Abstract:At present, the self-potential method is widely used in the measurement of marine electric fields. Non-polarized electrodes such as Ag/AgCl or super-stable electrodes such as carbon fiber electrodes are immersed in seawater to measure the voltage between the electrodes and indirectly obtain the electric field. On the one hand, the self-potential method requires the electrodes to maintain electrical contact with seawater during measurement. The characteristics of the electrode may change gradually under the influence of seawater, which will inevitably affect the measurement stability. On the other hand, in the self-potential method, since the measurement sensitivity is proportional to the electrode spacing, the electrode spacing is usually several meters to several kilometers. The difference in environmental temperature and salinity between the two electrodes will cause voltage drift, resulting in measurement errors. In response to the above problems, this paper proposes the principle of electrical modulation for measuring marine electric fields. By controlling the electrode connection through electronic switches to modulate the electric field, an alternating electric field is generated to induce an induced current as the electric field sensing signal. When the detection electrode is insulated from seawater and does not conduct, the marine electric field signal is picked up to prevent the electrode from reacting with seawater, and the characteristics of the detection electrode can remain stable for a long time. Moreover, since the intensity of the sensing signal is negatively correlated with the electrode spacing, the electrode spacing is much smaller than that of the self-potential method, which greatly reduces the influence of the potential difference and drift caused by environmental factors on the measurement signal and the resulting measurement error. A simulated underwater electric field test platform was built to experimentally verify the proposed principle, and the direct current electric field in the simulated seawater was measured by applying this principle. The designed electrode spacing was 25 mm and the electrode dimensions were 50 mm×50 mm×1 mm. The experimental results show that the proposed method achieves an electric field measurement sensitivity of 117.87 mV/(V/m), a minimum resolution of 5.786 μV/m, and a drift range within 10 hours of less than 50.9 μV/m.

Abstract:For phase-shifted full-bridge converters used in on-board power systems of electric vehicles operating under a wide input voltage range, conventional control strategies suffer from poor dynamic performance and large overshoot. To address these issues, an improved control strategy that combines model predictive control with load current feedforward compensation is proposed. Firstly, the operational principles of the transformer lagging(Tr-lag) zero voltage switching(ZVS) phase shifted full bridge converter and the mechanism of duty cycle loss are analyzed. On this basis, a mathematical model is established to decouple the control variable from the duty-cycle loss, from which the steady-state operating model of the converter is derived. Secondly, to address the degradation in converter model accuracy and system dynamic performance caused by duty cycle loss, an enhanced mathematical model for model prediction control is derived and applied to the current inner loop. Meanwhile, a duty-cycle compensation mechanism is introduced, which effectively improves the accuracy of the converter model and the dynamic performance of the system. Furthermore, by establishing a system load estimation model and incorporating load current feedforward compensation into the voltage outer loop, the excessive output voltage overshoot caused by sudden load changes is effectively suppressed. Finally, comparative experiments are conducted to evaluate the proposed control strategy against both traditional dual-loop control and sliding mode control strategies Experimental results demonstrate that, when the proposed control strategy is applied to a Tr-lag ZVS phase-shifted full-bridge converter, both output voltage overshoot and undershoot during load transients are suppressed to within 10% of the reference voltage across the entire wide input voltage range of 300~800 V, with recovery times limited to within 5 ms. These results indicate that the proposed control strategy effectively enhances the converter′s adaptability to different battery voltage levels in electric vehicles, thereby validating its feasibility and superiority and providing innovative theoretical insights and practical engineering references for on-board power supply systems.

Abstract:High-step-up DC-DC converters are widely used in photovoltaic (PV) power generation, fuel cells, DC microgrids and hybrid electric vehicles, among other fields. This study presents a zero-input current ripple (ZICR) high-step-up DC-DC converter with soft-switching technology. It achieves ZICR by clamping the input inductor voltage at zero voltage to eliminate the input current ripple. A lower input current ripple can enhance the power generation efficiency and extent the service life of PV panels and fuel cells. High voltage gain is realized through the integration of coupled-inductor and switched-capacitor step-up technology, while the output voltage can be flexibly regulated by adjusting the switch duty cycle and the turns ratio of the coupled-inductor. Meanwhile, all the switches in this converter have achieved zero voltage switching (ZVS), and all the diodes have achieved zero current switching (ZCS). This can reduce the switching loss and the reverse recovery loss of the diodes, thereby improving overall converter efficiency. This study conducted a detailed analysis of the working principle, ZICR characteristics, voltage-current stress and soft-switching characteristics of the converter, and compared the performance parameters of the proposed ZICR converter with other similar high-step-up DC-DC converters. Finally, an experimental prototype of 100 kHz, 200 W and 38~380 V was built to verify the topological structure performance and theoretical analysis correctness of the proposed ZICR converter at the rated power. Meanwhile, the experimental efficiencies of the converters with and without ZICR at the rated power were 95.4% and 96.1% respectively. The experimental results show that the ZICR converter has excellent steady-state performance, achieves high voltage gain and high-efficiency output, and satisfies the power conversion application requirements between new energy and DC microgrids. The proposed topology therefore represents a competitive and effective solution for high-step-up DC-DC power conversion.

Abstract:This article proposes and designs an improved magnetic hyperthermia device to address the problem of large switch losses, the feedback system hysteresis and poor control effect, and also the uneven magnetic field leading to poor treatment effect in traditional tumor magnetic hyperthermia device. Firstly, it addresses the issues of high switching losses and limited driving capabilities by using totem pole drive circuits in terms of circuit design. In addition, the proposed control system achieves the power control by the fiber optic temperature sensor combined with an adaptive fuzzy control PID algorithm, which improves the response speed without interference from magnetic field. The proposed strategy also overcomes the feedback lag and large overshoot in nonlinear and time-varying systems for traditional PID control by identifying the dynamic characteristics of the system in real time and dynamically adjusting control parameters. Finally, this article also conducts further research on the problem of uneven magnetic fields based on finite element modeling method, which is improved by designing the auxiliary coils both sides of main solenoid coil and adjusting the size and number of main coil turns. The experimental results show that the proposed magnetic hyperthermia device exhibits significantly enhanced driving capability and closed-loop feedback stability compared to the prototype device. The relative deviation of magnetic field uniformity (δB) is reduced from 5.1% to 1.3% over an operating frequency range of 120~300 kHz. In addition, the steady-state temperature fluctuation within the edge area has increased by approximately 2℃ compared to the original case. Furthermore, this proposed system achieves more efficient magnetocaloric conversion in the improved solenoid coil magnetic field compared to existing devices in the literature after combining the driving circuit with the adaptive fuzzy control PID algorithm. This faster and more stable heating performance provides a more efficient hardware solution for the clinical application of magnetic therapy device.

