Abstract:Prognostics and health management (PHM) enables proactive maintenance and risk mitigation by monitoring, analyzing, and forecasting equipment health, undergirding safe and stable system operation. While physics-based and data-driven PHM paradigms are relatively mature, classical approaches struggle to integrate expert knowledge and generalize when confronted with massive heterogeneous data, especially unstructured text and multimodal information generated by increasingly complex industrial systems. Recently, the rapid development of Transformer architecture and large language model (LLM) have opened a high-precision prediction paradigm that efficiently exploits domain expertise, offering advantages such as knowledge extraction and fusion, few-shot generalization, and intelligent decision support. This review comprehensively surveys the current status and future prospects of PHM empowered by LLM. First, canonical PHM tasks, together with Transformer architecture and general LLM, are introduced. Second, domain-specific LLM construction is elaborated with respect to domain-knowledge creation and injection, covering internal parameter optimization and external knowledge augmentation; a unified framework for PHM domain-specific LLM is presented. Third, current PHM domain-specific LLM-based frameworks and applications are dissected in-depth from a task-oriented perspective across components, subsystems, and complex systems levels, focusing on fault diagnosis, state of health estimation, remaining useful life prediction, and anomaly detection. Finally, future challenges and opportunities are outlined regarding model compression, edge deployment, and generalized complex systems.

Abstract:Industrial defect detection is a critical component of modern industrial production and operation, ensuring product quality, production efficiency, and safety. The complex logical reasoning and generalization capabilities of large models have positioned them as the critical force behind the new wave of artificial intelligence. With the emergence of large models, a new paradigm is established for industrial defect detection, bringing both fresh opportunities and challenges. This article provides a comprehensive review of the current application status of large models in the field of industrial defect detection. Firstly, the development process of large models is systematically combed, and the core technologies are introduced in detail, including model architecture, multimodal data processing, pre-training techniques, fine-tuning methods, alignment strategies and efficient reasoning mechanisms. Secondly, a survey of traditional methods based on machine learning and deep learning for industrial defect detection is provided, followed by a comparison with large model-based approaches and a summary of their respective strengths and limitations. Then, focusing on the industrial defect detection domain, the review introduces commonly used open-source datasets that support large model research and evaluation, as well as the performance evaluation methods of large models. Furthermore, it categorizes the current main application of large models into five directions, including defect detection and localization, defect detection in complex scenarios and micro-defect detection, few-shot and zero-shot adaptive detection, interactive defect analysis and decision support, and defect data generation with automatic annotation. Finally, this article thoroughly analyzes the challenges confronting large models in industrial defect detection, such as data quality and security, high-reliability requirements, cost constraints and sustainable development, and the lack of unified evaluation standards, while providing an outlook on their future trends. The review aims to provide valuable references and insights for the continued advancement and broader implementation of large models in industrial defect detection.

Abstract:As the functionality of oscilloscopes continues to expand, their operational complexity has correspondingly increased, posing a significant learning barrier for novice users. Even those with basic operational knowledge often struggle to fully utilize the advanced features. To reduce the operational complexity of oscilloscopes, this study proposes an intelligent control system for oscilloscopes based on the large language model. Firstly, the system employs a domain adaptation technique by constructing a structured knowledge graph for oscilloscope control to generate domain-optimized prompts, thereby enhancing the large language model′s ability to comprehend user instructions. Secondly, the system incorporates semantic retrieval techniques, utilizing vector space modeling and approximate nearest neighbor search to filter the most relevant knowledge fragments from the knowledge graph based on user instructions. This approach compresses the prompt size and improves inference efficiency. Finally, by integrating these two techniques, the system establishes a closed-loop control mechanism of “an natural language instruction-standard commands for programmable instruments-operational feedback”, enabling precise control of the full range of oscilloscope functions through natural language. Experimental results demonstrate that on a self-constructed dataset, compared to directly using a large language model to generate standard commands for programmable instruments, the generation accuracy of the qwen-max-latest model improved from 6.20% to 99.6% after applying the domain adaptation technique. Furthermore, compared to using only domain adaptation, the incorporation of semantic retrieval technique, when running the qwen2.5-32b-instruct model on a single NVIDA RTX 4090 GPU, reduced average inference latency from 296 s to 23.3 s, while maintaining a loss inference accuracy of less than 7%. In summary, the intelligent oscilloscope control system proposed in this study effectively lowers the barrier to using oscilloscopes, provides technical support for the intelligent and automated operation of laboratory instruments, and demonstrates promising application prospects.

Abstract:With the rapid expansion of carrier network scales, the complexity and risks of network operations continue to escalate, making traditional operation models inadequate for meeting the demands of efficient and low-risk network changes. While digital twin technology has emerged as a promising solution through virtual-physical network mapping, extant implementations still face fundamental limitations: the fidelity-scale tradeoff in simulations and latency in dynamic responsiveness. This paper introduces an instrumentation-enhanced large model (IELM) framework, integrating data-driven modeling, instrumented simulation, and large-model verification. The proposed approach leverages network instrumentation to ensure scalable high-fidelity emulation and observability of digital twin simulations. Meanwhile, large language models (LLMs) power a closed-loop simulation-measurement-optimization cycle—enabling autonomous configuration generation and real-time policy refinement. Validated in China Mobile′s network digital twin system, IELM achieved large-scale network twin pre-verification. It reduced validation cycles for configuration changes and service deployments from weeks to hours, improving network operation efficiency by approximately 40%. This research establishes a new paradigm for intelligent network assurance in hyper-scale carrier environments.

