Abstract:The underwater swimming manipulator (USM) is a new type of underwater robot composed of an underwater snake robot and several thrusters. The USM system has the characteristics of high nonlinear and uncertainty, and its dynamic model is difficult to establish accurately. Therefore, it is challenging to achieve high precision stabilization control of USM. To solve this problem, this paper designs a dynamic control framework based on feedback linearization and adaptive radial basis function neural network (RBFNN) for USM stabilization control. Firstly, the structure of the USM platform is introduced, the dynamic model of the USM is established based on the Lagrange equation, and the model of the vector thrust system is derived. Then, a dynamic controller based on feedback linearization and RBFNN is designed, and the weight of RBFNN is updated adaptively by backstepping method. Among them, the weight adaptive updating RBFNN is used to estimate the unmodeled part of the system, parameter errors and external disturbances, so as to compensate the dynamics controller. In addition, in order to convert the generalized forces and torques provided by the dynamic controller into the control inputs of each actuator, a thrust distribution strategy is given. Finally, lake experiments are carried out to stabilize the I-shape and C-shape of USM respectively. Compared with traditional methods, the steady-state errors of the proposed control scheme under both configurations are less than 0.08 m and 10°, which verifies the effectiveness of the proposed 6-DOF USM stabilization controller.

Abstract:Aiming at the problem of temperature drift of Hall effect force sensor, a new temperature compensation model of chaotic adaptive whale optimized BP neural network (CIWOA-BP) was proposed. This model uses Cubic mapping as the initial whale population generation method to improve the quality and distribution uniformity of the population. The adaptive weight was introduced to adjust the shrinking and bounding mechanism of the whale to improve the global search ability and convergence of the algorithm. The CIWOA algorithm is used to optimize the initial weights and thresholds of the back propagation (BP) neural network, so that the model has better measurement accuracy and stability. Research results indicate that after temperature compensation, the temperature coefficient of sensitivity for the Hall effect force sensor decreases from 5.08×10-3/℃ to 9.8×10-5/℃, reducing by two order of magnitude. The temperature-induced relative error decreases from 19.82% before compensation to 0.38%, which is reduced by over 52 times, effectively mitigating the influence of temperature on measurement results.

Abstract:Atmospheric polarization mode are stable natural attributes with widespread applications in navigation, detection, and other fields. However, due to the influence of natural environments and surrounding structures, the polarization information obtained at the same time is often local and discontinuous, impacting its practical use. Existing methods mainly focus on repairing large-scale images of atmospheric polarization mode, resulting in limited accuracy in restoring high-frequency signals and causing edge blurring. To address this issue, this paper proposes a method of soft segmentation and synthesis for polarization information, which avoids the loss of high-frequency signals by redundantly segmenting and synthesizing the polarization information, thereby mining the high-frequency signal features in each local region. Additionally, based on the spatiotemporal continuity of atmospheric polarization mode, reasonable inference is made to ensure consistency between the reconstructed information and the real information, thereby generating complete and continuous atmospheric polarization information. Experimental results demonstrate that this method effectively reconstructs missing polarization information in atmospheric polarization mode. In practical reconstruction experiments where cloud interference exceeds 40%, the proposed method shows a 26% improvement in SSIM and a 12% improvement in PSNR compared to other methods.

Abstract:Laser measurement enables efficient non-contact real-time measurement and finds extensive application in the field of 3D printing. However, laser measurement is susceptible to interference from various factors such as measurement condition and the external environment, which are complex and difficult to quantify and analyze. Therefore, based on the principle of direct laser triangulation and an analysis of the factors affecting measurement accuracy, this paper proposes a 3D printing in-machine measurement error correction method integrating self-attention and residual neural network (SRNN). Firstly, the factors that affect measurement accuracy are used as input variables to collect laser measurement values and obtain a sample dataset. Then, residual network is employed to extract deep-level features from the sample data, and a self-attention mechanism is introduced to establish connections between influencing factors, resulting in weighted extracted features. Subsequently, the weighted features are learned through a fully connected network to obtain the predicted values of measurement errors. Based on this predicted value, the measurement errors are corrected. A laser in-machine measurement system is built, and experimental verification is conducted using three types of color cards (red, green, and purple) made of the same material. The results show that, compared to convolutional neural network (CNN) and self-attention neural network (SelfNN), the method proposed in this paper achieves the smallest mean squared error (MSE), root mean squared error (RMSE), and mean absolute error (MAE), exhibits the best stability, and yields correction results that are closest to the ground truth. After the laser measurement result is calibrated, the error is reduced from the original ±28 μm to below ±9 μm, significantly enhancing the accuracy and stability of 3D printing laser in-machine measurement.

