Abstract:To address the issue of processing deformation at the bottom of grooves during discontinuous slot milling of large thin-walled components—caused by their weak rigidity and uneven wall thickness, which impacts processing accuracy—a comprehensive compensation method for processing deformation errors is proposed. Initially, a milling force measurement experiment is conducted using empirical methods. Through regression analysis, a mapping relationship between processing parameters and milling forces is established, and a milling force prediction model is created. To overcome the low simulation calculation efficiency for large thin-walled components, the equivalent stiffness theory is applied to simplify the deformation region. An improved substructure simulation method is then introduced by replacing the main structure to simulate multi-layer milling processes. Combined with the milling force prediction model, this method predicts processing deformation with a 27.27% improvement in calculation efficiency compared to the full-structure finite element method. Next, using an in-machine measurement system to collect wall thickness data at the groove bottom, a deformation correction model is developed. The predicted processing deformation is corrected by applying inter-layer and node correction coefficients. The inter-layer correction coefficients are iteratively calculated using the secant method, and discrete compensation points are fitted to the machining path using non-uniform rational B-splines. Finally, an in-machine measurement system tailored for the robotic milling of large thin-walled components is designed and implemented. Comparative experiments on milling processing error compensation are conducted. Experimental results demonstrate that, after applying the comprehensive compensation method, processing error is reduced by 92.09% and 77.63% compared to no compensation and the mirror iterative compensation method, respectively. These findings validate the effectiveness of the proposed comprehensive compensation approach for processing deformation errors.

Abstract:Nanophotonic imaging technology is one of the primary methods in modern optical imaging and measurement systems. However, traditional broadband achromatic metalenses face issues such as increased structural complexity, reduced imaging efficiency, and lower focusing resolution with the addition of working wavelengths, which limit the development of high-performance, integrated miniature optical systems. Graphene oxide is a two-dimensional material with high refractive index and high transmittance. By using laser direct writing technology, specific areas of graphene oxide are thermally reduced to reduced graphene oxide, altering the optical properties of the material and enabling the fabrication of ultrathin planar lenses. Addressing the discrete wavelength dispersion issue of graphene oxide lenses, this paper proposes a reverse design method based on an improved genetic algorithm. By setting optimization goals for the genetic algorithm and incorporating a penalty factor into the objective function, the lens structure can be targeted for optimization, designing a graphene oxide lens capable of focusing discrete wavelengths into a single focal point with equal intensity. The designed graphene oxide lens (approximately 200 nm thick) was fabricated using vacuum filtration and femtosecond laser direct writing techniques, and its focusing characteristics were characterized. Experimental results show that this lens can excellently control dispersion for incident wavelengths of 450, 550, and 650 nm, with a maximum deviation from the preset focal length of only 2.23%. The average radial full-width at half-maximum (FWHM) of the lens′s focal points for working wavelengths is 324.3 nm, achieving sub-diffraction-limited focusing. This design method offers new possibilities for miniature optical systems that require high resolution and broad bandwidth dispersion control.

Abstract:To address the issue of deviation in the maximum efficiency point in wireless power transfer systems, caused by overlooking losses in inverters and rectifiers, a composite control strategy is proposed. This strategy combines maximum efficiency tracking with constant voltage output to minimize the optimal load deviation. Initially, the relationship between the parameters of the LCC-S type wireless power transfer system and the output voltage and current of each component is analyzed. A nonlinear rectifier bridge load equivalent model is developed to understand the impact of rectifier diode voltage drops on load conversion.Next, a system efficiency quantification model that incorporates losses from both the inverter and rectifier is created to quantitatively analyze the system efficiency as the load varies, thus reducing the deviation error in the optimal load. Additionally, LCC-S compensation parameters are optimized to enable zero-voltage switching operation for the inverter. On the secondary side, an impedance matching technique based on a Buck-Boost circuit is implemented for maximum efficiency tracking, while a Buck circuit is employed on the primary side to ensure stable output voltage control.Finally, an experimental platform is built to validate the theoretical analysis. Compared to traditional methods, the proposed strategy reduces the optimal load deviation error by 24.6%, increases the maximum efficiency by 1.1%, enables zero-voltage turn-on operation of the inverter across a wide load range, and maintains an overall system efficiency of approximately 90% with constant voltage output.

Abstract:Angle measurement, as a pivotal technology in the field of precision metrology, holds significant application value in modern industries, such as aerospace, equipment manufacturing, and high-precision inspection. To address the challenge of high-precision small-angle measurement in the precision manufacturing and assembly of large-diameter components such as large gears, this article proposes an extended-field-of-view high-precision rotation angle measurement method based on binocular vision collaboration. Firstly, an improved two-dimensional four-parameter coordinate transformation algorithm is employed to formulate a calibration model for the measurement reference plate, achieving high-precision calibration and providing a reliable reference basis for angle measurement. Secondly, to overcome the limited field of view of a single camera when measuring large-scale workpieces, a binocular vision measurement system with complementary spatial fields of view is established. A feature point matching-based algorithm for solving the field-of-view center coordinates is proposed, enabling the determination of the camera′s field-of-view center coordinates in the measurement reference plate coordinate system. Thereby, a theoretical foundation for angle measurement is achieved. Finally, based on the aforementioned methods, a binocular vision-based rotation angle measurement system is developed. A large gear with a modulus of 20 mm and 30 teeth is used as the test object. Systematic accuracy tests and repeatability verification are conducted within a rotation range of 0° to 20°. The results show that the system achieves an absolute measurement accuracy of 6″ and a repeatability accuracy of 5″, demonstrating high measurement precision and stability. The proposed method and the designed system meet the requirements for high-precision dynamic angle measurement of large-diameter components, providing a new technical approach for such applications.