Abstract:Tool condition monitoring (TCM) is a key technology for ensuring machining quality, improving production efficiency, and extending the service life of computer numerical control (CNC) machine tools. As a core component of the machining system, the cutting tool is subject to failure modes such as wear and chipping, which directly affect machining accuracy and system reliability. Influenced by variations in cutting parameters, fluctuations in operating conditions, and environmental noise, the tool wear process exhibits continuous, irreversible, and uncertain characteristics, resulting in evident limitations of single-sensor signals in terms of information completeness and anti-interference capability. By integrating the complementary advantages of multi-sensor signals such as cutting force, vibration, acoustic emission, current, and power, multi-source signal fusion provides an effective approach for achieving highly accurate and robust online tool condition monitoring. Focusing on the application of multi-source signal fusion in TCM, this work presents a systematic review of relevant theoretical frameworks and research progress. Common sensor types, including cutting force, vibration, and acoustic emission sensors, as well as their integration methods, are analyzed, and the performance differences among various sensors in terms of signal acquisition accuracy, anti-interference capability, and response characteristics are compared. Subsequently, data-level, feature-level, and decision-level fusion strategies are discussed, including filtering algorithms, machine learning models, and uncertainty reasoning methods. Typical applications such as tool breakage detection, wear monitoring, and remaining useful life prediction are reviewed to reveal the advantages of multi-source signal fusion in improving monitoring accuracy and system reliability. Finally, current challenges and future directions are summarized, providing theoretical and practical references for tool lifecycle monitoring.

Abstract:Key components with complex curved geometries are widely found in various industrial structures, where internal defect detection faces significant challenges due to acoustic beam distortion and unstable coupling conditions. To address this issue, a flexible ultrasonic array and an adaptive imaging method for the inspection of complex curved components are proposed. A flexible ultrasonic array capable of conformal attachment to irregular surfaces is designed and fabricated. Considering that geometric deformation induced by conformal attachment of the flexible array introduces spatial position deviations of array elements, which consequently lead to inaccuracies in propagation delay calculations in total focusing method (TFM) imaging, an adaptive TFM imaging approach based on known boundary features is developed. The proposed method does not rely on external positioning devices. Instead, geometric boundaries such as the bottom surface of the inspected component are used as internal references. Relative time delays between array elements are estimated using normalized cross-correlation, enabling reconstruction of the actual array geometry and adaptive correction of the delay law in TFM imaging. As a result, phase mismatches caused by conformal attachment of the flexible array are effectively compensated. Numerical simulations and experimental validations are conducted to comparatively analyze TFM imaging performance before and after adaptive calibration. For curved specimens with an overall inspection scale on the order of hundreds of millimeters, the minimum localization errors of internal artificial defects after adaptive calibration are 2.16 mm in the lateral direction and 1.48 mm in the axial direction, respectively, allowing stable and clear reconstruction of defect positions. Meanwhile, geometric imaging artifacts are effectively suppressed, and the overall signal-to-noise ratio is significantly improved. The results demonstrate that the proposed flexible ultrasonic array combined with the adaptive imaging method enables reliable high-accuracy defect localization under complex curved surface conditions, providing an effective approach for ultrasonic nondestructive testing of complex curved components.

Abstract:High temperature strain gauges are precision sensors for measuring the strain of hot end components in aeroengines, and the structure of their sensitive grids directly affects the measurement accuracy and fatigue life. To reduce the measurement error of high temperature strain gauges and improve their fatigue life, a structural parameter optimization method and a fatigue life reliability assessment method are proposed. Firstly, the length, spacing and number of bends of the sensitive grid are taken as optimization variables, and the measurement error and fatigue life are taken as optimization objectives to establish a multi-objective optimization model. The multi-objective grey wolf optimization algorithm (MOGWO) is used for iterative optimization and solution. Based on the optimized structural parameter combination, high temperature strain gauge samples were prepared and vibration fatigue tests were conducted at 1 000℃ to obtain effective fatigue life data, verifying the effectiveness of the optimization method. In view of the small sample size, dispersion and right skewed distribution of the test data caused by the dispersion of material properties and manufacturing errors, a set of small sample fatigue life reliability analysis methods suitable for high temperature strain gauges is proposed. To compare the fitting performance of normal distribution, lognormal distribution and three-parameter Weibull distribution, the K-S test method and regression test method are combined to determine that the three-parameter Weibull distribution has the highest fitting degree for the test data. Then, the Bootstrap method is used to estimate the confidence intervals of the parameters of the three-parameter Weibull distribution, and the quantitative relationship between reliability and fatigue life is finally established. The results show that the measurement error of the optimized sensitive grid structure is reduced by 5.3%, and the fatigue life is increased by 22.4%. The research results provide a comprehensive theoretical method and experimental basis for the design and life reliability assessment of strain gauges in high temperature environments.

Abstract:The passive radio frequency identification (RFID) tag has advantages of compact node size, battery-free and wireless operation, low cost, and less restricted by line-of-sight conditions. To address the limitations of existing RFID vibration sensing research-such as the difficulty in analyzing multi-frequency fault features and the susceptibility of raw signals to environmental noise and spectral interference, this study proposes a multi-frequency fault feature detection method using passive RFID tags for centrifugal pumps. Firstly, a multi-frequency vibration sensing model based on the RFID phase signals is formulated to clarify the mechanism of vibration sensing using phase information. Then, by utilizing the non-uniform random sampling characteristics of the reader, a compressed measurement matrix and a sparse basis are established, transforming the denoising and reconstruction of feature signals into a sparse optimization problem, which is solved using the orthogonal matching pursuit (OMP) algorithm. Finally, the quantum particle swarm optimization (QPSO) algorithm is utilized to optimize the number of iterations and the data length in the reconstruction algorithm. Two indexes-the support reconstruction ratio (quantity of the completeness of diagnostic band recovery) and the diagnostic contrast (quantity of the prominence of fault features) are proposed. The comprehensive evaluation index is constructed as the fitness function to ensure accurate recovery of the fundamental frequency and its harmonics for different fault conditions. Experimental results show that, after processing with the proposed method, the signal-to-noise ratio of the phase signal at the rotating frequency and the second harmonic under the misalignment condition increases to approximately 40 dB, and the full width at half maximum of the spectral peak is reduced to 0.41 Hz, indicating a substantial enhancement in the separation and localization of fault frequency components. For both misalignment and bolt-loosening faults, the method clearly reconstructs the second- and third-harmonic components, with frequency errors less than 1.2 Hz, and the harmonic amplitude ratios are highly consistent with those measured by the accelerometer sensor. Further pump-room field tests show that the rotational and harmonic frequency components of the pipeline circulation centrifugal pump are highly consistent with the accelerometer results, confirming good engineering applicability and providing an effective new solution for centrifugal-pump fault feature detection.