Abstract:At present, most of the research on students’ classroom learning status focus on single-person online monitoring, and the monitoring of offline classroom with multiple students and complex environment is still in the exploratory stage. A monitoring system for students’ classroom learning status was designed for offline education to monitor students’ classroom attendance and fatigue state of students’ faces in real time. First, DSFD face detection algorithm combined with ResNet deep residual network was used to recognize students’ faces and record students’ attendance. Then, ERT regression tree set algorithm combined with head pose estimation was used to detect the fatigue behavior of yawning and drowsiness. Then, the improved YOLOv5 object detection algorithm added CBAM module was used to detect students’ closed eyes behavior. Finally, a complete set of integrated attendance, fatigue detection of student classroom learning state monitoring system is formed. In the actual classroom test environment, the system can accurately calculate the students’ attendance, and can real-time monitor the fatigue state of yawning, lower head and closed eyes on the face of the students. The detection accuracy rate is more than 90%, and the detection speed is about 14.1 fps, which proves that the system has important use value.

Abstract:Microalgae microscopic image target detection technology is one of the important research directions in fields such as biology and environmental monitoring. The dataset of microalgae images captured by electron microscope exhibits a long-tail data issue. Traditional methods for microalgae detection are notoriously labor-intensive, time-consuming, and heavily influenced by operator expertise. In this context, combining methods to address the long-tail distribution, this paper proposes a target detection algorithm called DDM-YOLO, which combines delayed resampling and knowledge distillation. The approach involves data augmentation for microalgae microscopic images and utilizes delayed resampling for long-tail data. In the second stage, reverse resampling is applied to focus on the challenging minority class samples, thereby enhancing the performance of target detection. Additionally, a lightweight target detection network architecture is designed, and knowledge distillation is employed to reduce model complexity and computational requirements. Experimental outcomes reveal that the DDM-YOLO algorithm achieves an mAP@0.5/% of 77.1%, surpassing the YOLOv5s algorithm by a notable 6.1%. The model parameter size is 3.88 megabytes, a significant 45.4% decrease. This proposed method significantly enhances performance on microalgae microscopic image data and efficiently performs target detection under resource-constrained conditions, substantially reducing the workload of detection personnel.

Abstract:Deep learning has made significant progress in fault diagnosis, but it is mostly an end-to-end intelligent diagnosis with limited application in signal denoising. This article proposes a denoising method based on fully convolutional network (FCN). Firstly, the overall model adopts the encoder decoder architecture, where the encoder part consists of three convolutional layers and the decoder part consists of four deconvolution layers. Secondly, residual connections were introduced to constrain the learning objectives of the model, allowing the model to focus more on noise information during propagation. And in order to enhance the feature extraction ability of the model, non-local blocks (NLB) are introduced in the encoder and decoder. Then, through simulation signal comparison experiments, select the hyperparameters of the network and compare them with current mainstream noise reduction methods to preliminarily verify the noise reduction effect of the model. Finally, the denoising effect of the proposed method was compared and verified through practical cases. The results showed that the method proposed in this paper achieved good application effects in both intuitive observation and denoising performance indicators, and can effectively improve the accuracy of fault diagnosis.

