is introduced to mutate the optimal position. This not only accelerates the convergence speed of the algorithm but also reduces the possibility of falling into the local optimum. Finally, the population diversity of the algorithm is enhanced by initializing the dung beetle population with chaotic mapping. The effectiveness of the SCDBO algorithm is investigated using 23 benchmark functions, and the experimental results show that the algorithm exhibits a better ability to find the optimum compared with other comparative algorithms. To further evaluate the performance of the SCDBO algorithm for practical applications, the algorithm was successfully applied to three engineering design problems. By comparing with other algorithms, the results show that the SCDBO algorithm has high potential in solving practical engineering problems.

Abstract:Aiming at the problem of low estimation accuracy of traditional estimation methods, this paper proposes a novel state of health (SOH) estimation method based on the framework of spatio-temporal convolutional network and Pyramid Squeeze Attention Fusion (TCT-PSA) model.Convolutional Transformer-Pyramid Squeeze Attention (TCT-PSA) modeling framework for a novel state of health (SOH) estimation method. The NASA battery dataset was first preprocessed, and then the health factor (HF) was extracted from the charging stage of lithium-ion batteries, and the correlation between HF and SOH of lithium-ion batteries was quantified by using Pearson′s correlation coefficient and grey correlation analysis, and the HF with high correlation was inputted into the TCT-PSA model, and SOH was the model output. In order to verify the validity of the model, the TCT-PSA model was used to estimate the capacity degradation of each group of batteries; the SOH of each group of batteries was estimated using different models and compared, and the quantile estimation was used to verify the accuracy and robustness of the TCT-PSA model. The experimental results show that the errors of the proposed models are within 2% by validating the average absolute estimation errors of the capacity decays in the test and training sets of each group of batteries; the average absolute error (MAE), the average absolute percentage error (MAPE), and the root mean square error (RMSE) of the proposed models for estimating the SOH of each group of batteries are within 0.035; and the highest accuracy of the quartile estimation of the SOH of lithium-ion batteries is 99.82%. The highest accuracy reaches 99.82%.

Abstract:At present, YOLO object detection algorithm is still the most mainstream method in the field of insulator defect detection, however, the existing YOLO model framework has a large number of parameters leading to the difficulty of outdoor deployment, and at the same time, the background of the insulator images taken outdoors is complex, and the defects are even more tiny, making it very difficult to be detected. To address the above problems, this paper proposes an improved insulator defect detection model YOLOv8-GCS based on the YOLOv8n object detection framework to reduce the number of parameters in the model and improve the detection precision of the model. Firstly, the C2f block in the model is replaced by a more lightweight Ghost convolution block to reduce the computational and parametric quantities of the model. Then the Coord Attention module is added at the end of the backbone network and at the second detection head to suppress the influence of the complex background on the defective parts of insulators and thus improve the detection precision of the model. At last, an SPD-Conv block is introduced so that the model of the network has no loss of important information in the process of two-fold downsampling and at the same time enhances the learning rate of the model of the network on the important features, which further improves the detection performance of the model. Analyzing the experimental results, it can be seen that the algorithm in this paper improves the mAP50 by 4% compared with the baseline model, the recall rate and the check all rate by 4.7% and 1.3%, respectively, the number of parameters is reduced by 26.7%, the size of the weight file to save the results is reduced by 1.5 MB, and the AP50 of insulator broken and pollution-flashover are improved by 4% and 8.1%, respectively.

Abstract:Aiming at the problem that patients with dysphonia lack clear and accurate pronunciation, which leads to low pathological speech recognition rate, an improved Gammatone Filter Bank map feature extraction algorithm based on LMD is proposed for three channel pathological speech recognition. Firstly, the algorithm uses LMD to decompose speech signals, performs short-time Fourier transform on each decomposed speech component, and synthesizes frequency to extract filter bank features and their first-order and second-order differential features, forming LMD-GFbank map features that can obtain effective local features of pathological speech. Secondly, in order to further improve the problem that the network model will miss some effective feature information during the training process, a three-way pathological speech recognition model is proposed. Finally, the pathological speech recognition model is trained and tested by combining the speech feature information. The experimental results show that the recognition rate of LMD-GFbank map features on the three channel pathological speech recognition model reaches 93.36%, which is better than the speech recognition performance of traditional MFCC, GFCC, and Fbank features, and verified that the proposed algorithm and recognition model can improve the accuracy of pathological speech recognition.

Abstract:In high-speed online visual measurement systems, it is difficult for small depth of field cameras to obtain clear images of calibration objects from different directions, resulting in the Zhang calibration algorithm having no solution or significant calculation errors. Therefore, a fast and accurate calibration algorithm for small depth of field cameras is proposed. Obtain a set of calibration object images located on the plane to be calibrated and parallel to it, linearly solve to obtain the height of the camera lens optical center to the plane to be calibrated, simplify the rotation matrix based on the camera posture, and finally use the calibration images located on the plane to calculate the camera′s internal and external parameters. The experimental results show that it is best to set the number of calibration image groups around 13, but using only one set of calibration images can also achieve high-precision calibration. In cases where the depth of field of the camera is small, this method has a root mean square error of less than 0.74 pixels for reprojection, and the calibration accuracy is improved by about 33% compared to Zhang′s calibration.