Abstract:The learning-based motion planning approach uses a data-driven policy trained on large-scale driving experiences and have demonstrated good performance. However, these methods often treat motion planning as a black-box problem, resulting in limited interpretability. They also face challenges such as dataset bias, overfitting, and getting stuck in a local optimum. In this paper, we exploit the powerful inference and interpretation capabilities of emerging large language models to propose a large language model-based motion planning framework for autonomous driving, called LLMs-Driver, to address the problem of poor interpretability in learning-based approaches. LLMs-Driver consists of three parts, namely, the reasoning module, the memory module, and the reflection module. In the reasoning module, we propose the important experience playback algorithm, which integrates the two influencing factors of experience priority and scene similarity, to improve the learning efficiency and performance of the LLMs-Driver. In the memory module, we propose an improved first-in-first-out experience storage algorithm to ensure the validity and novelty of the experience, ensuring that LLMs-Driver continuously learns from the most recent and effective strategies. Meanwhile, in order to fully enhance the transparency and credibility of the self-driving motion planning model, the ‘three-step chain of thought’ method is adopted, which divides the inference and reflection process into three steps, each accompanied by explanatory textual reasoning. Finally, we validate LLMs-Driver through closed-loop autonomous driving experiments on the Highway-env simulation platform. The experimental results show that LLMs-Driver has significant interpretability and motion planning capabilities, with the median number of successful steps on a task increased up to 2.19 times of the baseline algorithm. Additionally, it supports the customization of different driving styles based on the driver′s intention.

Abstract:The high coupling of components and the concealed nature of cascading faults in spacecraft electromechanical devices impose stringent demands on the reasoning efficiency and interpretability of fault diagnosis systems. To address the challenges of high construction costs associated with traditional knowledge graphs (KG), the lack of domain-specific expertise in general-purpose large language models (LLM), and the insufficient associative reasoning capability of retrieval-augmented generation (RAG) in textual knowledge-driven intelligent fault diagnosis, this study proposes an ontology-constrained knowledge graph-RAG fault diagnosis method. Firstly, a four-layer fault diagnosis ontology framework is constructed. Utilizing ontology-injected prompt learning, the LLM achieves standardized extraction of multi-source diagnostic knowledge. A dynamic integration and updating mechanism for the knowledge graph, based on dual-layer similarity calibration involving character comparison and an embedding model, is implemented to autonomously build an integrated diagnostic knowledge graph base. Secondly, leveraging entity fuzzy retrieval that combines LLM and word embeddings, along with a power-encoding-based instant knowledge graph distillation method, the approach incorporates fault subgraph structural features and contextual knowledge while visualizing fault propagation paths via graph nodes. This significantly enhances the logical completeness of fault root cause analysis and maintenance strategy generation by the general-purpose LLM. Validation using diagnostic texts and FMEA tables of the solar array drive assembly (SADA) shows that, compared with traditional RAG methods, the proposed KG-RAG method combined with fault subgraphs improves the keyword F1-score by 70.88% and semantic similarity by 11.60% in intelligent diagnostic Q&A. The results show superior accuracy and interpretability over using LLM or RAG alone, providing substantial theoretical support and a technical pathway for intelligent fault diagnosis of spacecraft electromechanical equipment.

Abstract:Aiming at the problems of large discrepancies between dynamic model response and the distribution of measured data, as well as the poor generalization in fault diagnosis methods driven by simulation data, a gearbox fault diagnosis method based on digital twin driven wavelet attention transfer network is proposed. Firstly, a virtual model of the gear transmission system was built using the lumped parameter method to map the physical system. Key parameters were optimized with measured normal data, and fault mechanism models were integrated to generate a rich twin fault dataset. Secondly, a discrete wavelet attention-based feature extraction network was designed, combining the multi-scale signal decomposition capability of discrete wavelet transform with the channel attention mechanism that dynamically focuses on strongly correlated fault features. This model effectively extracts domain-invariant fault features from both twin and measured data in the wavelet domain. Then, to address differences in marginal and conditional distributions between twin and measured data, a joint subdomain adaptation criterion was proposed by combining maximum mean discrepancy (MMD) and dynamic local MMD. This criterion measures the joint feature distribution discrepancy between the two domains, enabling the transfer of the gear twin model to real-world fault diagnosis. Finally, the proposed method was experimentally validated on a multi-stage parallel gearbox test bench. Results showed that it achieved superior diagnostic performance across all transfer tasks, with an average classification accuracy of 98.10%. The method effectively enables fault diagnosis transfer from twin data to measured data under conditions of limited labeled high-quality fault data.

Abstract:Mechanical equipment operating under complex working conditions is highly prone to failure. If such failures are not diagnosed in a timely and accurate manner, they may lead not only to performance degradation and economic losses but also to serious safety accidents. Therefore, developing efficient and reliable intelligent fault diagnosis methods is of significant engineering importance. However, in real industrial scenarios, the number of fault samples in monitoring data is usually limited, resulting in data imbalance problems that severely constrain the accuracy and robustness of traditional diagnostic models. To effectively mitigate this issue, this paper introduces spectral graph convolution and a hybrid attention module, and proposes an improved cycle-consistent generative adversarial network for generating high-quality fault samples, thereby enhancing intelligent mechanical fault diagnosis under imbalanced data conditions. Specifically, spectral graph convolution models global pixel dependencies through sparse adjacency matrices, improving long-range feature interactions while reducing computational complexity. Meanwhile, the hybrid attention module dynamically assigns weights at both channel and spatial levels to highlight critical regions and strengthen feature representation. With the proposed improved cycle-consistent generative adversarial network, more realistic and diverse fault samples can be generated, effectively augmenting minority-class data and alleviating the limitations imposed by data imbalance on intelligent fault diagnosis performance. Experimental results on the Beijing Jiaotong University metro bogie dataset and the Soochow University bearing dataset show that the proposed method significantly outperforms comparison approaches in three image quality evaluation metrics and fault classification accuracy. These results validate its diagnostic effectiveness under imbalanced data conditions and demonstrate that it provides a practical and feasible solution for addressing data imbalance challenges in industrial applications.