Abstract:Control valves serve as critical actuators in industrial process control systems, and their operational status directly impacts production safety and product quality. Addressing the limitations of existing control valve fault diagnosis methods—such as delayed pressure and flow signal responses, susceptibility of vibration signals to interference, and inadequate extraction of characteristic information—this study proposes a fault diagnosis method for pneumatic control valves based on microwave vibration measurement and time-frequency domain feature fusion. First, microwave vibration measurement technology is employed to achieve non-contact, high-precision acquisition of control valve stem vibration signals, overcoming the application limitations of traditional contact-based sensors. Stem vibration can more directly reflect the status of critical components such as the valve core, spring, and seals. Second, a network structure with multi-scale time-frequency domain dual-channel feature fusion is constructed. In the time-domain branch, multi-scale one-dimensional convolutions combined with bidirectional gated recurrent units are designed to fully extract the temporal dynamic features of the signal. In the frequency-domain branch, short-time Fourier transforms are used to convert one-dimensional signals into two-dimensional time-frequency spectrograms, and multi-scale two-dimensional convolutional networks are employed to extract spectral texture features. A channel attention mechanism is introduced to adaptively learn feature importance weights, and a cross-attention mechanism is employed to achieve deep fusion of time-frequency domain features, fully leveraging complementary information across different modalities. Experiments were conducted on a pneumatic control valve fault simulation test bench equipped with a microwave vibration measurement system. The experimental results show that the proposed method achieves a classification accuracy of 96.25% for six operating states on the fault simulation test bench, demonstrating superior diagnostic performance compared to common deep learning models. In the validation of 11 fault modes on the DAMADICS platform, the method achieved an average classification accuracy of 99.24%, demonstrating the model′s excellent generalization capability and providing a new technical approach for control valve fault diagnosis.

Abstract:In practical applications, control valves are often operated under multiple control modes, causing the same fault in valves to exhibit distinct characteristic information under different control conditions. This makes the machine learning based diagnostic models trained by a single-condition dataset difficult to generalize and leads to performance degradation. To address this problem, a multi-condition fault classification method is proposed in this article by combining the quantum attention-based sparrow search algorithm (QA-SSA) with an elastic adaptive random forest (EARF) model. The proposed EARF model is formulated on the adaptive random forest (ARF). By introducing a two-level decision mechanism, optimizing the placement of the global drift detector, implementing a local pruning strategy, and dynamically adjusting the number of trees, EARF reduces the computational cost of ARF, improves the modeling efficiency and diagnostic accuracy, and enhances the adaptability to variant operating conditions. To optimize the coupled hyperparameters of EARF model, QA-SSA optimization algorithm is designed by introducing quantum behavior and a Boltzmann selection strategy into the traditional SSA. The algorithm improves the searching efficiency and robustness in high-dimensional hyperparameter spaces. Finally, simulative experiments are implemented on the electric control valve fluid control system in our laboratory to evaluate the proposed method classifying six types of faults under flow, pressure, and level control conditions. The results show that the proposed QA-SSA-EARF method achieves a classification accuracy 97.47% under a single condition case, which is 9.65% and 3.64% higher than the optimized random forest and ARF models, respectively. In a multi-condition case, the average classification accuracy is 93.12% by the proposed method, which is 2.59% and 8.9% more than the other two methods. Therefore, the effectiveness and robustness of the proposed approach are verified for fault diagnosis tasks in multi-operating conditions.

Abstract:Fiber Bragg grating (FBG) sensors are widely used in the structural health monitoring due to the high sensitivity, electromagnetic interference resistance, and multiplexing capabilities. However, the FBG sensors are prone to fatigue degradation under the cyclic loading. Traditional diagnostic methods often rely on manual feature extraction or physical modeling, making it difficult to effectively capture the subtle damage signals at the early-stage. Thus this paper proposes an end-to-end spectral intelligence monitoring model named CABiL in order to solve the early diagnosis problem of FBG sensors during the fatigue degradation. The key contribution of CABiL lies in its deep integration of convolutional neural networks (CNN), multi-head attention (MHA) mechanism, and bidirectional long short-term memory networks (BiLSTM), which forms an automatic feature extraction and time-series modeling framework. The model employs a 1D-CNN to automatically extract local morphological features from spectral data by eliminating the need for manual feature selection. MHA enhances the model′s sensitivity to early spectral changes caused by subtle damage, allowing it to focus on key regions of spectrum where fatigue-induced distortions occur. BiLSTM effectively captures the temporal evolution patterns of spectral data during the loading process, integrates global dependencies and dynamic information, thus improving the modelling ability of complex damage processes. This end-to-end learning framework does not require complex physical modeling, offering the high inference efficiency. Experimental results show that CABiL achieves the state classification accuracy of over 95% for FBG, which provides the F1 scores above 0.93 for all categories. The proposed spectral intelligence monitoring method provides a highly reliable and real-time intelligent diagnostic solution for the health management of FBG sensors, which also advances the structural health monitoring towards smarter, lightweight systems.

Abstract:Using deep learning models for fault diagnosis of the transmission system helps improve the smoothness and safety of train operation. However, the operating environment of trains is complex and highly variable, and fault signals from the transmission system are easily corrupted or even obscured by noise, which leads to degraded performance of the diagnostic model. Therefore, a multi-level associative memory network is proposed for accurate fault diagnosis of train transmission systems under noisy conditions. First, a dual-domain feature extraction module is proposed to capture and fuse latent information from both the time and frequency domains, thereby enabling the extraction of multi-level features. Second, a feature fragmentation encoder is proposed to partition continuous features into partially overlapping fragments with fixed lengths and strides while embedding positional information to facilitate content addressability. Subsequently, a feature fragment association reconstructor is proposed to perform content-addressable association and prediction across fragments, complete those corrupted by noise, and reconstruct continuous features through windowing and overlap-add. In addition, a gated residual connection unit is incorporated to selectively inject the reconstructed features into the original multi-level features, enhancing detail recovery and noise robustness. Finally, extensive experiments are conducted on both a self-constructed dataset and a public dataset to demonstrate the effectiveness and superiority of the proposed method. Experimental results show that, under various noise interference, the proposed method achieves average diagnostic accuracies of 94.40% and 97.96% on the two datasets, representing improvements of at least 11.15% and 2.41% over seven comparative methods, respectively. The experimental results demonstrate that the proposed method can suppress noise and achieve superior diagnostic performance, indicating promising potential for application in fault diagnosis of train transmission systems under practical operating conditions.

Abstract:The three dimensional reconstruction of surface defects on complex-shaped components is challenging due to the varying curvature, non-uniform illumination, the absence of regular reference features as well as the low accuracy and slow speed of binocular vision matching. This study proposes a fast stereo matching algorithm named GAF-Census based on Gradient-Aware Fusion (GAF) to achieve the high-precision three dimensional reconstruction of dimensional quantification defective areas. First, an SIFT feature-guided disparity range constraint mechanism is introduced to narrow the search space and improve efficiency at the cost computation stage. Meanwhile, an adaptive Census transform based on key point median filtering is adopted, which replaces the contaminated center pixels and enhance the noise immunity with a dynamic threshold. Additionally a gradient-aware cost fusion mechanism is constructed by strengthening the gradient constraints in the edge regions to accurately locate defect contours, while the Census weight is increased in weak-texture regions to improve matching stability, thereby significantly enhancing the matching accuracy in key areas. Finally, in order to address the difficulty of defect quantification for complex-shaped components, a global fitting method based on a quintic polynomial combined with numerical integration is proposed, enabling the automated and high-precision measurement of defect dimensions. Experimental results show that the proposed GAF-Census algorithm achieves a mismatch rate as low as 5.25% for both standard and custom samples, and the computational efficiency is also improved by 96.7% compared to the conventional AD-Census algorithm. Furthermore the system can detect defects with a minimum width of 0.354 mm, and the average relative errors of defect width and length measurements are only 0.483% and 0.271%, respectively. Last but not least, the algorithm maintains the high reconstruction completeness and measurement stability under the complex lighting and geometric variation conditions, which demonstrates the strong practical applicability and provides a reliable technical solution for the automated high-precision monitoring of surface defects in the complex-shaped components.