Abstract:In the field of unmanned vehicles, point cloud strength and ground constraints play a very important role in mapping and positioning under large-scale environment. However, existing laser SLAM algorithms only consider geometric features when constructing maps, and neglect point cloud intensity information and ground constraints, resulting in blurry mapping details and drifting in the Z-axis direction, thereby reducing the accuracy of SLAM systems. To this end, this paper proposes a laser SLAM optimization algorithm based on point cloud intensity and ground constraints. Based on the ground measurement model, it is proposed to construct local conditional ground constraints, which not only improves the accuracy of ground point extraction but also reduces the drifting in the Z-axis direction; introducing point cloud intensity information to improve the reliability of non-ground point clustering, further improving mapping accuracy and positioning stability. A feature extraction method based on local smoothness is proposed, in which by introducing intensity factors to rank intensity features, features with consistent intensity information are selected preferentially, enhancing the robustness of feature extraction. The pose is optimized and estimated by constructing strength residuals based on a spherical strength map, together with geometric residuals, effectively solving the problem of blurring in map details in odometry. The matching distance and intensity difference based on feature projection are used to remove interference from dynamic point clouds, further improving the robustness of SLAM systems. Experiments on the public dataset KITTI and real scenarios have shown that the proposed algorithm has higher mapping and positioning accuracies by introducing ground constraints and point cloud strength information. Compared to the LVI-SAM algorithm that outperforms traditional LIO-SAM algorithm, the proposed algorithm in this paper is improved by 54.5% in accuracy, providing a reliable solution for SLAM tasks of unmanned vehicles in large-scale environment.

Abstract:A distributed adaptive iterative learning control strategy is proposed for the formation problem of nonlinear multi-agent systems with unknown time-varying parameters. Firstly, the uncertain parameters of the system are expanded through Fourier series, and a convergent series sequence is employed to handle the truncation error resulting from the Fourier series expansion. Combined with the formation error during the operation of the multi-agent system, the adaptive iterative learning control law and parameter update law are derived. Secondly, for scenarios where the dynamics of the leader are unknown to most agents, a new auxiliary control is designed to compensate for the unknown dynamics and avoid unknown bounded interference. Then, based on the Lyapunov energy function, it is proved that the formation error of the multi-agent system tends to be zero within a limited time as the number of iterations increases under the action of the designed control law. Finally, this control strategy is applied to multi-UAV formation systems, and its effectiveness is validated through the construction of a semi-physical experimental platform. Experimental results demonstrate that this control method can ensure rapid formation of the required formation by multiple agents, and each agent can accurately track the desired trajectory within a limited time. The proposed method fully considers the parameter uncertainty and anti-interference ability of multi-agent systems, providing an effective approach for the precise control of complex multi-agent systems in practical applications.

Abstract:To address the issue of output voltage disturbance in Buck converters under complex environments with load fluctuations, a composite adaptive prescribed performance control scheme is proposed to enhance control effectiveness. Initially, an adaptive law is utilized to predict and estimate the nonlinear function containing the load term within the model. Concurrently, a parallel estimation model is constructed during the adaptive law update process to acquire prediction errors, which are then integrated with tracking errors to design an adaptive parameter update law. Subsequently, a generalized proportional-integral observer is employed to estimate the remaining uncertainties and external disturbances, which are compensated for within the control law. Finally, combining command-filtering backstepping control and specified-time prescribed performance control techniques, a composite adaptive prescribed performance control scheme for Buck converters is proposed. The presented scheme ensures high-precision prediction of load fluctuations, preventing output voltage from exceeding the prescribed function range during sudden events, and also demonstrates the signal convergence within the closed-loop control system. Experimental results indicate that the composite adaptive prescribed performance control, when faced with a sudden reduction in load, limits the system’s maximum voltage deviation to 0.376 V, a 78.7% decrease compared to the traditional adaptive backstepping control’s 1.773 V, thereby validating the effectiveness and superiority of the proposed scheme.

Abstract:An edge-preserving smoothing algorithm based on local Gaussian mean-difference variation is proposed to address the issue of detail not being preserved during the process of image smoothing. Firstly, a local Gaussian mean-difference variational operator is established by statistical analysis. To differentiate between structure and texture, the operator is employed to quantify the difference between the local gradient and the gradient after Gaussian filtering. Secondly, a local Gaussian mean-difference variational smoothing model is developed, and a sparse solution is used to produce the initial smooth image. Finally, an isolated noise removal model is suggested to address the issue of texture residue in images with complex texture. The model adjusts pixel values using an adaptive window and eliminates texture residue from the initial smooth image without changing the structure. It has been demonstrated through subjective and objective experiments that this algorithm produces smoothing results of superior quality than traditional algorithms. Evaluation indicators improved by 0.7% overall. Extended experiments verify the algorithm's applicability and efficiency enhancement potential across various visual tasks, including compression artifact removal, HDR tone mapping, image dehazing, and accelerated Laplacian pyramid.