Abstract:A high-precision infusion monitoring method based on improved YOLOv8n network was proposed to address the difficulties of insufficient accuracy and inconvenient installation of visual sensor infusion monitoring methods. Based on the original network, the PuzzleMix data augmentation method was used to improve its generalization ability and to avoid cutting key features. The high-level screening-feature fusion pyramid networks structure was introduced to reduce the number of parameters and enhance the expression ability of droplet features. The Mixed local channel attention was included to enhance droplet feature extraction. The Inner-PIoU was proposed to improve the loss function, which utilized auxiliary regression anchor box to improve regression performance and accuracy. Meanwhile, a method used the ratio of geometric parameters of the detection box was proposed for accurately measuring infusion speed and the remainder. The experimental results show that compared with YOLOv8n, the mAP@0.5:0.95 is increased by 2.674%, and the model size is only 3.87 M. In various complex infusion environments, the proposed method can achieve accurate monitoring of infusion speed and the remainder.

Abstract:As the main means of Marine traffic warfare, it is of great significance to identify ship targets efficiently and accurately in remote sensing images. Although the optical remote sensing ship image contains rich information, the recognition rate is low because of its high complexity, large image and the influence of weather and day and night change. To solve this problem, this paper proposes a more efficient optical remote sensing ship image classification method by improving ResNet18. The ResNet18 network is simplified and its parameter number is reduced. The parallel Pooling is used to reduce the dimensionality of the feature graph space to speed up the network convergence while keeping less feature loss. The multi-scale convolution is introduced to extract feature information of different scales, and the ECA attention mechanism is used to improve the multi-scale convolution module and residual module to solve the problem that features can′t interact well between channels in branch network branch fusion. Experiments were carried out on the FGSCR.42 dataset, and the experimental results show that the improved algorithm converges faster, and the accuracy and F1-score are up to about 95%, which is about 7% higher than that of the ResNet18 network, while the number of parameters is only about 20% of that before the improvement. Compared with the performance of other networks in ship target recognition, the proposed method also has better performance.

Abstract:Improving image quality and visualization in complex deep-sea environments is of great importance for underwater scientific research and engineering applications. In order to solve the problems of scarcity of deep-sea datasets caused by the special environment of the deep-sea, as well as the problems of color distortion and low contrast of deep-sea images, a paired deep-sea image dataset DSIEB was created, and on this basis a generative adversarial DM-GAN was built algorithm proposed for networks combining DC attention and MSDR multi-scale dense residuals. First, the DC dual-channel attention mechanism is built in the network hopping connection part to strengthen the connection between channels and extract the texture features of image details. Second, the MSDR multi-scale residual block is embedded in the generator structure to improve the attention is on local information and the ability to reuse features. Finally, a new loss function is reconstructed and the smoothing fidelity SF loss is introduced to guide the network to learn the mapping from the original image to the target image from multiple perspectives. Experiments are carried out on the self-built dataset DSIEB, and compared with seven advanced underwater image enhancement algorithms. The experimental results show that the proposed algorithm has stronger generalization ability and is suitable for various deep-sea images.

Abstract:This paper presents a segmentation network that utilizes global feature guidance and attention to enhance the precision of cell segmentation, addressing the variability in size and shape of normal and abnormal nuclei in cervical cells as well as interference in cell images. Firstly, the U-shaped network structure is utilized as the main framework, with the introduction of a global feature guide module to comprehensively extract features at each stage for obtaining global context information at different levels. This overcomes the limitation of insufficient extraction ability of single-stage context information in U-shaped networks, enabling better handling of nuclei with different shapes and improving edge segmentation accuracy. Secondly, an improved attention-gate structure is incorporated to suppress interfering information in images, emphasize nucleus information, and enhance model discrimination against interfering data. Experimental results on the Herlev dataset demonstrate that our proposed method effectively enhances nuclear segmentation precision, achieving a Dice coefficient of 0.941 3 in quantitative analysis which presents certain advantages compared to other methods.

Abstract:Bidirectional Reflection Distribution Function (BRDF) describes the scattering and reflection characteristics of material surfaces. Considering the complex structure, difficult disassembly, and difficult optical path debugging of current BRDF measurement systems, a portable BRDF measurement system has been designed. The measurement system mainly consists of three parts: the light source system, the detection system, and the angle device. The mechanical structure of the angle device and the detection part has been optimized. In order to verify the comprehensive performance of the designed measurement system, experimental measurements were conducted on the surface of typical coating materials using the measurement system, and the influence of incident angle on the surface reflection characteristics of objects was analyzed. The experimental results show that the larger the surface roughness of typical coating materials such as quartz, alumina, and zirconia, the smaller the proportion of mirror reflection. Between 0° and 60° incidence angles, the surface BRDF value also increases with the increase of incidence angle. The measurement data predicted by the improved Torrence Sparrow model was used for fitting analysis with the experimental data. The fitting results showed that the measured experimental data had a good fit with the theoretical model prediction data, with a minimum fit greater than 96%.