Abstract:To address the efficiency problem in batch 3D reconstruction of fixed workpieces in industrial sites, a registration optimization method integrating binocular fringe structured light, a high-precision turntable and a scanning path planning strategy is proposed to solve the problems of low efficiency, redundant scanning time and precision degradation caused by the insufficient overlap rate of traditional scanning methods. The method eliminates the interference of initial pose differences on registration results by presetting the fixed position of the workpieces, which achieves the efficient coarse registration of multi-view point clouds with the high-precision turntable and completes the fine registration by iterative closest point (ICP) algorithm. Firstly, a 3D point cloud model of measured workpieces and tri-pyramid model of binocular fringe structured light scanner are constructed, and the system parameter calibration is completed to improve the registration accuracy and path planning. Then, the mapping relationship between the turntable rotation angle and the scanning field of view is established. The ray casting method is used to simulate the projection of real light onto the surface of target point cloud, accurately calculate the visible point cloud, and compute the effective scanning area under different angular poses. Finally, based on this mapping relationship, the minimum number of scans required to complete the full 3D reconstruction of workpieces with guaranteed overlap rate and the corresponding optimal rotation angle are solved to realize scanning path optimization. Compared with the traditional uniform rotation point cloud scanning method, this method shortens the average multi-view registration time of workpiece1 to 24.2 s, improves the efficiency by about 43%, reduces the number of scans by 3, and achieves an average error of 0.011 4 mm with a precision improvement of 64.64%. For workpiece 2, the average registration time is reduced to 58.2 s, efficiency is improved by about 40.5%, the number of scans is reduced by 7, the average error is 0.008 2 mm, and precision is improved by 81.62%, respectively. In conclusion, this method improves the scanning efficiency while ensuring the high precision, which is suitable for the batch rapid detection of fixed-position workpieces in industrial sites.

Abstract:To address problems of multiple engineering constraints, high technical iteration cost, and prolonged development cycle in the development of miniature quadrupole mass spectrometers, and meet the stringent requirements of compact size, light weight, low power consumption, long service life, and high reliability in special scenarios, a physical model and corresponding mathematical expressions for the quadrupole mass spectrometer are established based on a comprehensive analysis of product characteristics and engineering difficulties. An independent performance simulation system for quadrupole mass spectrometers is developed using the Lua language. The system achieves the coupled calculation of temperature field, fluid field, vacuum field, and electromagnetic field, and integrates 12 core functions, including ion full-life-cycle trajectory tracking and batch parameter processing, thereby enhancing engineering applicability. Furthermore, batch simulations are implemented to systematically investigate the effects of three key engineering parameters on instrument performance, i.e., the exit aperture of the EI source lens (1.5~3.5 mm), the insertion depth of the EI source into the QMA (-0.5~0.5 mm), and the sample injection rate (0~1 mL/min). The optimal value range of each parameter is determined. Simulation results show that the ion transmission efficiency reaches the maximum when the lens exit aperture is 3.0 mm; the edge field effect can be effectively suppressed with the EI source insertion depth of 0.2 mm, leading to a more concentrated spatial distribution of the ion beam. The system vacuum degree can be maintained while ensuring high signal intensity when the sample injection rate is 0.3 mL/min. An experimental platform is constructed to verify the effectiveness of the proposed simulation system. The optimized mass spectrometer has a mass range of m/z 10~180, exhibits excellent signal-to-noise ratio (SNR) in the detection of perfluorotributylamine(PFTBA), and its overall performance meets the design requirements. The developed performance simulation system provides key technical support for the precise research and development of miniature quadrupole mass spectrometers.

Abstract:This study proposes an automatic laser frequency stabilization method using dualfrequency modulation transfer spectroscopy(DF-MTS) to overcome the limitations of conventional MTS methods, such as insufficient sensitivity of the error signal and difficulty in determining the laser lock status. In this method, a dual-frequency sinusoidal phase modulation is applied to the pump beam, and two resulting error signals are demodulated and combined to construct a highly sensitive dual-frequency modulation transfer error signal. Furthermore, a second-harmonic-signal-assisted discrimination strategy is introduced for laser lock status determination. By utilizing the second-harmonic component in the absorption signal along with the amplitude information of the error signal, this approach enables accurate judgment of whether the laser frequency is aligned with the absorption peak, thereby achieving automatic absorption peak identification and rapid lock status detection. Based on the full-hardware real-time signal processing technology of field programmable gate array(FPGA), a fully automatic frequency stabilization control module is designed, featuring dual-frequency error signals demodulation, second harmonic signal demodulation, identify, control, and monitor (ICM), scanning control, proportional integral derivative control (PID), and fast relocking functions. When the loss of lock is detected, it will quickly start the relock process, enabling fully automatic frequency stabilization control. Theoretical derivation and simulation analysis are conducted, and a laser frequency stabilization experimental system is constructed to validate the proposed method. Experimental results show that the DF-MTS method improves the sensitivity of the error signal by 20.24% compared with conventional techniques. Further verification via beat-frequency measurements using a femtosecond optical frequency comb confirms that the laser is successfully locked to the absorption peak of the 85Rb D2 line (F=2 → F′=3). The standard deviation of the frequency difference between the stabilized laser and the theoretical value of the absorption line is 17 kHz, and the relative Allan deviation reaches 1.09×10-11. Auto-relock experiments show that the system can monitor the locking status of the laser in real time and complete relocking within 40.192 ms upon detecting loss of lock, indicating excellent dynamic response and robustness.

Abstract:To overcome the limitations of complex calibration procedures and low accuracy in light-plane calibration for existing sheet-of-light 3D reconstruction systems, a regional multimodal flexible mapping-based light-plane calibration method is proposed. The proposed method uses the inner corner coordinates of a checkerboard image as control points for light-plane calibration, transforming the problem of solving the light-plane equation during calibration into a mapping problem between control points in the pixel and world coordinate systems. Based on this, a regional multimodal flexible mapping model is constructed to achieve flexible mappings between pixel and world coordinates for feature points in different regions of the light plane. This approach addresses the loss of calibration accuracy caused by distortions, while accounting for non-orthogonal distortions and high-order nonlinear deformations. The method requires only a single comprehensive flexible mapping, achieving both distortion correction and high precision calibration of the optical plane. The method obviates the need for light stripe center calculation and the utilization of intrinsic and extrinsic parameter matrices, thereby eliminating the impact of image-processing errors on calibration outcomes. Experimental results demonstrate that the proposed method is simple to operate and supports a large number of feature points. The average distance residual between corresponding feature points before and after calibration is 0.01 μm. Compared with mapping models of different orders, the mapping accuracy is improved by an order of magnitude. After calibration, the measurement system exhibits a standard deviation of approximately 0.1 μm over multiple sets of 800 repeated measurements, with repeatability accuracy within 8 μm and a root mean square error of 6.5 μm. Compared with traditional invariance of cross-ratio methods, the proposed method improves measurement accuracy by 83.26%. A sheet-of-light 3D reconstruction platform was established. The planar and depth ranges of the system were characterized, and the measurement accuracy in the depth direction was experimentally validated. It fundamentally meets the requirements for stable, reliable, and high-precision sheet-of-light 3D reconstruction systems.