Abstract:Aiming at the current problems of low efficiency and poor detection accuracy of steel surface defects, a model, named ECC-YOLO, is proposed for steel surface defects detection based on YOLOv7. Firstly, in order to improve the capability of feature map information characterization of the backbone network, a feature enhancement module ConvNeXt is introduced, which enhances the feature extraction capability of the model for fine cracks by fusing the depth separable convolution and the large kernel convolution, secondly, a C2fFB module is designed, which enhances the capability of extracting the feature information of the target and at the same time, reduces the computational volume and parameter complexity of the model significantly. Finally, the MPCE module is designed with the help of the ECA attention mechanism to weaken the interference of the complex background information on the steel surface defect detection and improve the detection efficiency. Finally, extensive experimental results show that the mAP of the model of ECC-YOLO reaches 77.2% on the NEU-DET dataset, and compared with YOLOv7, the detection accuracy of ECC-YOLO is improved by 10.1%, and the number of model parameters is reduced by 9.3%, which gives the model a better comprehensive performance in steel surface defect detection.

Abstract:The series arc fault is mainly caused by poor contact of the electrical contact points in the circuit, which is one of the main causes of electric vehicle fires, directly threatening the life safety of the occupants. In order to study it, an experimental platform for DC series arc fault of electric vehicles was established. The voltage signals of the power supply terminal were obtained under various working conditions, and the impact of arc faults on the power supply terminal voltage was analyzed. When constructing the detection model, the paper used a convolutional neural network, introduced a lightweight convolution operation, and considered its limitations in practical applications. Combining conventional convolution and lightweight convolution operations, a preliminary model for arc fault detection was constructed. Then, with the scale and accuracy of the network as the evaluation index, the genetic algorithm with elite preservation strategy was used to search for the external structure and internal parameters of the model. Finally, the model AFDNet suitable for arc fault detection (AFD) of electric vehicles was established. The detection accuracy of the model is 93.73%, and the running time on the embedded device Jetson Nano（JN） is 10.82 ms. After establishing the model, the paper compared the search results of the algorithm with other network structures in relation to network size, accuracy, and real-time performance, verified the validity of the results obtained by the search algorithm. By comparing AFDNet with other detection methods, it was proven that the performance of the electric vehicle arc fault detection model was superior.

Abstract:Aiming at the problems for which the first predicting time (FPT) of bearing remaining useful life (RUL) is based on subjective selection and maintenance risks caused by predictive lag. A stochastic configuration networks (SCNs)-based bearing residual life prediction method is proposed. Firstly, the complementary ensemble empirical mode decomposition (CEEMD) is used to decompose the original bearing horizontal vibration signal, then extract its time-domain and frequency-domain signals to construct fusion features. Secondly, the health state is divided by wavelet clustering to find the appropriate FPT, and the health data set is constructed by combining the characteristics of the energy response bearing degradation. The prediction is made by SCNs network offline modeling, and the prediction results are corrected according to the slope of the fitted curve and the RMSE index. Through experimental analysis, the comprehensive score of the proposed method is as high as 0.83, and the mean absolute deviation (MAD) and standard deviation (SD) of the error percentage are 5.26 and 3.38. Compared with other prediction methods, the proposed method has higher prediction accuracy.

Abstract:Aiming at the problem that a single intensity image lacks polarization information and cannot provide sufficient scene information under bad weather conditions, this paper proposes a dual-attention mechanism to generate an adversarial network for fusion of intensity and polarization images. The algorithmic network consists of a generator containing an encoder, a fusion module and a decoder and a discriminator. First, the source image is fed into the encoder of the generator, after a convolutional layer and dense block for feature extraction, then feature fusion is performed in the texture enhancement fusion module containing the attention mechanism and finally the fused image is obtained by the decoder. The discriminator is mainly composed of two convolutional modules and two attention modules, and the generator network parameters are iteratively optimised by constant gaming during the network training process, so that the generator outputs a high-quality fused image that retains the sparse features of the polarimetric image without losing the intensity image information. Experiments show that the fused images obtained by this method are subjectively richer in texture information and more in line with the visual perception of the human eye, and that the SD is improved by about 18.5% and the VIF by about 22.4% in the objective evaluation index.