Abstract:Laser ranging has become the mainstream technology in the field of high-precision spatial perception due to its excellent directionality, long-distance transmission capability, and anti-interference performance. Especially in recent years, the growth of smart city construction and military reconnaissance demands have driven the technology of high-precision ranging and positioning for UAV-mounted systems to become a research hotspot. Addressing the issues of wide-ranging error sources and large positioning errors in airborne laser rangefinders, this paper proposes a multi-dimensional joint calibration and guidancepositioning method. First, the error sources, generation mechanisms, and spatiotemporal correlations of airborne rangefinder calibration and positioning are systematically studied and analyzed, providing a theoretical basis for algorithm research. Second, a calibration method for airborne laser rangefinders is designed, which integrates mechanical alignment, pod spatial calibration, and sensor time synchronization, reducing the ranging error of unmanned aerial vehicles in low altitude flight scenarios. Subsequently, a guidance-positioning method is proposed, which achieves high-precision target positioning through active detection, laser guidance, and geometric calculation. Finally, the calibration and positioning methods are experimentally validated in scenarios of variable speed and turning motion scenarios. The experimental results show that when measuring targets within a 10 000 m range, the proposed method yields a distance measurement error of less than 2 m. Compared with the single drone positioning method, the average distance error of the guided positioning algorithm in variable speed motion is 2.22 m, and the positioning accuracy is improved by 80%, meeting the high-precision calibration and positioning requirements of drone detection.

Abstract:Panoramic illuminance refers to the average luminous flux per unit area received from all directions in three-dimensional space at a given measurement point. It plays a significant role in lighting quality assessment, particularly in evaluating the overall illumination received by objects and the general lighting level perceived by individuals. However, there has been a lack of simple and effective practical measurement tools. This paper proposes a method for measuring panoramic illuminance based on a regular tetrahedral illuminance meter. This method involves measuring the illuminance on the normal directions of the four faces of a regular tetrahedron, allowing for straightforward calculation of the panoramic illuminance value. This study begins by reviewing existing methods for panoramic illuminance calculation and measurement, followed by a theoretical demonstration of the proposed approach. Subsequently, over 120 000 measurement simulations were conducted using software based on 2 233 indoor and 205 outdoor high dynamic range panoramic images from the Laval database. Results show that approximately 96% of the measurements had an absolute error within 5%, with an average absolute error of 1.7%, validating the method′s accuracy. Moreover, the measurements remained stable under different orientations of the tetrahedral illuminance meter. Finally, we proposed a simple construction plan for the tetrahedron illuminance meter and verified it in four real lighting environments. Experimental results indicate that tools based on this measurement theory are easy to construct, with a straightforward measurement process and calculation, enabling a relatively accurate measurement of the panoramic illuminance values in lighting spaces. The method has opened up a new approach for the measurement of panoramic illuminance in practice.

Abstract:The mode localization effect is a proven approach for enhancing the sensitivity and resolution of MEMS physical quantity sensors. However, the lack of in-depth research on electric field introduction methods has limited the resolution of currently reported mode-localized electric field microsensors, preventing them from meeting the requirements for weak electric field detection. This study proposes a high-resolution electric field microsensor leveraging mode localization. The sensor consists of an electric field introduction structure, a three-degree-of-freedom weakly coupled resonator, and a packaging base. A floating and discrete highresolution electric field introduction design is adopted, supported by a theoretical model for optimizing structural parameters. The arithmetic square root of the amplitude ratio of the weakly coupled resonator is used as the output of the sensor to address nonlinear output challenges. Finite element simulations are utilized to analyze the vibration modes of the sensor and investigate the impact of the rotation of the electric field induction electrode on measurement accuracy. Experimental testing of the fabricated sensor under vacuum conditions demonstrates a sensitivity of 0.068 /(kV·m-1) and a background noise level of 0.012 1 (V·m-1)/Hz within the measured electric field range of 0~7 kV/m. The sensor achieves a resolution better than 0.4 V/m, representing the highest level reported domestically and internationally to date. Additionally, the sensitivity of the electric field induction electrode is tested by rotating it within the measured electric field; An in-depth study of the mode localization phenomenon in the resonator revealed that this phenomenon can be precisely controlled by selectively exciting and perturbing specific resonators to induce localization in targeted modes.

Abstract:To achieve the rapid airspeed recognition requirements of loitering munition fuzes, a high-efficiency and accurate airspeed measurement method based on pressure sensors and the Bernoulli equation is proposed. A compact and low-cost pressure measurement device is designed accordingly. Numerical simulations of the aerodynamic characteristics of a fixed-wing loitering munition model are conducted to determine the optimal layout of pressure sensors based on surface pressure distribution and flow field characteristics. To reduce flow field interference and improve measurement stability, a metal square tube is installed around the sensor to form a pressure measurement channel, with pressure compensation applied. Simulation studies on different pressure channel length combinations identify 15 and 20 mm as the optimal configuration. Further simulations under varying speed and angle of attack conditions show that the proposed method maintains a velocity measurement error within 9.43% to 14.07% in the range of 0.1~0.3 Ma and -15° to +15° angle of attack. The method significantly reduces measurement error under low-speed and small-angle conditions. It exhibits strong error suppression capabilities under high-speed, and large-angle conditions, with an average error reduction of 3.21% compared to the configuration without a pressure channel. To further validate the accuracy and reliability of the method, both wind tunnel tests and field flight tests are conducted. The wind tunnel tests cover a wide range of speeds and angles of attack, showing that this method reduces airspeed measurement error by 3.6% compared with the traditional Pitot tube. In the field tests conducted under actual flight conditions of the loitering munition, the measured airspeed is compared with data from the Pitot tube and the flight control system. Results show that the proposed method achieves a mean squared error of 1.429 m/s, which is 2.06 m/s lower than that of the Pitot tube. These results show that the proposed method has strong environmental adaptability and measurement stability, providing an efficient and reliable solution for airspeed measurement in loitering munition fuzes.