Abstract:To address the capability bottlenecks of multi-line structured light caused by fixed line spacing angle and narrow depth of field, this paper proposes a novel multi-line structured light projection system regulated by liquid lenses and develops a method for adjusting the characteristic parameters of light stripes. Firstly, based on the given system design configuration, a structured light projection parameter model based on dual liquid lenses is established, and a bi-level optimization method for solving structural parameters is formed by combining nonlinear programming and genetic algorithm. Then, to adjust the structured light stripes under varying projection distance, a control strategy for the liquid lens is provided, and a neural network structure is proposed to model the mapping from system parameters to line spacing angles. Finally, a physical prototype of a liquid-lens-controlled variable structured-light projection system is constructed, and a series of experiments are designed and conducted to evaluate system performance and verify the effectiveness of the proposed method. Compared with traditional schemes, this method has the ability to substantially extend the effective projection distance of multi-line structured light (keep the linewidth of light stripes within 1.25 mm). By coordinately regulating the input currents of the dual liquid lenses (without moving components during operation. The system is able to generate dynamically adjustable structured-light fringes with inter-line spacing angles ranging from 0.537° to 0.986°, achieving an angular magnification of up to 1.836. Based on the dataset obtained through calibration, a mapping network model is constructed to relate the dual liquid-lens parameters to the inter-line spacing angle. Experimental evaluation shows that the model achieves an average prediction error of 1.07% for the inter-line spacing angle. These results demonstrate that the proposed system and method can flexibly adjust the density of multi-line structured-light fringes according to factors such as the target′s size and position, while strictly controlling the stripe linewidth. The proposed approach provides a new technical pathway for enhancing the detection performance of multi-line structured-light systems.

Abstract:Pipeline leakage is an inevitable problem. However, the small leakage signals of natural gas pipelines are weak, which are inclined to be overwhelmed by strong background noises. The timely and accurate identification of leakage is a crucial and formidable challenge. Therefore, an effective leakage recognition research based on iterative self-updating multivariate variational mode decomposition combined with wavelet energy transform and dual-channel neural network was carried out in this study. Firstly, the MVMD algorithm with self-updating parameters is introduced. Specifically the number of internal modes and the penalty factor are continuously iteratively adjusted through a double-loop strategy, which realizes the adaptive and high-fidelity decomposition of multi-channel infrasonic leakage signals. This effectively avoids the modal aliasing and parameter dependence problems of conventional detection methods. Furthermore, an adaptive continuous wavelet transform enhancement strategy is proposed. Specifically the eigenmode functions are distinguished into high-energy leakage components and low-energy background components by using K-means method, where the enhancement strategy is only implemented for signals of high-energy mode and the feature integrity of low-energy signals is retained.Thereby, the information extraction ability of key features is enhanced. Finally, it is input into the designed dual-channel neural network. Specifically the high-energy channel integrates the attention mechanism and the Max pool to enhance the perception of important features. Meanwhile the low-energy channel utilizes a large receptive field convolution to extract the global background information and perform the fusion and pooling operations for information of different channels, and then utilizes the dual-path collaborative fusion to improve feature perception capabilities. Finally, the experiment was verified. The identification accuracy rate of small leaks (pore size ≤2 mm) reached 97.1%, which was 10% higher than that of the mainstream detection methods. Moreover, it maintained good performance and a short reasoning time in the cross-scenario migration experiment, proving its effectiveness and robustness in the practical engineering applications.

Abstract:To address the non-intrusive requirement of measuring arcing time of low-voltage circuit breakers, it′s crucial to overcome the interference of strong acoustic events such as mechanical collisions of opening sound signal on the identification of weak arcing acoustic events as well as the difficult identification of arcing sound signals′ start and end boundaries. Thus an arcing time measurement method based on the characteristic frequencies of acoustic-electric field signals is proposed. First, the acoustic signal segments corresponding to the arcing stage are obtained according to the division results of acoustic events during the complete opening process of the circuit breaker. Then, a kurtosis-permutation entropy index is constructed as the fitness function of bitterling fish optimization-based variational mode decomposition, which is used to adaptively decompose the acoustic signal segments. Combined with the characteristic frequency of arcing acoustic events obtained from power spectrum analysis and correlation coefficient criterion, effective modal components are selected. These components are then denoised with the singular value decomposition and reconstructed to suppress mechanical collision interference and highlight arcing components. Then a band-pass filter is designed based on the frequency characteristics of electric field signal to extract the very low-frequency components, thereby improving the distinguishing ability of arcing events′ boundaries. Taking the reconstructed acoustic signal and the very low-frequency electric field signal as inputs, a one-dimensional convolutional neural network based binary classification model is built for the arcing events. the model outputs the event probability of arcing duration, which exhibits the high precision and recall performance. To verify the effectiveness of proposed method, tests were conducted at different phase breaking current conditions. The results show that the mean absolute error, mean squared error, and root mean squared error do not exceed 0.25. Furthermore all indicators are improved by more than 76.2% compared with other measurement methods. In conclusion the proposed method possesses the high measurement accuracy and robustness, which provides the potential application value of non-intrusive online condition monitoring of low-voltage circuit breakers.

Abstract:With the consumption of energy and land resources, the demands for offshore exploration and mining, infrastructure installation, maintenance, and repair continue to increase. Automated robotic platforms equipped with manipulators have emerged as a key solution, where operational precision relies heavily on the system′s environmental perception capabilities. However, object perception in underwater environments is considerably more challenging than in terrestrial environments, and optical imaging systems will fail in turbid waters. The underwater environment presents unstructured features, and low-light and turbid underwater environments pose significant challenges to the effectiveness and accuracy of optical perception. Although existing acoustic perception methods are robust, they suffer from limited resolution and incomplete spatial information, constraining high-precision manipulator servo control. To address these limitations, this article proposes a real-time, fine-resolution 3D ultrasonic imaging method for turbid water based on a sparse array. This method introduces the total focusing method (TFM), which is traditionally used in ultrasonic non-destructive testing, into marine perception for the first time. A uniform sparse sampling strategy is first applied to the matrix array to significantly reduce data volume and computational load. An adaptive directivity correction algorithm and a sign coherence factor weighting scheme are introduced to mitigate the sidelobe artifacts caused by periodic undersampling and large element sizes. The method is implemented on a GPU-based parallel computing platform to ensure real-time performance. Experiments are conducted on critical subsea components, a wet-mateable connector and valve, demonstrating that the proposed method effectively overcomes the limitations of optical imaging in turbid conditions and compensates for the resolution and geometric information deficiencies of existing acoustic perception approaches, achieving signal-to-noise ratios of 57.03 dB and 62.54 dB, respectively. This work offers a promising solution for refined environmental perception in robotic operations under extreme ocean conditions.