Abstract:Precise condition assessment of brushless direct current motors (BLDCM), as critical power units in industrial drive systems, is essential for ensuring their safe and stable operation. However, insulation degradation between adjacent stator windings can lead to inter-turn short circuit(ITSC). Early-stage ITSC present significant challenges, including spatially scattered and highly redundant characteristic information, posing new difficulties for equipment evaluation. To address these issues, this paper proposes an intelligent ITSC identification method for inter-turn short circuits in BLDCM based on electromagnetic field feature distillation. Firstly, an electromagnetic equivalent model is established to analyze the physical relationship between leakage flux signals and winding states, revealing the nonlinear law governing the spatial redistribution of leakage flux relative to the number of shorted turns. Secondly, multi-source spatial leakage flux signals are fused using recursive phase space reconstruction techniques to construct high-information-density electromagnetic field feature maps, overcoming the limitations of traditional single-dimension analysis.Finally, an augmented training set covering diverse noise scenarios is constructed to significantly improve the model′s robustness in complex industrial environments. Through the synergistic optimization of multi-source feature fusion, lightweight architecture, and anti-interference strategies, this research provides an innovative solution for highly reliable online monitoring of motors in industrial settings.

Abstract:To address the issues of low detection accuracy in existing flotation foam-layer thickness measurement, large errors in interface determination, and poor adaptability to complex operating conditions, this paper proposes a thickness-detection method for flotation foam layers based on a Mann-Kendall adaptive threshold and proportional-fitting compensation. Firstly, an electrical conductivity model for industrial flotation processes is established, and the differences in conductivity characteristics between the pulp layer and the foam layer are analyzed. Secondly, a trend-testing method based on the Mann-Kendall statistic is introduced. By assessing the monotonic trend of the overall voltage sequence, the presence of a foam layer is identified. Combined with the voltage difference at the pulp-foam interface, an adaptive threshold suitable for various operating conditions is constructed based on the Pauta criterion to accurately determine the interface position. Finally, an equivalent-resistance model of the interface is established according to the relationship between gas holdup and foam-layer conductivity. A foam-thickness compensation algorithm based on the proportional coefficient between the pulp layer and the foam layer is proposed to correct measurement deviations caused by operating-condition disturbances and foam nonuniformity, thereby enabling accurate measurement of flotation foam-layer thickness. Simulation and field-measurement results show that the Mann-Kendall adaptive-threshold method can accurately identify the pulp-foam interface under different flotation conditions. Compared with existing methods, the proposed proportional-coefficient fitting approach provides better overall foam-layer thickness estimation with higher measurement accuracy. The absolute error of thickness compensation is within 5 mm, and the actual measurement error of the final flotation foam layer thickness is within ±1 cm, which meets the requirements for measuring the foam layer thickness in the industrial field of the flotation process and provides accurate and reliable technical support for the real -time monitoring and automatic control of the flotation process.

Abstract:Currently, the multi-load magnetic coupled wireless power transfer (MC-WPT) system is designed for AC application scenarios, which involves an increase in the cascading links, making the system structure more complex and unable to achieve multi-frequency output. This paper proposes a hybrid envelope and sine wave pulse width modulation (HESPWM) control method and constructs a multi-frequency and multi-load MC-WPT system for AC/DC output. Firstly, this paper presents the basic structure of the multi-frequency and multi-load MC-WPT system for AC/DC output, elaborates on the HESPWM modulation mechanism, and provides the expression of the effective values of each frequency component of the inverter output under this control mode. Secondly, a half-wave energy pickup topology based on dual active full bridge and dual diodes is proposed, and the working mode of the AC output channel is analyzed in detail. The positive and negative half-wave power of the received end envelope signal are extracted through diodes, and the polarity switching is carried out with the active full bridge circuit, and then the two AC output channels are filtered to achieve interference-free mutual operation. To address the impact of inter-frequency interference on the output waveform quality, the design of a band-stop filter is optimized to suppress the inter-frequency interference. Then, an equivalent model of the dual-frequency system with AC and DC synchronous output is established, and the system transmission characteristics are derived. Finally, a simulation and experimental platform is built for verification. The simulation and experimental results show that the system achieves 20 kHz and 60 kHz dual-frequency three-channel AC/DC output under the control of the HESPWM modulation strategy. The inter-frequency interference in the AC/DC output channels is suppressed with the action of the band-stop filter, and the output waveform quality is significantly improved. Moreover, the system maintains stable operation under load switching conditions, and the peak efficiency of the entire machine under different working conditions reaches 75.9%.

Abstract:To address the decline in coupling coefficient and transmission efficiency of traditional flat solenoidal magnetic coupling structures in wireless power transfer (WPT) systems under lateral misalignment, this article proposes an engineering-oriented concave flat solenoidal coil design method with enhanced misalignment tolerance. First, based on the first harmonic approximation (FHA), the voltage gain characteristics of typical compensation topologies are analyzed, and an analytical relationship between voltage gain G and mutual inductance M is established. It shows that reducing the sensitivity of M to lateral displacement can improve system tolerance to misalignment. Secondly, from the perspective of coil arrangement, the effects of winding distribution, turn spacing, and concave end angle on magnetic field uniformity and coupling coefficient retention are investigated, and trade-off and optimization methods are proposed. Subsequently, magnetic coupling structures are modeled in Ansys/Maxwell, and their magnetic field distributions and coupling variations under lateral misalignment are compared. The results shows that a non-uniform winding distribution with a 30° concave end angle can maintain a high coupling coefficient and stable transfer performance under ±60% lateral displacement. For lightweight and integrated design, the secondary side is implemented with a flexible printed circuit board (FPC) coil, meeting the miniaturization and high-power density requirements of medium- and small-scale portable devices. Finally, a 100 W prototype shows that within a misalignment range of ±15 mm along the X-axis and ±30 mm along the Y-axis, output voltage fluctuation is within 4%, and maximum transmission efficiency reaches 87.3%. These findings validate the effectiveness and engineering applicability of the proposed magnetic coupling structure in enhancing lateral misalignment tolerance and system performance.