Abstract:In low-voltage distribution substations, users′ neutral and ground wire wiring errors often cause leakage faults, which can easily lead to electric shock casualties. Although multivariate regression analysis has been used to locate users with wiring errors and leakage faults, it is limited by the insufficient sampling frequency of the current monitoring equipment in the substation area, and has the inherent defect of poor positioning timeliness. Therefore, a fast localization method based on super-resolution reconstruction of metering data is proposed. By reconstructing low-resolution data, this approach overcomes the time-resolution limitations of traditional methods. First, the composition of the residual current in the substation area during wiring errors and leakage faults is analyzed, and the correlation characteristics between the residual current and the user load currents are clarified. Then, the generalization performance of traditional multivariate linear regression, Lasso regression, ridge regression and elastic network regression models is systematically evaluated, revealing the impact of independent variable collinearity on the stability of parameter estimation. The time series current data is further mapped into a two-dimensional feature image, and the enhanced super-resolution generative adversarial network (ESRGAN) model is used for super-resolution reconstruction. The data reconstruction quality is verified by root mean square error, peak signal-to-noise ratio and structural similarity index. Finally, the reconstructed high-resolution data was used to establish an elastic network regression model to locate users with wiring errors and leakage. The comparative analysis based on the laboratory simulation platform and the field measured data showed that the proposed method has higher data reconstruction quality, higher model fitting degree and higher accuracy in locating users with wiring errors and leakage. Moreover, the fault localization time is reduced by several multiples compared to traditional methods.

Abstract:To address the challenges in the quantitative damage evaluation and the uncertainty in damage evolution of composite laminate structures, a fatigue damage evolution tracking method for composite laminates based on multi-feature interactive fusion is proposed. This paper proposes a method for tracking the damage state of composite plates by constructing a damage index observation equation based on the interactive fusion of multi-domain features, combined with a strain energy release model and a particle filter algorithm. By extracting multi-domain features of Lamb wave signals, such as time-frequency domain features, dynamic time warping features, and transfer entropy features, the fatigue damage state of composite plates is comprehensively characterized. These features are used as observations of the damage state, and a damage state space model for the composite plate is established. To better capture the linear correlation between multi-domain features and the degree of damage in the composite plate, a multivariate interactive prediction model is innovatively introduced to fuse the multi-domain damage features interactively. This establishes a mapping relationship between Lamb wave signal features and the damage evaluation index of the composite plate, forming the damage index observation equation. Building on the strain energy release rate model of the composite plate and considering the uncertainty factors in damage evolution, the particle filter algorithm is employed to track the damage state, such as crack density and delamination size. The effectiveness and accuracy of the proposed method in tracking and predicting the damage state of composite plates are validated through finite element simulations and fatigue test data analysis of composite plates made from T700G unidirectional carbon fiber prepreg. This research not only reveals the evolutionary of the damage index but also provides a new technical approach for real-time monitoring and evolution prediction of damage in composite plates.

Abstract:This study investigates pipeline defect imaging using ultrasonic guided waves based on a sparse-array total focusing method (TFM) integrated with a phase coherence factor (PCF). Conventional TFM suffers from significant limitations, including excessive fullmatrix data volume, prolonged imaging processing time, and relatively low signal-to-noise ratio (SNR), which hinder the effectiveness of pipeline nondestructive testing. To address these issues, this research proposes an optimized TFM method by employing a sparse-array configuration, reducing the number of transducer elements from 32 to 8 while maintaining sufficient acoustic coverage, thereby decreasing the data volume by a factor of 16. Furthermore, to ensure imaging quality with the reduced array, a phase coherence factor (PCF) is introduced to suppress noise interference by leveraging the phase consistency of echo signals, significantly improving computational efficiency without compromising imaging resolution. Compared with conventional 32-element TFM, experimental results show that the proposed method achieves an approximately 120% improvement in SNR and a 38% enhancement in imaging efficiency, effectively reducing processing time while improving detection performance. For scenarios involving multiple coexisting defects, this study develops a multi-defect superposition imaging technique based on the TFM-PCF data matrix. By utilizing the phase coherence characteristics of defect signals, this method enables phase-synchronized enhancement of multiple defects, improving both SNR and detection rates. Specifically, the proposed method increases the SNR for double-hole defects from 32.97 dB (conventional TFM-PCF) to 42.69 dB, representing a 30% improvement. The reliability and effectiveness of this method are experimentally validated for various pipeline defect conditions, demonstrating its potential for practical industrial applications in high-precision pipeline inspection.

Abstract:Aluminum electrolysis cells operate in complex environments influenced by the coupling effects of strong magnetic, electric, and thermal fields, making them prone to grounding faults that significantly affect both production efficiency and safety. Existing detection methods have limitations in precise fault localization and real-time diagnosis. To address these issues, a grounding fault diagnosis method based on excitation coupling is proposed. An alternating voltage signal with an amplitude of 30 V and a frequency of 100 Hz is injected at the front end of the aluminum electrolysis series. The dynamic response of the alternating voltage across the cell and its ground voltage is analyzed to examine the fault characteristics of the grounding system. It is found that, under normal operating conditions, the circuit exhibits capacitive characteristics, whereas, after a grounding fault occurs, the circuit′s characteristics transition from capacitive to resistive. A series of simulation experiments for both single-point and multi-point grounding faults are designed, introducing the capacitive-to-resistive ratio as a fault diagnosis indicator. By comparing the electrical characteristics of the system under normal and fault conditions, the following diagnostic criteria are proposed. When the alternating voltage across the cell exceeds a threshold, the phase difference between the alternating voltage across the cell and the ground voltage is less than 10°, and the capacitive-to-resistive ratio exceeds 104, a grounding fault can be confirmed. Experimental prototypes are established, and field evaluate have validated the effectiveness of the method. The proposed method is suitable for online monitoring and real-time diagnosis, offering significant improvements in both safety and economic efficiency in aluminum electrolysis cell operations. The results show that the method provides an efficient and reliable solution for grounding fault diagnosis in aluminum electrolysis cells.