Abstract:Refractive-index random fluctuations in the oceanic turbulence channel can cause wavefront distortions and intermodal crosstalk of vortex beams, thereby reducing the transmission probability of orbital angular momentum modes and undermining the stability of underwater optical communication systems. To address the limitations of the conventional GS algorithm—namely, its reliance on the far-field Fraunhofer diffraction assumption, its inability to characterize Fresnel diffraction in short-range underwater propagation, and its tendency to get trapped in local optima—an improved GS algorithm for underwater vortex-beam wavefront correction is proposed. In this algorithm, Fresnel forward/inverse diffraction is used to replace Fraunhofer diffraction; the amplitude of an ideal Laguerre-Gaussian vortex beam is introduced as an amplitude constraint; and a restricted region together with a negative-feedback mechanism is incorporated to achieve rapid and stable convergence. The method is validated theoretically and experimentally. Simulation results show that under moderate turbulence, the transmission probability of a vortex beam with topological charge L=1 increases from 0.40 to 0.98. Across variations in topological charge, propagation distance, turbulent kinetic energy dissipation rate, mean-square temperature dissipation rate, and temperature-salinity ratio, the improved GS algorithm consistently shows better robustness and higher correction accuracy than the traditional GS algorithm. Results based on simulating ocean turbulence using a spatial light modulator indicate that the improved GS algorithm converges in about 120 iterations on average, and the intensity correlation coefficient increases from 0.64 to 0.82. Compared with the conventional GS algorithm, the transmission probability is improved by approximately 20%, the convergence speed by about 25%, and the intensity correlation coefficient by roughly 3.8%. In water-tank experiments, the average intensity correlation coefficient and variance are 0.77 and 8.2×10-5 before correction; 0.79 and 3.23×10-5 after correction using the traditional GS algorithm; and further improved to 0.80 and 1.4×10-5 with the proposed algorithm. These results demonstrate that the proposed method exhibits superior robustness under a wide range of turbulence parameters, providing a useful reference for wavefront correction in underwater vortex-beam optical communication systems.

Abstract:To address the challenges of distortion and insufficient accuracy in image reconstruction of lung electrical impedance tomography (EIT), a multi-scale dense attention network (MsDA-Net) is proposed in this study to improve the reconstruction accuracy of lung ventilation and lesions based on EIT technology. As a direct estimation framework, MsDA-Net integrates dilated convolution, multi-scale dense connection, and attention mechanism to establish the end-to-end image reconstruction architecture with strong feature representation and reuse capabilities, aimed at fully exploiting deep nonlinear features from voltage measurements to improve the accuracy of lung EIT imaging. Both simulation and mapping model experiments are implemented to comprehensively evaluate the performance of MsDA-Net. Simulation results show that the lung contours and lesion structures can be effectively reconstructed by MsDA-Net. Compared with traditional imaging algorithms, the reconstructed images achieve significant improvement in visual quality and quantitative indicators. The average correlation coefficients (CCs), structure similarity index measures (SSIMs), root mean square errors (RMSEs), and peak signal-to-noise ratios (PSNRs) can reach 0.987 1, 0.903 5, 0.060 5, and 31.671 6 dB, respectively. The accuracy of MsDA-Net is similar to that of the frontier model (two-branch hyper-convolution U-Net and attention-based deep convolution neural network), which further confirms the effectiveness and progressiveness of MsDA-Net. Meanwhile, MsDA-Net shows excellent noise robustness, and the images can still maintain basic usability under 20 dB Gaussian white noise interference. Constructing the mapping models within a circular domain based on lung CT images to validate the practicality of MsDA-Net, the results indicate that the shapes and sizes of targets within the field are more accurately reconstructed by MsDA-Net. As the conductivity distribution within the field becomes more complex, the reconstruction accuracy shows a decreasing trend. However, the average CCs, SSIMs, RMSEs, and PSNRs of the reconstructed images can still reach 0.943 1, 0.857 5, 0.109 6, and 19.392 1 dB, respectively.

Abstract:Addressing the challenges of low resolution, severe occlusion and significant changes in personnel pose or shape variations, this paper proposes a new method for personnel re-identification (PR) in surveillance videos based on multimodal information fusion, using YOLOv9 as the backbone network and combining it with The Multi-Modal model CLIP (contrastive language-image pre-training). The method is divided into two stages. In the first stage, a ReID-YOLO network is constructed to enhance person feature detection performance under challenging conditions. A receptive-field enhancement module and deformable convolution are introduced to improve feature extraction for personnel with diverse poses and shapes. A spatially enhanced attention mechanism is employed to model relationships among person features and restore occluded information. In addition, a normalized Gaussian distance-based loss function is designed to increase sensitivity to low-resolution person features. These strategies jointly improve the accuracy and robustness of person feature detection in surveillance videos affected by low resolution, pose variation, shape deformation, and occlusion. In the second stage, the Multi-Modal model CLIP is introduced to improve the overall accuracy and scene generalization ability. By leveraging CLIP′s image-text alignment ability, personnel targets extracted in the first stage are predicted using discriminative features provided by ReID-YOLO. This fusion strategy mitigates CLIP′s excessive reliance on global scene information while compensating for the limited scene-awareness and target semantic parsing capability of YOLO-based networks. Experimental results under challenging conditions such as low resolution, ablation studies, and cross-identity scenarios demonstrate that the proposed method achieves outstanding performance in video-based person re-identification. It outperforms YOLO-series networks and seven other state-of-the-art video re-identification models, showing considerable promise for practical applications.

Abstract:To address the reliance of high-precision camera-line-laser joint calibration on complex highaccuracy targets in industrial 3D color reconstruction, this paper proposes a high-precision calibration scheme based on a multi-feature, weakly constrained calibration block. The overall method comprises a multimodal feature extraction and registration framework, together with a two-stage optimization-based solver for the calibration model. Circular-hole centers are introduced on the calibration block and jointly detected with corner points. Corners are localized with sub-pixel accuracy using geometric constraints, while circular-hole centers are precisely estimated via a two-stage ellipse fitting strategy. Subsequently, a pose-adaptive projection-based method for 3D feature reconstruction is presented Following a dimensionality-reduction-detection-lifting pipeline, the 3D localization problem is transformed into 2D feature detection and then back-projected to reconstruct the 3D point cloud, thereby improving robustness to noise and pose variations. Finally, unambiguous 2D-3D feature point registration is achieved by incorporating geometric priors. For parameter estimation, a linear decomposition-nonlinear reconstruction two-stage optimization is adopted. The initial mapping matrix is linearly estimated from single-frame feature correspondences; after separating intrinsic and extrinsic parameters via RQ decomposition, lens distortion is incorporated for global nonlinear refinement to enhance global optimality and generalization. Experimental results indicates that, the proposed method achieves a normalized mean reprojection error of 0.84 pixels, corresponding to a physical distance error of 0.019 4 mm. Compared with the baseline method, those two error metrics are reduced by approximately 65% and 61%, respectively. The proposed method also yields consistent calibration results with small error fluctuations under three illumination conditions, indicating strong robustness. Ablation results further confirm that center features are significantly more stable than corner features under perspective transformation. In the gear tooth-surface color reconstruction task, point-cloud color texture mapping based on the obtained mapping matrix faithfully reproduces microscopic impressions and scratches on the tooth surface, thereby validating its engineering applicability.