Abstract:With the increasing complexity of mixed-signal chips, the number of test items has also grown significantly. However, rapid production cycles and shortened product lifespans impose stringent constraints on the time available for engineers to refine the testing strategies. Existing methods primarily rely on fully characterized results, particularly from failed chips, making them incompatible with stop-on-first-failure mechanism where complete failure characterization is impractical. Another widely used index, the process capability index, assumes normally distributed results, limiting its applicability for non-normal characteristics. To address these limitations, this paper proposes a fuzzy comprehensive evaluation method. It integrates the extra information gain cost(EIGC), the distribution characteristics of test results, and the proximity to upper and lower limits of metrics, making it applicable to a wider range of test item evaluations. Unlike previous methods, our approach solely utilizes the database of passing chips to compute EIGC, enabling the identification and removal of less informative test items without requiring failed chip data, and therefore suitable for stop-on-first-failure mechanism. The methodology has been validated on the binning processes and final test of two different mixed-signal chips. Experimental results show that the fuzzy comprehensive evaluation method with EIGC information can effectively handle non-normally distributed characteristics, and achieves test-time reductions of 66.23% and 28.12%, respectively, in the chip binning process and final test, while maintaining an extremely low defect-escape rate.Our findings provide a scalable and efficient solution for mixed-signal chip test optimization, reducing reliance on failure characterization and improving test efficiency in high-volume manufacturing environments.

Abstract:In electric vehicles, the electrical contact points within the main circuit are susceptible to series arc faults as a result of poor contact and other factors, which poses a significant threat to the safety of vehicle occupants. First, this study constructed an experimental platform centered on the Geely Emgrand EV450 electric vehicle and undertook experimental investigations on arc faults in electric vehicles. With the power supply terminal voltage serving as the research focus, a one-dimensional convolution based arc fault detection model was developed. However, traditional deep learning models encounter two core challenges in practical applications. First, the models have poor interpretability, making it difficult to clarify the key basis for fault detection. Second, they have a large number of parameters and high computational complexity, making it difficult to meet the strict requirements for real-flortime and lightweight fault detection in electric vehicles. To address these issues, this study adopts a network structure search strategy integrating multi-objective optimization. which incorporates accuracy, interpretability indicators, and floating point operations into the search objectives of the network structure search. Through multidimensional trade-offs, adaptive optimization of the network structure is achieved, which effectively improves the initial performance of the model. Subsequently, a feature channel merging strategy was developed by integrating dynamic time warping, particle swarm optimization, and simulated annealing algorithms. Among these methods, Dynamic time warping can measure the similarity of outputs from different channels; Particle swarm optimization, with its global search capability, quickly locates potential optimal channel merging combinations; and Simulated annealing further enhances the rationality and effectiveness of channel merging. Using this strategy, an accurate, interpretable, and lightweight arc fault detection model for electric vehicles has been successfully developed. Finally, generalization analysis and comparative analysis with other detection methods confirm that the model exhibits excellent performance in electric vehicle arc fault detection.

Abstract:The analysis and evaluation of the power performance of tidal energy converters are crucial for promoting the iterative upgrading of tidal current energy generation technology. However, analysis of field testing datasets of tidal energy converters shows that the tidal current velocity data and output electric power data do not always have a one-to-one correspondence, which is not conducive to refined analysis of the power generation performance of tidal energy converters. In view of this, based on previous field test data, methods such as the Savitzky-Golay （SG） filtering algorithm, data interpolation algorithm, comparative analysis, and mathematical statistics were adopted to conduct research on the analysis methods of the power performance of tidal energy converters, focusing on indicators such as the output power, overall conversion efficiency, annual power generation, capacity factor, and annual equivalent full-load hours of the devices. The research results show that the SG filtering method has a good effect on filtering out the fluctuations of tidal current velocity data, especially for the tidal current velocity data in significantly oscillating intervals. The cut-in velocity characterized by the scatter plot of the output power of the tidal energy converter after interpolation is 0.5 m/s, and the cut-in velocity index is decreased by 7.4% compared with that before interpolation. The maximum output power in the dataset after interpolation is 542.9 kW, which is 0.6% higher than the maximum output power before interpolation. The maximum overall conversion efficiency in the interpolated dataset is approximately 43.4%, which is 1.9% higher than the maximum overall conversion efficiency of 42.6% before interpolation. The differences in the annual power generation, capacity factor, and annual equivalent full-load hours of the tidal energy converter between the datasets before and after interpolation are 700.8 kWh, 0.000 2, and 1.6 h, respectively. These findings provide a valuable reference for the refined performance analysis of tidal energy converters.

Abstract:This article investigates the voltage rise problem induced by regenerative braking in heavy-load electrified railways with long gradients and proposes a stability evaluation method based on the rise mechanism. An iterative model is formulated by deriving vector relationships among no-load voltage, rise voltage, load voltage, and braking current, revealing the formation of overvoltage during braking. The effects of power factor, regenerative power, load position, and system instability on voltage stability are analyzed. Results show that a higher power factor facilitates voltage balance under combined traction and braking, significantly mitigating voltage rise, while the influence of regenerative power variation is limited. Voltage rise is most severe when the load is located at the end of the power supply arm and weakest near the substation. Parallel power supply enhances system redundancy and suppresses fluctuations. Further analysis indicates that no-load voltage level plays a dual role, which is that higher values improve adaptability to regenerative power but increase overvoltage risk. However, lower values reduce rise but constrain traction capacity. A recommended range of 27.5~28.5 kV is suggested. In addition, introducing inductive loads is shown to effectively suppress voltage rise by enlarging the no-load voltage angle and reducing the load voltage angle. An engineering application requires dynamic adjustment of inductive loads between regenerative and traction states. This study provides theoretical and practical guidance for voltage stability regulation in long-gradient electrified railways.