Abstract:Aiming at the existing bearing defect detection algorithms with low accuracy, misdetection as well as serious leakage, a bearing defect detection algorithm based on YOLOv8n (LASW-YOLOv8) is proposed to solve these problems. Based on YOLOv8n, the algorithm introduces a lightweight and efficient LiteShiftHead detection head, which is combined with SPConv, REG and CLS modules to improve the efficiency and accuracy of feature extraction, target frame regression and category classification. In addition, the algorithm also introduces the Adaptive Rotation Convolutional Kernel module (ARConv), which enhances the detection of multi-directional defects; the Neck Network Optimisation module (SAF), which further improves the efficiency of feature extraction; and the Inner-WIoU loss function, which is used to optimise the bounding box localisation accuracy and to enhance the detection of small targets and complex shape defects. Experimental results show that the LASW-YOLOv8 algorithm outperforms other mainstream algorithms in several performance indicators. The algorithm achieves an accuracy of 97.2% and a recall of 96.6%, representing improvements of 3.4% and 4.5%, respectively, compared to the original YOLOv8n. Meanwhile, mAP0.5 and mAP0.5:0.95 achieved 98.9% and 73.3%, respectively, and ran at 83 fps. These results fully demonstrate the effectiveness of the proposed algorithm, which not only effectively reduces the phenomenon of false detection and missed detection, but also meets the requirements of high accuracy and real-time performance in industrial inspection. In addition, in the experiments on the Northeastern University public dataset (NEU-DET), the LASW-YOLOv8 algorithm outperforms other mainstream algorithms in the four key metrics of accuracy, recall, mAP0.5, and mAP0.5:0.95, which are 79.3%, 79.9%, 84.1%, and 49.1%, respectively. This performance proves that the LASW-YOLOv8 algorithm has excellent generalisation ability and robustness.

Abstract:The modular multilevel converter (MMC) is a key power conversion component in flexible DC transmission and distribution systems. However, its cascaded submodule topology, which incorporates a large number of switching devices, presents reliability challenges and contributes to a higher failure rate. Traditional open-circuit fault diagnosis methods for MMC switching devices often rely on additional sensors and are susceptible to interference due to threshold sensitivity. To address these limitations, this paper introduces a novel open-circuit fault diagnosis and localization approach based on tensor feature extraction and a two-dimensional broad learning system (2D-BLS), enabling fast and highly accurate fault identification.The proposed method constructs a third-order tensor from submodule capacitor voltage data to efficiently handle multi-channel MMC signals. Through Tucker decomposition, the method separates fault-type classification from fault-location identification while extracting meaningful tensor features. Each subtask′s tensor features are then processed using dedicated sub-classifiers built on the 2D-BLS framework. The 2D-BLS employs a bilinear transformation to maintain structural information while significantly reducing the number of parameters. The outputs of all sub-classifiers are subsequently fused to accomplish fault diagnosis and localization.This approach eliminates the need for additional sensors and empirical thresholds, reduces the model′s class complexity, and enhances both diagnostic accuracy and computational efficiency. It is particularly well-suited for handling multiple open-circuit faults in switching devices. Simulation and experimental results confirm that the proposed method achieves a diagnosis and localization time of less than 15 ms with an accuracy exceeding 98.5%, demonstrating its effectiveness and superiority.

Abstract:In dynamic settings such as human-machine hybrid intelligent warehouses, Automated Guided Vehicles (AGVs) often face challenges in accurately detecting randomly appearing obstacles like pedestrians and forklifts, which can jeopardize both operational efficiency and safety. This study introduces a lightweight target detection method based on the fusion of LiDAR and image data, termed L-BEVFusion. Firstly, a lightweight feature extraction network is designed to derive 2D image information for constructing bird′s eye view (BEV) features. To reduce localization errors caused by relying on single-scale semantic information, multi-scale semantic features are incorporated. Secondly, an explicit supervision strategy utilizing depth ground truth is applied to project image features into 3D space. Predictive features from both image and point cloud data are then extracted. A BEV feature fusion network concatenates these image and point cloud BEV features along the channel dimension, enabling bounding box regression and classification for dynamic obstacle detection in human-machine collaborative warehouses. The proposed algorithm is evaluated on both the KITTI dataset and real warehouse-collected data. Experimental results show that, compared with common point cloud-image fusion methods, L-BEVFusion improves detection accuracy for workers and forklifts by 3.46% and 2.22%, respectively, on the warehouse dataset, with an overall average accuracy increase of 2.97%. It also demonstrates superior inference speed and detection size accuracy, achieving an average normal distance error of 4.02 mm and a tangential absolute error of 1.75 mm. These improvements enhance the real-time detection performance and reliability of AGVs, ensuring efficient and safe logistics operations in intelligent warehouses and highlighting strong practical value.

Abstract:Austenitic stainless steel small-diameter tubes, with their thin walls, generate multiple echoes due to ultrasonic wave reflections and mode transitions at the interfaces, leading to significant artifacts in phased array imaging that hinder defect detection. To address the impact of ultrasonic mode type, number, and their amplitude spatial variability on multimodal compound imaging performance, this paper proposes a multimodal ultrasonic total focus compound imaging method with spatial compensation for defect response. A defect spatial response model is developed, and compound imaging is performed on the full matrix data acquired using this model, which shows good agreement with results from CIVA simulation software. The defect response values, determined by maximum intensity near the defect in each imaging mode, are compared, with deviations within ±3 dB relative to the simulation. Additionally, the spatial distribution of imaging sensitivity in different modes is analyzed, and spatial compensation parameters suitable for each mode are identified. Compound imaging is carried out using both calibrated direct modes and half-skip modes. Compared to single-mode imaging and direct-sum composite imaging, the proposed method, based on defect spatial response compensation, achieves higher defect amplitude and reduced artifact amplitude, with no artifacts in imaging results of circular-hole defects at various positions. The method is further applied to crack and porosity defect experiments, where it effectively suppresses artifacts and enhances the signal-to-noise ratio compared to single-mode imaging.