Abstract:To address the challenge of inertia tensor measurement for irregular rigid bodies, this article proposes a novel integrated method combining binocular vision and torsional pendulum techniques. First, high-resolution industrial cameras synchronized with an atomic clock are employed to capture sequential images of the torsional motion during a single measurement cycle. Feature points are extracted from the images, and a high-precision angular displacement-time curve is derived based on the geometric relationships within the measurement system. From this curve, the torsional vibration period and damping ratio are extracted. A linear damped torsional vibration model under linear damping conditions is subsequently applied to calculate the moment of inertia for a single measurement. Furthermore, binocular structured light 3D reconstruction technology is utilized to obtain the point cloud data of the measured object and the torsional pendulum. A point cloud registration algorithm is used to accurately align the real-measured point cloud of the object with the point cloud of the object′s computer-aided design (CAD) model, solving for the homogeneous transformation matrix. The axis direction is determined through cylindrical axis fitting, and the homogeneous transformation matrix is used to transform the data into the coordinate system of the CAD model′s product center of mass, effectively avoiding the mechanical positioning errors inherent in traditional measurement methods. The cosine values of the angles between the centroidal coordinate system axes and the torsional pendulum axis are computed. Combined with the measured moments of inertia, these values formulate an inertia ellipsoid equation. Ultimately, a system of equations encompassing all parameters of the inertia tensor is formulated through six rotational configurations and solved to achieve high-precision measurement of both the moments of inertia and inertia products. Extensive experiments are conducted on the proposed method and system. The experimental results evaluate the feasibility and effectiveness of the proposed method. The absolute error in the measurement of the moment of inertia is less than 0.5×10-5 kg·m2, and the maximum deviation in the principal axis orientation angle is 0.99°. The measurement proposed scheme proposed in this paper achieves high accuracy, no longer relying on mechanical positioning, significantly improving both measurement efficiency and safety. It is suitable for the measurement of inertial parameters of various products.

Abstract:To address the high cost, time-consuming nature, and specialized expertise required by marker-based optical systems, this article proposes a visual kinematic analysis method utilizing two viewpoints to achieve convenient, low-cost kinematic evaluation. First, a two-dimensional feature extraction architecture is established by integrating the global context modelling capability of the Swin Transformer, the precise positional awareness of coordinate attention, and the multi-scale feature fusion capability of the bidirectional feature pyramid network. It overcomes challenges such as occlusion and small target detection for keypoints, enabling effective extraction of two-dimensional features. Secondly, a triangulation method is proposed, employing joint contextual constraints based on keypoints position plausibility and limb length consistency. This is combined with a parametric human model to reconstruct 3D keypoints, enhancing estimation accuracy. Finally, a keypoint augmentation model is formulated to obtain an anatomical label set, which is then integrated with a musculoskeletal model for kinematic analysis. Kinematic evaluation on public datasets demonstrates an average joint angular error of 8.59° and average joint positional error of 42.02 mm, outperforming existing high-performance methods. To validate real-world applicability, commercial motion capture system Xsens serves as the evaluation benchmark against the mainstream OpenCap method, with analyses conducted on shoulder joint and gait kinematics, respectively. Experimental results show that for shoulder joint and gait kinematics, the proposed method achieves correlation coefficients of 0.92 and 0.86, respectively, with Xsens, representing improvements of 9.52% and 7.40% over OpenCap. Angular errors are reduced to 13.97° and 3.12°, respectively, marking decreases of 27.01% and 25.18% compared to OpenCap. In summary, the proposed method achieves more accurate kinematic analysis than current mainstream approaches on both public datasets and in real-world scenarios, holding significant implications for advancing applications related to kinematic analysis.

Abstract:UAV can efficiently perceive the fire environment and obtain the fire scene information. To improve the intelligence level of firefighting work, a fire cannon pre-aiming system based on UAV visual information is proposed. Pre-aiming is a control process consisting of three stages: Fire scene perception, pose calculation, and angle adjustment. In the stage of fire scene perception, considering the real-time requirements of firefighting work, a perception model combining lightweight object detection and dehazing processing is proposed to address the problems of small target size and smoke interference in UAV images of fire scenes. Regarding image dehazing, considering the characteristics of non-uniform distribution and diverse gray levels of haze in fire scenes, the atmospheric scattering model is improved. A neural network with an encoder-decoder structure is designed to solve transmission map and haze gray value, which significantly enhances the image quality. Regarding object detection, YOLOv8s is used as the baseline. In the backbone network, the convolution operations in shallow layers are replaced by PSConv module with a concentrated receptive field to extract more information of small targets; the convolution operations in deep layers are replaced by GhostConv, and the SimA-former module is employed to substitute the deepest C2f structure to achieve model lightweighting. During the feature fusion stage of the neck network, the coordinate attention mechanism (CA) and the small target detection head are combined to construct a high-resolution multi-scale feature fusion module. Based on the acquired fire scene information, the camera model is utilized to compute the relative position and orientation of the fire cannon and fire source. Subsequently, the required horizontal and pitch angles for the fire cannon adjustement are determined. Experiments were conducted in a custom-built fire scenario outside an industrial facility. The perception model achieved a mAP50 score of 92.3%, representing a 6.2% improvement over the YOLOv8s without dehazing preprocessing. The pre-alignment error in the horizontal angle was within ±4°, while the distance estimation error for the fire source remained below 6%. Those results demonstrate the effectiveness and practical applicability of the proposed method.

Abstract:In varying-illumination and repetitive-texture environments, existing visual-inertial navigation systems (VINS) suffer from insufficient feature extraction and high feature mismatch rates, failing to meet application requirements in pose estimation accuracy and system robustness. To address these challenges, an improved PL-VINS is presented to enhance feature extraction in varying-illumination scenes and feature matching in repetitive-texture environments. In the image preprocessing module, a closed-loop gamma correction method iteratively adjusting image brightness until the desired level is proposed to increase the number of extractable features, thereby enhancing system robustness under varying illumination conditions. In the line feature detection and tracking module, the intersection points of spatially parallel line pairs are first calculated in the image plane and clustered to obtain intersection-point clusters and their weighted centers. Then the line features are clustered based on their distance and direction relative to these weighted centers to enhance the robustness of line feature matching in repetitive-texture environments. In the backend optimization module, the intersection points of intra-cluster line features are incorporated into optimization as additional features. Reprojection residuals that jointly fusing point, line, and intersection features are constructed to improve pose estimation accuracy in repetitive texture scenarios. Comparative experiments on public datasets demonstrate that the improved PL-VINS reduces the average absolute pose error by 17.4% on the EuRoC dataset compared to PL-VINS and by 12.2% on the UMA-VI dataset compared to SuperVINS. To further verify the effectiveness of the proposed method, an experimental platform using a mobile robot was constructed for real-world testing. The results indicate that the improved PL-VINS exhibits superior accuracy and robustness compared to state-of-the-art algorithms in environments with illumination changes and repetitive textures.