Abstract:The hydrophone is an indispensable component of sonar systems for acquiring underwater acoustic information, and its inherent noise level directly impacts sonar performance. It is essential to accurately determine the equivalent noise spectrum level of a hydrophone, both during its development phase and throughout its operation and maintenance stages. This article addresses the challenge that hydrophone equivalent noise is extremely weak and its measurement process is susceptible to interference from external environmental sounds, vibrations, and electromagnetic waves. This study proposes a measurement environment construction scheme combining sound insulation, vibration reduction, and electromagnetic shielding. Building upon this well-constructed measurement environment, we further employ low-noise preamplifier circuit design and the averaged periodogram power spectrum estimation method to achieve accurate measurement of the hydrophone equivalent noise spectrum level. The article provides estimation methods for vibration reduction and electromagnetic shielding effectiveness, along with measures to effectively suppress circuit noise. A measurement system is developed using the proposed method. The sound insulation and electromagnetic shielding effectiveness of the system are tested, showing an acoustic attenuation greater than 70 dB and an electromagnetic attenuation greater than 60 dB, effectively satisfying practical requirements. The effectiveness of the averaged periodogram power spectrum estimation method is evaluated through experimental results. It is also concluded that for different measurement frequency bands, users should select appropriate total acquisition duration and segment numbers to balance measurement efficiency with accuracy. Finally, the measurement system is used to determine the equivalent noise spectrum level of the RESON TC4014 low-noise hydrophone. The results show that the experimentally obtained equivalent noise spectrum curve for the TC4014 coincides with the curve provided by the manufacturer, validating the effectiveness of the proposed method.

Abstract:To address the limitations of existing ocean models for underwater unmanned system research, such as single parameter, non-dynamic, cannot realistically simulate the marine environment of arbitaryre sea area, a multi-parameter marine database based on ocean big data are established. Leveraging the efficient retrieval capability of the database, marine data blocks are rapidly acquired, and conversion of kinematic positions to geographic coordinates was performed. A smooth interpolation algorithm of temporal and spatial ocean data based on slippage was proposed to build a RTPDCOM, which avoids the issue of data overshooting at great depth; On this basis, a profiling float dynamics model was further established, and prediction of the water outlet location was carried out. Comparisons between simulated and sea trial data in the western Pacific region reveal that, under 4 000 m profile motion conditions, the horizontal distance between the modeled and measured water inlet and outlet points remainu largely consistent, the angular deviation of water outlet points is mostly within 20°, and the distance deviation is mostly within 650 m. Moreover, the temperature and salinity profile data exhibit high concordance, with an average temperature deviation of 0.126℃~0.185℃ and an average salinity deviation of 0.027 6~0.031 4 PSU, validating the RTPDCOM model′s consistency with the real-world marine environment. Comparisons between simulated and sea trial data in the Indian Ocean region demonstrated that the angular deviation is mostly within 20°, and the distance deviation is mostly within 600 m, indicating excellent prediction performance and highlighting the RTPDCOM model′s universality across different sea areas. Additionally, compared to traditional layered models, the RTPDCOM yields smaller deviations in water outlet angle and distance. Its strong portability holds significant application value for ensuring the reliability and stability of unmanned undersea systems, and improving the precision of trajectory planning and formation control.

Abstract:In the process of attitude dynamic measurement for steering drilling tool, the bottom drilling tool combination interacts with the rock and the collision between the drilling string and shaft wall generates strong vibration, which leads to multi-frequency and high-amplitude noise interference in the original measurement signal, resulting in the extremely low signal-to-noise ratio of the drilling tool attitude measurement signal, or even completely annihilated in the noise. This severely affects the accuracy of attitude parameter calculation such as inclination angle. In order to solve this problem, this paper proposes an array tri-stable chaotic system detection method used for the dynamic measurement signal. Firstly, a variable-scale processing of the drilling tool measurement signals is performed on the drilling tool measurement signals, which involves reconstructing and transforming the frequency values of the characteristic signals to meet the constraints of phase transition of the tri-stable system output.Second, considering the deviation of traditional frequency detection methods caused by the random variation of the initial phase angle of drilling tool measurement signals,a frequency detection model based on the array tri-stable chaotic system is proposed, relying on the collaborative work of the different chaotic equations for driving signals, a full-phase coverage frequency detection method is realized to eliminate the influence of the initial phase angle of drilling tool measurement signals on the frequency detection results.Finally, a parameter estimation model based on another array tri-stable chaotic system is designed to synchronously estimate the amplitude and phase of the drilling tool measurement signal, and then recovers the complete drilling tool measurement signal. Simulation and real drilling data experiments show that this method can detect the signal-to-noise ratio threshold of the measured signal as low as -18 dB, and that the errors of the inclination are lower than 1°, Compared with the signal distortion problem of traditional filtering methods in extremely low SNR scenarios and the detection deviation problem of bi-stable chaotic systems due to incomplete phase coverage, this method exhibits significant advantages in both signal detection accuracy and inclination angle calculation precision.

Abstract:To address the issue that the induction coil of existing magnetic-field-based time-grating angular displacement sensors couples with different magnetic-field characteristics at different air-gap heights, resulting in low utilization of the effective time-varying magnetic-field area, weak magnetic-field pickup capability, small induced-signal amplitude, and consequently increased residual errors within a pole pitch, this study proposes a time-grating angular displacement sensor featuring a three-layer complementary coil-shape assembly. A mathematical model is established to analyze the spatial distribution of the excitation magnetic field in the air gap. Based on this, a stratified coupling theory is developed, enabling the air-gap magnetic field to be categorized into three types. Based on this theory, a measurement model of the proposed sensor is constructed. The excitation coil adopts a double-layer complementary winding structure that enables mutual compensation of magnetic-field constraints at the ends of the winding, resulting in a more uniform excitation field. The induction unit employs a three-layer complementary coil-shape assembly, in which coils of different geometries are placed at different air-gap heights to couple with their corresponding air-gap magnetic-field types. This design significantly improves the amplitude and stability of the induced signal. The principle of magnetic field coupling of planar induction coils and the sensor signal processing method are analyzed: two channels of excitation signals are applied to the excitation coil, the measured traveling wave signal is obtained from the induction coil, and the angular displacement is calculated through phase discrimination. The error analysis and structural parameter optimization of the sensor are carried out through electromagnetic simulation, and the sensor prototype was made by PCB process for experimental verification. The simulation and experimental results show that compared with the traditional single-layer coupling structure, the measurement accuracy of the sensor is improved by 12.6%, the harmonic error introduced by the space air gap magnetic field is reduced, the amplitude and stability of the induced signal are improved, and the signal-to-noise ratio is improved. The optimal installation gap of the sensor is 0.6 mm, and the measurement accuracy of the sensor is ±83″.