Abstract:In recent years, functional nearinfrared spectroscopy (fNIRS) and electroencephalography (EEG) have been extensively employed to measure and analyze neurocognitive responses and mental states during natural and social interactions. While fNIRS-EEG bimodal fusion has shown promise, challenges remain, including limited channel counts, low sensitivity, and poor data synchronization. To address these issues and enable more effective investigation of neurophysiological dynamics in everyday scenarios, we developed a lightweight fNIRS-EEG imaging system capable of synchronized acquisition, real-time data transmission, and visualization. The system integrates 80-channel fNIRS and 32-channel EEG coverage across the whole brain. System performance was first validated through a series of benchmark tests. For fNIRS, the signal fluctuation under fully parallel measurement remained below 1%, with linearity exceeding 0.99, successfully detecting 20% signal variation at a 10 mm depth in a two-layer brain-mimicking phantom. For EEG, the system achieved a signal-to-noise ratio of 52 dB (with a 1 μV reference noise level) at an input of 100 μV, and a common-mode rejection ratio of up to 112 dB. In integrated mode, the system supports simultaneous sampling at 20 Hz (fNIRS) and 500 Hz (EEG), with timestamping based on the terminal clock to enable synchronous hyperscanning. To further demonstrate in-vivo capability, a canonical steady-state visual stimulation experiment was conducted, confirming the system′s ability to monitor neural activation. In summary, this integrated system provides a novel, real-time platform for simultaneous monitoring of brain electrical activity and hemodynamic responses, facilitating research on perception and cognition in real-world contexts.

Abstract:In dynamic environments, the presence of moving objects such as pedestrians and vehicles within the sensor′s field of view causes significant interference in the creation of static point cloud maps. These dynamic objects often result in “ghost” artifacts, severely affecting the accuracy and completeness of the generated maps. To tackle this challenge, this paper proposes a high-precision algorithm for static point cloud map generation specifically designed for complex dynamic scenarios. The goal is to effectively eliminate dynamic interference points while preserving map construction accuracy. The method begins by serializing multi-frame point cloud data and applying region-based ground segmentation to minimize the impact of ground points on dynamic point identification. A multi-stage offline dynamic point cloud filtering strategy is then implemented. In the first two stages of dynamic point identification, two distributed descriptors (D-POD and D-PODV) are used to capture the spatial occupancy and distribution patterns of the point cloud. These descriptors are combined with the scan ratio test (SRT) and radial ratio test (RRT) to precisely identify both strongly and weakly dynamic points. In the third stage, an improved adaptive DBSCAN clustering algorithm, based on regional density, is employed to further refine the removal of irregular dynamic point clouds. Experimental results using the publicly available SemanticKITTI dataset show that the proposed algorithm effectively filters out dynamic points, resulting in high-precision static point cloud maps across diverse complex scenarios. Compared to state-of-the-art dynamic point cloud filtering algorithms, ERASOR and Removert, the proposed method achieves average improvements of 3.95% and 14.56% in static point retention, and 13.44% and 17.46% in dynamic point rejection, respectively. By employing staged filtering for both strongly and weakly dynamic point clouds, the proposed method successfully eliminates various types of dynamic objects while maximizing the preservation of static information in the original point cloud. This ensures the structural integrity of the global map, providing robust support for the creation of high-precision and highly reliable static point cloud maps across a wide range of applications.

Abstract:To fulfill the need for high-precision positioning capabilities of indoor service robots, this article proposes an indoor lightweight mapping method for laser point clouds based on direct optimization. This method fully considers the structural characteristics of indoor environments and the unique advantages of lidar. First, redundant information, such as ground points, is filtered out through point cloud filtering. Then, long point cloud sequences are segmented into multiple fragments. Inside each fragment, the NDT algorithm is used to provide an initial estimation for registration. Subsequently, nonlinear optimization of poses is conducted based on pixel brightness information to construct accurate local maps. Finally, combined with OPENGL′s multi-layer technology, a complete indoor map is assembled. To evaluate the feasibility and performance of the proposed algorithm, a dedicated point cloud processing software is developed and tested in the internal areas of an experimental building. The results show that, under lightweight and low-configuration conditions, the map constructed by this algorithm maintains a high degree of consistency with the currently well-known algorithm LIOSAM. Meanwhile, the relative error of mapping is controlled within 1%, and the average computation time between frames is 95.8 milliseconds. While demonstrating high precision, it also maintains excellent real-time performance, thus exhibiting potential and value for practical applications.

Abstract:Laser 3D measurement is an important means of modern high-precision measurement technology for large and complex surfaces, which is widely used in vehicles, ships, aircraft, and other fields, especially in the digital imaging of large scenes. To address the problem that the accuracy and range of visual tracking 3D measurement cannot be optimized at the same time, this article proposes a 3D measurement method based on multi-camera visual tracking. The method adopts the multi-camera joint calibration algorithm to formulate the global co-baseline position conversion model to realize the position tracking of the laser sensor, establishing the optimal field-of-view decision model with the highest accuracy of stereo target positioning as the optimization goal. The spatial geometric relative position relationship between the target and the laser sensor is established according to the structural characteristics between the stereo target and the line laser sensor. The position-matching algorithm of geometric distance is adopted to obtain the position of the line laser sensor and combines the multi-camera field of view to complete the laser point cloud data splicing. Finally, the three-dimensional point cloud imaging is realized. In order to evaluate the effectiveness of the method, a four-eye laser 3D measurement system is built within the range of 1.8 m×2.5 m×1.5 m. The experimental results show that the measurement method achieves a translational positioning uncertainty of 0.054 3 mm and a rotational positioning uncertainty of 0.047 8° for the stereoscopic target; the lateral measurement range of this method is 0.93, 1.18 and 1.46 m at the distance of 0.5, 1 and 2 m from the binocular system, respectively. Under the condition of using a camera of the same focal length in order to achieve the same accuracy, it is respectively larger than the monocular measurement range of 0.57, 0.7 and 0.82 m; 0.35, 0.46 and 0.49 m larger than the binocular measurement range. It shows that the stereo target position measurement method based on binocular vision has good engineering advantages in terms of accuracy and measurement range.