Abstract:The dropper clamp bolt is an important component of railway power supply line, which can affect the flow quality of electric locomotive. Therefore, this paper improves the SSD algorithm: Firstly, a lightweight neural network MobileNetV3 is introduced for front-end feature extraction to reduce the model complexity and improve the detection speed; secondly, CA attention mechanism to replace the SE module of the linear bottleneck layer with inverted residuals structure, aggregate the position information in the two directions of space, and the adjusted feature layer can capture the global remote feature information. Finally, the feature fusion module for reconstructing the feature layer is designed to adjust the small target detection layer to improve the detection effect of small targets. This paper also expands the training sample with CycleGAN to solve the problem of insufficient data set. The experimental results show that the model complexity of the improved algorithm decreased, and mAP @ 0.5 and FPS reached 95.5% and 81 fps, respectively. This study helps the transformation of catenary detection instruments to small mobile embedded devices.

Abstract:To mitigate the influence of nonline of sight (NLoS) errors in ultra-wideband (UWB) ranging, this study presents a method that utilizes a genetic algorithm backpropagation neural network (GA-BP) for error identification and optimization. This method effectively detects and rectifies ranging errors and system deviations occurring in the NLoS propagation link, and subsequently improves the ranging outcomes through the application of Kalman filtering (KF). On this basis, this paper proposes a weighted concentric circle clustering localization (WCCGT) method to address the problem of no intersection or multiple intersection points in multilateral positioning caused by ranging errors. The method solves the problem of no intersection points through weighted concentric circle generation (WCCG). Then, it uses the mean shift clustering localization method to achieve a localization solution and improve localization accuracy. The experimental results show that the improved ranging optimization method effectively reduces the ranging error in the NLoS propagation link, and the ranging accuracy based on UWB is improved by more than 60%. Analyze through static positioning experiments and dynamic experiments, the positioning results of the WCCGT method were compared with the least squares (LS) method. The proposed method can achieve a positioning accuracy of 10.78 cm in NLoS environments, and the positioning performance has been improved by 17.32%.

Abstract:In order to meet the demand of voiceprint recognition in flammable and explosive environment. A linear Sagnac interference optical fiber acoustic sensor system has been designed. Speech data was denoised using the Wiener filtering algorithm, and pitch features were extracted through three-level clipping. Speaker samples were screened using dynamic time warping, and Mel-frequency cepstral coefficients were extracted as features. Voiceprint recognition experiments were conducted utilizing the Gaussian mixture model-expectation maximization algorithm, concurrently investigating the frequency response characteristics of the optical fiber acoustic sensor system and their relationship with voiceprint features. The influence of the amplitude of acquired speech on voiceprint recognition outcomes was studied. Experimental results demonstrate that the system can realize the sound signal perception in the frequency range of 300~3 500 Hz. When the sound amplitude decreases from 0.9 to 0.15 V, the difference between the maximum and second-largest log-likelihood values drops from 35.5 to 10.9, the recognition result changed from success to failure. Repetition experiments show that, at a distance of 2 meters from the sound source along a 10-kilometer sensing fiber, the system accurately recognizes 400 speech segments of 3 to 5 seconds duration, unrelated to any specific text, achieving an overall identification accuracy rate of 94.75%. This system holds promise as a solution for voiceprint recognition in applications such as equipment fault diagnosis and emergency response within flammable and explosive environments.