Abstract:Optical voltage sensors(OVS) represent the developmental direction of next-generation voltage transformers. The primary challenge currently facing OVS is the ineffective resolution of linear birefringence in electro-optic crystals induced by temperature drift, which has become a bottleneck restricting their practical application. To address this issue, this paper proposes a dynamic decoupling and compensation method for linear birefringence in OVS electro-optic crystals under the influence of temperature drift. First, a mathematical model characterizing the linear birefringence of bismuth germanate oxide(BGO) crystals is established based on the crystal′s optical indicatrix equation. The generation mechanism and dynamic characteristics of linear birefringence and electro-optic phase retardation are analyzed, providing a theoretical basis for the decoupling of linear birefringence and electro-optic phase retardation in OVS. Then, a linear birefringence decoupling method based on linear OVS demodulation is proposed. Its demodulation result is the linear superposition of electro-optic phase retardation and linear birefringence. The electro-optic phase retardation is directly related to the measured electric field, while linear birefringence is introduced by ambient temperature variations. Due to the distinct generation mechanisms of the two, linear birefringence can be separated from the phase retardation and compensated for. On this basis, an OVS linear birefringence compensation method based on AC voltage zero-crossing detection is proposed. Experimental results demonstrate that the linear birefringence of electro-optic crystals in the proposed method is measurable and can be compensated. Within the temperature range of -20℃ to 70℃, the temperature characteristics of linear birefringence in BGO crystals are studied and analyzed. After compensation using the proposed method, the ratio error of OVS is less than 0.257 9% and the phase error is less than 14.254 7′. Finally, an experimental platform for optical power fluctuation and vibration is built to test the optical power independence and vibration stability of the proposed method. The linear demodulation mode based on OVS provides a new perspective for solving the linear birefringence problem in OVS electro-optic crystals.

Abstract:Conventional electromagnetic acoustic transducers (EMATs) typically empoy either permanent magnets (PMs) or direct current (DC) including unidirectional pulse currents and unidirectional sinusoidal currents, to provide the required bias magnetic field. However, PMbased transducers are inherently rigid, which limits their applicability for in-situ inspection of curved structures, while DC bias fields generally require large current magnitudes and impose stringent experimental conditions. To address these limitations, a curved dual-AC magnetostrictive transducer is proposed, which generates the bias magnetic field using alternating current. A theoretical model for ultrasonic excitation based on the dual-AC magnetostrictive mechanism is established. Finite element analysis demonstrates that the proposed transducer can excite a pure shear horizontal (SH) guided wave mode without modal conversion after interacting with defects. As the curvature radius increases, the curved transducer exhibits a smaller standard deviation of the excitation magnetic field and lower signal amplitude fluctuation. Using 95% of the peak magnetic field intensity as the criterion for field uniformity, the uniform magnetic field region of the curved transducer ranges from 22 to 24.34 mm, compared with 17.32 to 20.29 mm for the planar electromagnetic acoustic transducer (EMAT). Furthermore, orthogonal experimental results indicate that the influence of structural parameters on signal amplitude follows the order: number of layers > number of turns > length > wire diameter > height. A dual-AC magnetostrictive SH guided wave system is subsequently constructed. Experimental results indicate that when the magnetization current is 12.3 A with 40 coil turns, the defect signal amplitude of the curved transducer is enhanced by 18.64% compared to that of the planar transducer. Increasing the magnetization current to 30 A yields an amplitude increase of 35.71%, and further increasing the number of turns in the magnetization coil to 100 results in an amplitude enhancement of 85.59%. Repeated experiments with different curvature radii show that the signal amplitude of both transducers increased with curvature radius, while the curved transducer exhibited smaller overall errors and amplitude fluctuations. These results confirm the feasibility of dual-AC magnetostrictive excitation for SH guided waves and demonstrate the effectiveness of the proposed curved transducer for defect detection on curved surfaces.

Abstract:To address the challenges of limited space in fully mechanized excavation faces and difficulty achieving long-term high-precision roadheader pose monitoring with single sensors, a dual filtering fusion method using heterogeneous sensors is proposed. First, a laser vision subsystem employing an improved adaptive Canny operator identifies and analyzes laser spot geometry to obtain lateral offset, vertical offset, and three-axis attitude. Second, an ultra-wideband (UWB) subsystem based on an improved weighted Chan-Taylor hybrid method effectively compensates for non-line-of-sight errors and reduce dependency on initial values in traditional UWB positioning. A roadheader body attitude calculation compensation model is also established to reduce the impact of environmental signal reflections on positioning accuracy. Building upon these subsystems, a complete dual filtering combined pose fusion framework is constructed. An improved adaptive extended Kalman filter algorithm is applied to perform primary filtering and noise reduction on the pose outputs from both the laser vision and UWB subsystems. Subsequently, an adaptive weighting algorithm conducts secondary fusion filtering on the redundant pose parameters, compensating for UWB calculation errors while overcoming laser vision data loss from temporary target loss, thereby achieving complementary advantages of multi-source sensors. Finally, based on a scaled prototype of the EBZ200 roadheader, an experimental platform for multi-source sensor combined pose perception is established. Results demonstrate that the multi-source heterogeneous sensor filtering fusion pose perception system achieves a machine body position detection error of less than 13 mm and an attitude detection errors of less than 0.8°, providing an effective technical solution for continuous precise positioning of roadheader in fully mechanized excavation processes.