Abstract:Meniscus-confined electrodeposition (MCED) technology has gained considerable attention in micro/nanoscale structure fabrication due to its low cost and high precision. However, the limited displacement range of piezoelectric stages restricts its applicability in large-scale manufacturing scenarios. To overcome this limitation, a cross-scale continuous deposition method based on visual inspection is proposed. This approach integrates a microscope camera and a threeaxis motion system into existing MCED equipment, creating a cross-scale continuous deposition platform that combines coarse probe adjustments with fine platform positioning, guided by microscopic vision measurements. Next, by using predefined equidistant displacements of the micro/nano platform as a reference, pixel shifts corresponding to these displacements are calculated through image grayscale analysis. This data is then used to develop an Adam-optimized gradient descent model, which establishes an “object-to-image” distance mapping relationship, enabling high-precision positioning and measurement of the deposited material and the probe from a single camera image. Visual feedback control is employed to achieve precise alignment and deposition positioning, facilitating cross-scale continuous deposition on the MCED platform. Finally, a confocal microscope is used to measure and analyze the length and junction quality of the deposits. The results demonstrate that this method enables the deposition of millimeter-scale line segments using an 80 μm-range piezoelectric stage, with deposition length errors below 3% , while maintaining high quality and precision at the segment endpoints. This approach offers a promising solution for large-scale precision manufacturing.

Abstract:Surface defect detection of steel strips is crucial for ensuring product quality in steel manufacturing. The achievement of efficient and accurate detection is significant for product performance. While deep learning methods have made significant progress in defect detection, two challenges persist in practical applications: First, due to the pursuit of high yield rates in industrial production, defect samples are limited and sample annotation is time-consuming. Secondly, the collected samples may contain redundant features, affecting model training efficiency and generalization performance. To address feature redundancy, a dynamic sample selection method based on feature representation and learning feedback is proposed. Initially, a multi-dimensional feature quantification model incorporating geometric morphology, grayscale distribution, and directional features is formulated to characterize defect features. Subsequently, a feature representationbased sample selection strategy is designed to select diverse and representative training samples through feature clustering. Finally, a confidence-based dynamic optimization strategy is proposed to obtain supplementary samples through learning feedback, achieving adaptive optimization. Experimental results on the NEU-DET dataset show that the method achieves a mean average precision of 76.99% while reducing the training sample size by 52%. It is comparable with using the complete dataset and decreases training iteration time by 62%. Evaluation of various detection models shows that the method effectively improves efficiency while maintaining performance across different architectures.

Abstract:The steered response power with phase transform (SRP-PHAT) algorithm is widely used for sound source localization due to its strong robustness in reverberant environments. However, the traditional SRP-PHAT algorithm has insufficient localization accuracy and large computational volume in multi-microphone array source localization scenarios, which cannot meet the demand for high-precision real-time source localization. Aiming at the above problems, a SRP-PHAT algorithm based on quadratic cross-correlation is proposed. In this approach, the autocorrelation and cross-correlation between the signals of two groups of channels in the array are subjected to quadratic cross-correlation. The generalized cross-correlation phase transform function (GCC-PHAT) is used as the basis for further calculation to obtain the improved SRP-PHAT function, followed by peak detection to achieve sound source localization with improved accuracy. To reduce computational load, the microphone array is divided into reference channels and sound source channels, and correlation operations are performed only between these groups. This avoids the redundancy of traditional pairwise computations across all channels, significantly lowering the computational cost. The traditional SRP-PHAT algorithm and the SRP-PHAT algorithm based on quadratic inter-correlation are transplanted in the self-developed 128-array multi-spiral-arm array source localization system, and the experiments of source localization are conducted indoors at four source frequencies (10~25 kHz). The experimental results show that the improved algorithm reduces the azimuth estimation error by an average of 2.5°, the pitch estimation error by an average of 2°, and the spatial resolution of the localization by an average of 45.78% for the localization of sound sources at four different source frequencies. Compared with the conventional algorithm, the improved algorithm improves the localization accuracy and reduces the computation amount significantly, which provides an effective solution for the SRP-PHAT algorithm for real-time sound source localization in multi-microphone arrays.

Abstract:To address the noise interference during the demodulation of fiber Bragg grating (FBG) by the CCD-based spectral diffraction method, which leads to the reduction of the demodulation accuracy and stability of FBG center wavelength, a fusion noise reduction algorithm based on the variational mode decomposition (VMD-WT-SG) is proposed. By combining the advantages of different noise reduction algorithms and adjusting algorithm parameters, the effective removal of noise components in the spectral signals is realized. The spectral waveforms are smoother and more continuous while retaining most of the features of the original signals, and the noise reduction effectiveness is obvious. Compared with the traditional three methods, such as the VMD noise reduction, the SG noise reduction, and the Kalman filter, the spectral SNR of the spectral signal-to-noise ratio processed by the VMD fusion noise reduction method (VMD-WT-SG) is 20.14 dB, and the root-mean-square error is 0.017. The signal-to-noise ratio is improved by 23.94%, 41.14%, and 94.97%, respectively. The root mean square error is reduced by 39.29%, 45.16%, and 67.92%, respectively. It is the best among the four noise reduction methods. To improve the demodulation rate, an intersection peak-finding algorithm is proposed for the noise-reduced multi-peak spectra based on the commonly used peak-finding algorithm. The center wavelength is obtained by calculating the coordinates of the intersection of the peaks with the intersection of the left and right sag lines of the next largest points. The peak-finding accuracy is compared with the polynomial fitting, the center-of-mass, and the Gaussian fitting algorithms through experiments. The results show that the average peak-finding deviation of the proposed algorithm is 3.3 pm, which is better than that of the center-of-mass and the polynomial fitting. The average running time of the algorithm is 0.261 ms, which is better than those of the polynomial fitting and the Gaussian fitting. In real-time applications, the demodulation rate can reach 4 kHz while ensuring accuracy, Meanwhile, the algorithm has good stability, the average standard deviation of the demodulated wavelengths is 1.8 pm, and it can meet the requirements of practical applications, which has some reference values for the fast multi-peak real-time detection in the FBG sensing networks.