Abstract:To address challenges in valve pose estimation based on point cloud registration—such as complex backgrounds, occluded or missing features, and noise interference—the article proposes a lightweight graph-spatial attention-hierarchical feature fusion network (LGSA-HFFNet) algorithm. This approach designs and employs a multi-scale parallel convolutional feature extraction layer to enhance feature extraction, prevent gradient explosion during training, and accelerate convergence. It further incorporates a lightweight graph-spatial attention (LGSA) module, a streamlined enhancement combining graph and spatial attention, to overcome neural network difficulties in extracting features from disordered point cloud information, enabling effective extraction of local point cloud features. Finally, the system is validated through pose estimation experiments and deployed in real-world valve pose estimation tasks. Experimental results demonstrate that the LGSA-HFFNet algorithm achieves an average relative translation error of 0.05 m and a rotation error of 0.984 degrees in valve point cloud registration experiments. It exhibits excellent robustness, with translation and rotation registration performance degrading by only 2% and 7.5%, respectively, under complex backgrounds. Registration time is reduced by 80.32% compared to the iterative closest point (ICP) algorithm, with performance significantly outperforming traditional methods like ICP and semidefinite-based randomized sample consensus (SDRSAC). In ModelNet40 benchmark tests, rotational and translational errors were reduced to 2.293 degrees and 0.006 m, respectively, with rotational accuracy reaching an advanced level and translational accuracy showing a significant advantage over existing models. In experiments using real-world valve pose estimation datasets with high noise interference, the proposed method achieves errors of 2.175 7 degrees and 0.036 m, representing reductions of at least 28.98% and 17.81% compared to existing models.

Abstract:The operation status of coal mills in coal-fired power plants is influenced by factors such as coal quality variations, load fluctuations, and equipment aging, which can lead to dynamic shifts. Existing monitoring methods typically rely on the “offline modeling, online deployment” approach, which is difficult to realize the adaptive, continuous and precise condition monitoring. It′s khown that knowledge distillation methods use lightweight student models to inherit the superior performance of complex teacher models, facilitating the rapid model updates and online deployment. Therefore, we propose an adaptive condition monitoring method for coal mills based on the continuous distillation. This method enables the quick adaptation to dynamic changes of coal mill during the operation process by continuously guiding the student model through the teacher model and updating the student model online. Considering the characteristics of coal mill data, we ropose the graph temporal convolutional network as the teacher model by combing the feature extraction advantages of graph convolutional networks and temporal convolutional networks. The student model is constructed based on a composite loss function, inheriting the knowledge from the teacher model via distillation loss and ensuring the monitoring accuracy with supervised loss. The new parameter fusion strategy is designed to periodically update the student model′s parameters based on real-time data, achieving the iterative optimization of the parameters. Validation with operational data of power plant shows that the proposed method outperforms comparison methods in both monitoring accuracy and adaptability. At the normal operating conditions, the standard deviation of prediction residuals under the continuous distillation approach is reduced by an average of 8.45% compared to the offline modeling method, significantly enhancing the stability of model. While in the abnormal operating scenarios, the proposed method successfully captures the fault symptoms and issues early warning signals 116 hours in advance, while maintaining a zero false alarm rate. In conclusion, the proposed method can improve the intelligence level of equipment operation and maintenance, demonstrating the broad prospects for engineering applications.

Abstract:Motion interpolation methods functioning as a key technology in motion control directly impact the trajectory accuracy and operational efficiency of motional objects. The current circular interpolation method of variable synchronization ratio does not consider the establishment of synchronous transition process, leading to the large interpolation errors in the circular interpolation and low sensitivity of interpolation accuracy to the speed. Thus this study proposes a circular interpolation method with the variable synchronization ratio based on the optimization of synchronization establishment timing. Based on the acceleration and deceleration control strategy, the establishment of synchronous transition process time is calculated in each interpolation cycle, which is required for each synchronous slave axis to reach the new synchronization state after acceleration or deceleration. Then, the angle increment of each synchronous slave axis during the establishment of synchronous transition process time is obtained, thus the precise position of the synchronous master axis corresponding to the changed synchronization ratios of the slave axes is calculated, which facilitates the determination of optimal timing for such changes. By changing the synchronization ratio of the master axis in advance to optimize the synchronization establishment timing, the slave axes reach the new synchronization state with the master axis when it moves to the circular arc, thereby effectively reducing the trajectory deviation caused by synchronization lag. The circular interpolation is achieved by continuously changing the synchronization ratio. The experimental results of variable synchronization ratio circular interpolation on a two-dimensional motion platform show that the proposed method achieves a significant improvement in interpolation accuracy compared to the reported variable synchronization ratio circular interpolation method. When completing the full-circle interpolation with radii of 20 mm and 100 mm, the average offset distances of the interpolation trajectory center at a speed of 10 mm/s and an interpolation step length of 1 mm are reduced by approximately 84.7% and 85.6%, respectively, and the average mean squared error (MSE) values are decreased by about 58.7% and 68.1%, respectively. Moreover, the change of average MSE value is relatively small when the interpolation speed changes. This further enhances the circular interpolation accuracy and effectively reduces the sensitivity of interpolation accuracy to speed.

Abstract:To reduce the dependence of the finite control set model predictive control performance of grid-connected inverters on system model parameters, a model-free predictive control method for grid-connected inverters based on an adaptive sliding mode observer is proposed. Firstly, based on the ultra-local model theory, an adaptive sliding mode observer is designed to accurately observe the total disturbance of the system, effectively avoiding the dependence of traditional methods on precise model parameters and enhancing the robustness and anti-interference ability of the system. To optimize the system architecture, an extended Kalman filter is introduced to replace the grid-side voltage sensor, providing key parameters for model-free prediction by real-time estimation of the grid-side voltage state, while reducing the complexity of system design. To address the performance degradation caused by control delay in digital controllers, an improved delay compensation method based on the first-order linear extrapolation is proposed. By using historical current data to predict the system state after delay, the compensation current is used for the next cycle prediction, improving the real-time performance and accuracy of current tracking. Finally, a simulation model and an experimental prototype are established for comparison and analysis with traditional methods. Experimental results show that, compared with the traditional model predictive control strategy, the proposed method reduces the total harmonic distortion rate of the grid-connected current by 36.67% when the inductance suddenly increases and by the sudden 47.84% decrease. When the reference current undergoes a sudden change, the system dynamic response speed is increased by 21.78%, and the total harmonic distortion rate of the current in steady-state operation is as low as 2.37%, meeting the grid connection standards. The proposed strategy effectively reduces the negative impact of model parameter dependence and control delay through multi-module collaborative optimization, providing a reliable control solution for the efficient and stable operation of grid-connected inverters.