Abstract:Cable joints represent the most vulnerable component in power cable systems. Detecting partial discharge (PD) in cable joints is the most effective method for preventing cable system failures. However, the unique structure of cable joints makes it extremely challenging to apply conventional sensors for PD detection within the insulation of cable joints. Building upon the flexible sensor unit, this paper proposes a PD signal sensing method designed for deployment in the confined spaces between the outer shielding layer and the protective sheath of the joint. Firstly, finite element simulation software was employed to design and analyze the effective operating frequency band of the flexible sensor unit within the cross-linked polyethylene (XLPE) medium. Through adjustments and testing of its radiation pattern, the operational bandwidth was optimized to cover typical partial discharge signals generated during cable joint failures, thereby ensuring that its detection capability is maintained when the sensor is in a bent state. Secondly, the propagation characteristics of high-frequency electromagnetic radiation from PD signals through the elongated multi-layer dielectric space of a cable joint were calculated to determine the optimal installation position for the flexible sensor unit. This analysis confirmed that this position for the flexible sensor unit to be near the stress cone of the cable joint. Finally, an experimental platform for cable joint PD was established to analyze the performance of the flexible sensor unit. Simulation and experimental results demonstrate that the flexible sensor unit operates effectively within a frequency band of 50 to 640 MHz in the XLPE medium, with a bending radius of 6~10 cm without compromising detection performance. The amplitude of high-frequency electromagnetic radiation from PD signals is significantly enhanced near the stress cone of the cable joint. The flexible sensor unit can effectively detect composite defects located 12~24 cm from the PD source within the cable joint. The flexible sensor unit exhibits satisfactory detection performance in practical applications on 35 kV cables. It enables successful detection of PD signals within the confined spaces of cable joints.

Abstract:Micro-electro-mechanical system (MEMS) gyroscopes have found widespread application across various fields due to their core characteristics of low cost, compact size, and high practicality. However, their output accuracy deteriorates significantly under coupled temperature-vibration environments. Under combined wide-range temperature variations and persistent vibration, temperature drift and vibration errors overlap in the gyroscope output, weakening traditional single-factor compensation. Currently, research on systematic error compensation under coexisting temperature and random vibration is limited, with technical solutions still lacking. To address signal aliasing under temperature-vibration coupling, this paper proposes a variational mode decomposition-long short-term memory-Sage-Husa adaptive Kalman filter (VMD-LSTM-SHAKF) error compensation model based on virtual gyroscope technology, achieving simultaneous compensation of temperature drift and vibration errors in MEMS gyroscopes for the first time. In the proposed method, VMD is employed to separate the coupled signal, extracting temperature drift components while suppressing vibration interference. An LSTM network is then used to compensate the temperature drift, and an improved SHAKF fuses the outputs of a four-gyroscope array to further suppress random vibration errors and enhance overall accuracy. A static temperature-varying experiment was designed to collect gyroscope array data under different temperature and vibration conditions. Experimental results show that after VMD-LSTM-SHAKF processing, the system′s 1σ standard deviation decreases to 0.033 9°/s, angle random walk to 0.555 8°/h, with a 94.32% error reduction compared to the best single gyroscope. This study provides an effective solution for MEMS gyroscope error compensation in complex environments, offering engineering application insights.

Abstract:To address the challenge of degraded positioning accuracy and reliability in smartphone global navigation satellite system (GNSS) signals caused by multipath and non-line-of-sight effects in complex environments such as urban canyons and dense high-rise areas, this paper proposes a resilient graph optimization-based global navigation satellite system/pedestrian dead reckoning (GNSS/PDR) autonomous navigation method within a multi-source integrated navigation system framework. Focusing on system detectability and reconfigurability, the approach designs an elastic fault-tolerant architecture that incorporates state-adaptive fault detection and gradient descent regression-based autonomous reconfiguration, aiming to enhance the robustness of the navigation system under dynamically changing environmental conditions. In terms of detectability, a state-correlated dynamic fault detection mechanism is introduced. When no fault is detected in the previous epoch, a sliding-window-based dynamic 3σ statistical detection method is applied. When a fault is detected in the previous epoch, a dynamic threshold strategy based on exponentially weighted moving average is employed for continuous anomaly monitoring. In terms of reconfigurability, once a fault is identified, the system performs fault diagnosis and autonomous reconfiguration using a gradient descent regression-based GNSS/PDR algorithm. The reconfiguration process first utilizes historical innovations to predict the system state, then performs magnitude correction based on the dynamic relationship between abnormal and repaired innovations, and finally achieves dynamic recovery of abnormal observations. Experimental results demonstrate that the proposed method reduces the average positioning error by more than 20% compared to traditional extend Kalman filter (EKF), Huber-based M-estimation EKF, factor graph optimization (FGO), and Huber-based M-estimation FGO algorithms. These findings indicate that the proposed method offers significant advantages in enhancing the positioning accuracy and robustness of smartphone-based pedestrian navigation under challenging multipath and NLOS environments, providing a valuable reference for the development of high-precision positioning applications in future consumergrade devices.

Abstract:To address the shortcomings of existing gecko-inspired adhesive structures in real-time state sensing and active control capabilities, this paper proposes a bio-inspired multiscale adhesive structure with sensing capability, based on by the adhesion regulation method and sensing mechanism of the gecko′s lamella-setae hierarchical structure. The structure consists of a millimeter-scale boot-shaped elastic substrate integrated with a micrometer-scale mushroom-shaped adhesive array. By utilizing shear motion to achieve controllable adjustment of the contact area, the structure uses forward shear to increase the contact area for strong adhesion, while reverse shear enables easy detachment through interface peeling. A simplified inclined prism model of the multiscale structure was established to theoretically analyze the evolution of normal stress at the bottom surface of the boot-shaped structure during the preloading, shearing, and peeling stages. It was found that the normal stress at the inner edge of the bottom surface is sensitive to preload and shear forces, and its abrupt change can serve as an effective feature for identifying the peeling stage. Based on the bilinear traction-separation theory, a finite element model of the multiscale adhesive structure was developed to simulate the complete adhesion process, validating the controllable adhesion mechanism and the phased characteristics of stress evolution at the bottom surface. Guided by theoretical and simulation analyses, a thin-film pressure sensor was integrated into the inner backing layer of the bio-inspired multiscale adhesive structure, and experimental tests were conducted on its adhesion and sensing performance. The results demonstrate that shear motion significantly enhances the adhesion performance of the multiscale structure, with the adhesion force reaching a maximum value of 1.98 N at a displacement of 1.4 mm. Moreover, the sensing signals exhibit a clear correlation with the normal preload and shear displacement, enabling effective identification of adhesion, slip, and detachment states. The developed opposing-grip bio-inspired adhesive gripper prototype has successfully achieved stable grasping, controllable release, and state sensing on various smooth surfaces, such as glass and silicon wafers, with a maximum load capacity of 1 kg, verifying its potential for robotic manipulation applications.