Abstract:As China′s oil and gas exploration and development continues to move towards unconventional oil and gas reservoirs, it is vital to develop core equipment and key technologies for deep resource exploration. Acoustic logging while drilling tool is a key equipment for deep earth exploration, while its core technology is blocked by foreign countries. The quality of acoustic signals received by domestically developed instruments has declined due to the influence of various noise sources such as drilling operations, tool eccentricity, mud circulation, and circuit transmission noise. This article introduces an overall design of an acoustic logging while drilling instrument, including a transmitting circuit, transducer, and array receiving module. A convolutional neural network algorithm based on encoder-decoder architecture is adopted to reduce the noise of acoustic data in the time-frequency domain, which enhances the signal-to-noise ratio of the instrument′s received signals. The time-frequency features of the noise-containing signal after a short-time Fourier transform are used as the inputs to the network. The neural network with a U-shape architecture learns the sparse representations of the signal and noise in the data and generates the time-frequency masks at the same time. In this way, the separation of the signal and noise is realized. To address the lack of open data sets, theoretical modeling is implemented for the instrument addressed in this article, and many theoretical simulations are carried out to obtain the signal database for different model parameters, and high-quality noise data are collected to establish the noise database. After training, the neural network is able to intelligently reduce the complex multi-source downhole noise, and this algorithm can achieve a good noise reduction effect on the test data and the acoustic signals collected from the instrument site, which greatly improves the quality of the received signals of the instrument under the interference of multi-source noise, such as low-frequency noise, circuit overshooting, complex oscillations, and mutant noise.

Abstract:Prediction models based on deep learning often suffer from “catastrophic forgetting” caused by the inability to adapt to new degraded data. This is currently a hot and difficult topic in the field of artificial intelligence research, and also one of the constraints on the development of intelligent equipment. The accuracy evolution process of CNC machine tools has chaotic characteristics. Therefore, a chaotic self-evolution prediction method for the motion accuracy of CNC machine tools based on chaos theory and incremental learning is proposed. Firstly, it has been proven that the dynamic variation of motion errors in CNC machine tools is a chaotic evolution process of a complex nonlinear dynamic system. It is proposed to reconstruct the evolution trajectory of the precision system in the chaotic phase space through phase space reconstruction. Then, a motion accuracy chaotic evolution model based on deep long short-term memory (LSTM) network is established, utilizing the LSTM network's excellent ability to accurately capture long-term dependencies of time series, to track the inherent laws of the evolution trajectory of CNC machine tool motion accuracy in chaotic phase space. Finally, a learning without forgetting incremental learning method is proposed in the chaotic evolution model to establish a motion accuracy chaotic self-evolution prediction model. This model uses joint optimization and knowledge distillation methods to update parameters, allowing the model to adapt to new degraded data while also conveying the soft objectives of old tasks. It solves the problem of “catastrophic forgetting” during dataset updates, improving the accuracy and robustness of long-term predictions. The experiment shows that the evaluation indicators MSE,MAE and MAPE predicted by the method proposed have decreased in fluctuation amplitude by 70.56%, 33.31%, and 35.77%, respectively, compared to the non-LwF module. This proves the accuracy of the proposed prediction method. Further module ablation experiments also verify that the method proposed is superior to traditional methods in terms of prediction accuracy and robustness.

Abstract:To address the challenges of incomplete geometric feature information of two-dimensional visual images of roadside facilities and traffic participants, as well as the lack of scene semantic information inaccurate perception and understanding of complex traffic scenarios for autonomous driving, a semantic understanding model for complex autonomous driving scenarios based on element information completion is proposed. Firstly, a dense connection network (DenseNet) is utilized to extract multi-scale 2D features from visual images. Then, the feature line-of-sight projection (FLoSP) module is used to inverse-map voxels to 3D space. A dimension decomposition residual (DDR) module is utilized to construct a 3D UNet, extracting 3D features of scene objects and enabling the transformation of singleframe 2D visual image features into 3D features. Additionally, a contextual residual prior (3D CRP) layer is introduced between the 3D UNet encoder and decoder. Atrous spatial pyramid pooling (ASPP) and Softmax layers are used to output scene semantic completion results, thereby enhancing the spatial semantic understanding capability of the model. Meanwhile, image caption generation technology is utilized to formulate a context-aware semantic embedding scene understanding language description model based on an improved VGG-16 encoder and a long short-term memory (LSTM) decoder. The improved VGG-16 encoder integrates and concatenates features of traffic scenes at different scales and inputs them into the LSTM decoder via a projection matrix, establishing a semantic representation between scene object images and predicate relations, and automatically generating natural language descriptions of object detection results and autonomous driving decision-making suggestions. Finally, the proposed complex scene semantic understanding algorithm is validated on the Semantic KITTI dataset and through real vehicle experiments. Compared with the JS3CNet algorithm, the results show that the proposed algorithm achieves a relative improvement of 11.27% in mean intersection over union (mIoU), realizes accurate perception and semantic understanding of complex scenarios in autonomous driving through semantic completion, and provides a reliable basis for autonomous driving decision-making and planning.

Abstract:The piston pressure gauge is a critical instrument for pressure traceability and measurement, with the piston system being its core component. Accurate positioning and rapid stabilization of the piston present significant challenges due to the properties of gas media. To address these challenges, a novel ultra-local mode-free adaptive control (ULMFAC) method is proposed, which analyzes the nonlinear characteristics of the piston system in gas piston pressure gauges. This method combines super-twisting nonsingular terminal sliding mode control (STNTSMC) with a finite-time disturbance observer, effectively preventing the chattering phenomenon typical of sliding mode control and greatly enhancing the system’s dynamic response. To account for parameter variations such as the effective piston area, temperature, and medium leakage, an improved second-order dynamics model is developed using an ultra-local model-free approach, eliminating the need for a precise system model, as required in model-based control methods. A nonsingular terminal sliding mode surface is designed to overcome singularity issues in terminal sliding mode control, while an adaptive supertwisting algorithm is applied to mitigate chattering and enhance system dynamics. A disturbance observer is used to estimate lumped uncertainties, ensuring finite-time stability. The stability and convergence of the proposed control scheme are confirmed through Lyapunov analysis. Simulations and experimental results demonstrate that the ULMFAC method significantly improves the robustness, piston positioning accuracy, and dynamic response speed under varying working conditions (0.5, 3 and 6 MPa). This method holds considerable theoretical importance and practical value for achieving high-precision and high-efficiency pressure measurements.