Abstract:Person re-identification is highly used in the areas of traffic management, searching for lost people, etc. It is hard for existing algorithms to deal with the problem of human pose change, occlusion and feature misalignment, and a pose-guided and feature-fused pedestrian re-recognition algorithm is proposed. The proposed algorithm includes three branches, including global branch, global branch based on pose estimation guidance, and local alignment branch. The global branch extracts the global features of pedestrians and captures the coarse-grained information of pedestrians. The global branch based on posture estimation guidance uses the posture estimation network guidance model to focus on the global visible area of pedestrians and reduce the interference of occlusion to pedestrian recognition. Local alignment branch uitilizes a pose estimation algorithm to establish aligned local features while distinguishing visible local regions to reduce occlusion as well as the influence of postural changes. Through a multi-branch structure, integrated local characteristics with global ones to augment feature diversity is achieved and enhanced model robustness. Finally, network training is conducted using cross-entropy and triplet loss functions. The viability of the proposed algorithm is validated by the test results on Market-1501 and DukeMTMC-ReID datasets, during which the Rank-1 and mAP of the DukeMTMC-ReID dataset reached 91.2% and 81.8%, respectively, which has a better practicality.

Abstract:To address various challenges associated with conventional laminar flowmeters (LFM), such as inadequate linearity, significant length-diameter, inconvenient processing and use, and susceptibility to fluid-induced effects, a novel annular-gap laminar element structure is proposed, drawing inspiration from the double-cone flowmeter. This innovative design is accompanied by a comprehensive elucidation of its measurement principles and an analysis of the sources contributing to non-linear pressure losses within the flow conduit. Central to this design is the maintenance of coaxial alignment between the outer jacket cylinder and the cone and circular cylindrical, resulting in a flow channel characterized by concentric circular annual. Computational fluid dynamics (CFD) simulations were leveraged to ascertain the optimal cone angle of the conical guiding structure and establish the dimensional parameters of the laminar element. Furthermore, pressure taps were strategically positioned within the fully developed laminar segment of the flow channel, thereby theoretically mitigating localized losses at the inlet and outlet typical of conventional capillary-type LFM, as well as kinetic energy dissipation within the laminar development region. Experimental validation involved the fabrication of three distinct test specimens with varying gap dimensions, followed by rigorous testing. Results revealed that for flow rates below 53 mL/min, the measurement error of the laminar element remained within an acceptable margin of 3%. Likewise, within the flow rate range at (130~6 189) mL/min, the measurement error was constrained within the range of ±2%. Notably, a robust linear relationship between pressure drop and flow rate was observed, affirming the efficacy of the proposed design in circumventing the non-linear influences inherent in traditional LFM. The elucidation asserts the structural efficacy of annular gap laminar flow elements in effectively mitigating the nonlinear influences characteristic of traditional LFM. Simultaneously, it highlights the adaptability of the measured flow range, which can vary with alterations in gap size.

Abstract:Aiming at the problem that vibration signal of inter-turn short circuit fault in permanent magnet synchronous motor (PMSM) is easily affected by noise and it is difficult to accurately extract the fault feature of it, an improved whale optimization algorithm (IWOA) optimized variational mode decomposition (VMD) denoising method is proposed and applied to vibration signal of inter-turn short circuit fault in PMSM. Firstly, the nonlinear convergence factor, adaptive weight and the Cauchy operator are introduced into the traditional whale optimization algorithm, and the IWOA algorithm is used to optimize the VMD parameters to achieve adaptive signal decomposition. Secondly, according to the principle of selecting the optimal intrinsic mode function based on multi-scale permutation entropy and variance contribution rate, the signal components are divided into the noise-dominated components and effective signal components. The noise-dominated components are denoised by the non-local mean filtering (NLM). Finally, the denoised and effective signal components are reconstructed as denoised signal. A motor short circuit fault model is established using ANSYS finite element software, and a short circuit fault experimental platform is built. Using this method to denoise the simulated and measured signals, it is further compared with many denoising methods such as wavelet threshold denoising method. The signal to noise ratio of the simulated signal is improved from 8 dB to 20.273 8 dB, and the signal to noise ratio of the measured signal is improved by 77.01% compared with wavelet threshold denoising method, which proved the effectiveness and practicality of the proposed method.