Abstract:To tackle the challenge of efficient flight path planning for unmanned aerial vehicle (UAV), an enhanced hybrid genetic-particle swarm algorithm (IHGPA) is proposed. This algorithm, based on particle swarm optimization (PSO), integrates multiple strategies to enhance both convergence performance and solution quality. Firstly, to improve global optimization, a partition optimization strategy is introduced into the IHGPA, and a dynamic parameter adjustment mechanism is employed to optimize the particle velocity and position update methods. Secondly, the genetic algorithm’s selection, crossover, and mutation operators are refined to further boost optimization capabilities. During selection, a combination of the roulette wheel method and simulated annealing algorithm is used to preserve elite individuals. In the crossover phase, probabilistic arithmetic crossover and an improved simulation binary crossover are integrated to increase population diversity. For mutation, Lévy flight long-step perturbation and polynomial mutation are fused to prevent premature convergence. Finally, by drviding the search area to exchange optimal solution in formation and implementing a convergence detection mechanism is implemented, where particles undergo secondary optimization if their fitness value falls below a predefined threshold, preventing the algorithm from getting trapped in local optima. Experimental results show that, in environment 1 with scattered obstacles, the best fitness value of the IHGPA undperforms genetic algorithm, particle swarm optimization, wolf pack algorithm, artificial bee colony algorithm, and dung beetle optimizer by 78.130%, 46.190%, 53.990%, 41.124%, and 67.376%, respectively. In environment 2, with dense obstacles, IHGPA′s best fitness value is reduced by 89.990%, 75.088%, 76.503%, 71.048%, and 81.061%, respectively. The IHGPA effectively generates safe, smooth, and optimal flight paths while demonstrating outstanding stability and reliability across multiple verification trials.

Abstract:To address the degradation of endeffector accuracy in supernumerary robotic arms (SRAs) caused by human motion disturbances during humanrobot collaboration, we propose a compensation control method based on modeling and prediction of these disturbances. First, we design a sensory scheme using T265 visualinertial odometry, which integrates an inertial measurement unit (IMU) with visual estimation to accurately measure humaninduced disturbances. Next, we develop a kinematic model of the humanSRA system, expressing the endeffector pose as a function of both human motion disturbances and SRA joint movements. The control objective is to maintain a stable endeffector pose, and for this, we develop a disturbance compensation strategy using feedforward proportionalintegralderivative (PID) control. To further enhance the compensation control response speed, we propose a predictive approach utilizing a Kalman filter to estimate human motion disturbances. The Kalman filter algorithm is used to accurately predict human motion trajectories by formulating a statespace equation for human motion. Finally, we conduct experiments on both human motion disturbance prediction and disturbance compensation control. Experimental results show that the absolute error between predicted and actual disturbances is 048±032 mm. A comparison of compensation performance with and without prediction shows that the proposed method reduces the absolute error of the SRA′s endeffector on the working plane from 318±217 mm to 123±091 mm. These findings confirm that the proposed compensation control strategy effectively improves endeffector accuracy, with the Kalman filterbased prediction method significantly reducing control delay.

Abstract:The efficient operation of a variable pitch control system is an important basis to ensure the stable power output of wind turbines, optimize the operating conditions, and reduce the mechanical load fatigue. Before the operation of the wind turbine, it is necessary to complete the refined design and tuning of the pitch control system parameters offline. In engineering, these parameters mainly rely on engineers to perform manual tuning through experience knowledge and simulation software. This method has high personnel training and optimization time costs and faces low accuracy and poor consistency. However, traditional proportional-integral-derivative optimization algorithms are limited in the process of intelligent tuning of variable pitch control parameters, and it is easy to fall into local optimality. Therefore, based on the idea of proportional integral and derivative control, this article sets convergent and random controller parameter, and further introduces new control targets, control errors, and Levy flight strategies. An improved proportional-integral-derivative optimization algorithm is proposed. IPIDOA and PIDOA, Harris Hawks optimization, whale optimization algorithm, gray wolf optimizer, particle swarm optimization, and genetic algorithm are tested and verified on 4 single-peak reference functions, 4 multi-peak reference functions, and optimization examples of wind turbine pitch control parameters. The results show that IPIDOA has faster convergence speed, better parameter optimization ability, and stronger optimization stability in multi-class optimization cases. Concurrently, by calculating the time complexity of the IPIDOA and comparing the convergence curves of the algorithms in the parameter optimization research of the wind turbine pitch control system, it shows that the IPIDOA algorithm has excellent computational efficiency.

Abstract:Multi-piezoelectric drive is an effective scheme to break through the displacement travel limit of nano-piezoelectric drive mechanism. However, there are some problems such as inherent hysteresis nonlinearity, piezoelectric drives intercoupling, nonlinear and linear coupling, overdrive redundancy and so on. Inspired by those challenges, this paper presents an electromechanical coupling dynamics modeling and overdrive control strategy of a multi-piezoelectric parallel driving mechanism. Firstly, an Hammerstein electromechanical coupling dynamics model is built to separate and describe the linear and nonlinear characteristics of the multipiezoelectric drive mechanism. Meanwhile, it is proposed that the parameter estimation method for the linear and nonlinear parts of the model. Secondly, it proposes a comprehensive overdrive control strategy, which integrates feedback linearization, control distribution algorithm, and upper level control rate. Especially, the proposed least square control distribution algorithm can minimize the 2-norm of error sequence. Finally, parameter estimation experiments and over-actuated control experiments were carried out for the proposed modeling and control methods. The results of the parameter estimation experiments showed that the proposed model′s output curve closely matched the experimental output curve of the multi-piezoelectric driving mechanism, effectively describing its hysteretic nonlinear input-output characteristics. The results of the over-actuated control experiments demonstrated that the trajectory tracking performance of the proposed least squares control allocation algorithm was superior to both direct allocation algorithm and optimal allocation algorithm. Specifically, when the desired trajectory was a sinusoidal signal with an amplitude of 130 μm and a frequency of 10 Hz, the accuracy of the proposed least squares control allocation algorithm was 56.63% higher than that of the direct allocation algorithm and 47.83% higher than that of the optimal allocation algorithm.