Abstract:At present, China has not yet publicized the national calibration specification of respiratory tester, which makes it difficult to carry out the metrological traceability of various parameters of respiratory tester. Some self-compiled calibration methods for gas flow calibration have the problem that the flow calibration point is affected by the size of the plunger, which in turn limits the flow calibration range. To solve this problem, this paper proposes a calibration method based on the standard meter method and plunger method for static flow and tidal volume, respectively, and integrates two sets of pipelines and designs a calibration system. The basic principle of flow measurement of the ventilator detector is introduced, the calibration method is proposed based on this principle and the device is actually built and tested, and the test results show that the static flow measurement range of this system is 5~200 SLPM, with the extended uncertainty Ur(Qv)=0.602% (k=2); the tidal volume measurement range is 0~2 000 mL, with the extended uncertainty Ur(V)=0.174% (k=2), and the relative error meets the technical specifications, and this system can realize the calibration of respiratory tester calibration of flow parameters. The design of this calibration system lays the foundation for the research and development of the calibration device of respiratory tester and provides certain reference significance for the establishment of the calibration specification of respiratory tester.

Abstract:The heat exchange tubes of the steam generator, as a key component of the pressure boundary in the primary circuit of the high temperature gas-cooled reactor nuclear power plant, plays an important role in heat exchange and radiation barrier, and its structural integrity seriously affects the safe operation of nuclear power. In response to the in-service detection difficulties of this type of special structure heat exchange tubes, a dedicated electromagnetic ultrasonic guided wave automatic detection system has been designed, a magnetic field enhanced electromagnetic ultrasonic guided wave probe with builtin single point detection has been developed, a five axis linkage multi degree of freedom automatic transport device with modular components has been developed, a dynamic positioning method for tube holes based on machine vision has been proposed, a full-scale simulation test platform for steam generator was established, and positioning accuracy test and defects detection test were carried out. The experimental results indicate that the designed automated detection system can achieve high-precision positioning and automatic walking of target tube holes at any position, and can identify the notch defects at the weld of dissimilar steel and about 60 m away from the detection end on the simulator. The effective detection range covers the entire length of the heat exchange tubes, which is expected to provide technical support for the quality and health evaluation of the special structure heat exchange tubes of the steam generator in high temperature gas-cooled reactor nuclear power plants.

Abstract:In this paper, a novel all-fiber temperature sensor based on Fabry-Perot interferometer (FPI) and Michelson interferometer (MI) is proposed and fabricated. The sensor is formed by sequentially fusion-splicing a single-mode fiber, a segment of suspended-core fiber (SCF), and another segment of single-mode fiber, with each connection having an offset. The reflecting surfaces formed by the fusion and cutting constitute the FPI and MI, and the optical path of MI is about 2 times (slightly more than 2 times) that of FPI, so the two interferometers produce the first-order harmonic vernier effect. The experimental results show that the double envelopes with obvious first-order harmonic Vernier effect appears in the interference spectrum of the sensor, and with the increase of temperature, the double envelopes of the interference spectrum gradually red-shifts, and the red-shift amount is much greater than that of a single MI. In the temperature range of 20 ℃~120 ℃, the temperature sensitivity of the sensor is 208 pm/℃, about 20.8 times that of a single MI, and about 245 times that of a single FPI. The temperature rising and falling experiments and multiple measurement experiments at the same temperature show that the sensor has good repeatability and stability. The sensor realizes the parallel connection of FPI and MI within the same optical fiber, with a total length of the sensor head being only about 584.4 μm. Additionally, it features an all-fiber tip structure, making it particularly suitable for high-temperature detection in small space environments.

Abstract:The multi-scale feature pyramid can alleviate the problems of semantic segmentation in complex traffic scenes, such as missing segmentation, wrong segmentation and unclear boundary segmentation. However, the existing multi-scale feature pyramid has to downsample the feature maps and sacrifice the spatial detail information for rich semantic information, leading to the limited accuracy of the final segmentation result. Aiming at this problem, a feature enhancement module is proposed to further reinforce similar features based on cosine similarity between different vectors before downsampling, alleviating the negative influence of downsampling. In addition, combined with the principle of dilated convolution and strip convolution, the large convolution kernel is modified to build a new multi-scale feature pyramid module for semantic information with different scales and larger receptive fields. The proposed segmentation method is real-time and efficient, and can meet the requirements of automatic driving. Experiments on the VOC2012 dataset show that the mIoU of the proposed method reaches 74.36%, and the FPS reaches 43, which is superior than the current prevailing semantic segmentation methods.