Abstract:Dimensional measurement is a critical step in the visual inspection of industrial products. Traditional contact-based measurement methods suffer from low efficiency and significant susceptibility to subjective factors. Meanwhile, vision-based measurement requires tailored dimensional boundary extraction schemes for different objects, and high-precision 3D measurement techniques often involve high development complexity and limited applicability. To address these challenges, this study proposes a non-contact method for measuring main workpiece parameters based on the segment anything model 2 (SAM2) with RGB-D coordinate transformation. First, the mask segmentation performance of four traditional image segmentation algorithms—threshold segmentation, edge segmentation, color space segmentation, and GrabCut segmentation—is evaluated. Among them, GrabCut segmentation, identified as the optimal traditional method, is compared with mainstream deep learning segmentation algorithms and SAM2 to demonstrate the superiority of SAM2. Subsequently, a binocular stereo vision experimental platform is constructed to capture high-precision point clouds of workpieces. The point clouds undergo processing steps such as filtering, smoothing, and hole filling. Depth maps and RGB images of the target workpieces are then acquired. Leveraging SAM2′s zero-shot generalization capability, high-precision target segmentation is achieved on RGB images through positive and negative point interaction guidance, yielding initial masks. These masks are further refined via morphological optimization and connected component analysis to generate topologically closed smooth masks. A feature skeleton is extracted using principal component analysis (PCA). Finally, perpendicular segments are generated along the skeleton, and geometric parameters are calculated by integrating 3D coordinates from the aligned depth maps. The measurement results of the proposed method are analyzed by comparing them with those obtained using digital calipers and point cloud data. Experimental results demonstrate that, in measurements of sleeves, pliers, and motors, the mean absolute error for sleeve diameter is 0.0175 mm, while the mean absolute errors for plier and motor parameters are 0.028 3 and 0.023 7 mm, respectively, all meeting the required precision standards.

Abstract:To address the challenge of accurately reconstructing micro-defects on highly reflective surfaces during in-service inspection of aero-engine blades, a three-dimensional (3D) reconstruction method based on a binocular endoscopic system is proposed. To overcome the limitations of insufficient calibration accuracy in endoscopic scenarios, a concentric dual-ring calibration target is designed, and a feature point extraction and optimization algorithm based on concentricity constraints is developed to achieve high-precision binocular calibration. Experimental results demonstrate that the proposed method achieves average single-camera reprojection errors of 0.095 and 0.103 pixels, respectively, while significantly reducing binocular calibration error and improving the accuracy of system geometric parameters. For stereo reconstruction, a deep learning-based detection model, YOLO11, is integrated to automatically locate defect regions and obtain prior information of the detection bounding boxes in both left and right views. A region-constrained and prior-disparity filtering strategy based on the detection boxes is proposed, which determines the intersection of the corresponding epipolar search regions and disparity ranges. This approach confines the matching computation to the defect areas, preventing the propagation of smoothing costs from incorrect matches outside the detection boxes, thereby enhancing stereo matching stability and local reconstruction accuracy. Moreover, the AD-Census stereo matching algorithm is adaptively modified to further suppress noise. The reconstructed point clouds of scratches and pits exhibit relatively uniform density. Measurement results indicate that the relative errors of pit diameter and scratch length are both less than 1%, and the depth measurement error does not exceed 8%, meeting engineering accuracy requirements. The proposed method demonstrates strong robustness and precision under complex illumination and spatially constrained conditions, outperforming conventional approaches in terms of calibration accuracy, local reconstruction quality, and measurement reliability. This work provides an effective technical foundation for high-precision 3D surface morphology measurement of in-service aero-engine blades.

Abstract:Multi-sensor simulataneous localization and mapping (SLAM) mitigates single-sensor limitations, yet current methods still face challenges such as monocular scale ambiguity, inaccurate intrtial measurement unit (IMU) initialization, and limited local mapping precision. This paper proposes a tightly-coupled, factor graph-based SLAM approach that fuses data from three heterogeneous sensors: a 3D light detection and ranging (LiDAR), an IMU, and a camera. For initialization, LiDAR data provides depth for visual features, and outliers are removed through neighborhood selection and statistical optimization to improve accuracy. Visual, LiDAR, and IMU data are then fused to jointly estimate IMU biases and gravity direction, reducing vertical map drift. For local optimization, factor graphs dynamically maintain keyframes and local maps within sliding windows. Visual constraints are refined through co-visibility projection matching, efficiently purging redundant map points while boosting accuracy and robustness. Global optimization incorporates loop-closure factors detected via specialized algorithms and applies incremental optimization to the factor graph, suppressing cumulative error without compromising real-time performance. The proposed method is evaluated on KITTI, M2UD extreme weather, and real-campus datasets. It reduces the absolute trajectory error by 53.1% on KITTI, 66% in M2UD rain/snow scenarios, and 20.3% in campus environments compared to LIO-SAM. The resulting maps exhibit higher structural consistency and geometric accuracy in both overhead and side views.

Abstract:To address the inherent limitations of traditional division-of-focal-plane polarimeters, namely low spatial resolution and an inability to acquire complete polarization information, this paper presents a hardware-software co-designed full polarization super-resolution imaging system. On the hardware side, a dual-channel, common-aperture full polarization imaging system was developed to simultaneously capture a low-resolution division-of-focal-plane linear polarization image and a high-resolution circular polarization image in a single exposure. Compared to conventional full polarization imaging systems, this design simplifies the system architecture, reduces costs, and eases assembly and alignment. To process the unique mixed-resolution data from this camera, a complementary feature-fusion-based full polarization super-resolution network was designed. This model employs a dual-branch architecture that simultaneously processes the two captured images for feature extraction. The high-resolution circular polarization image serves as precise guidance, and a feature fusion module enables effective integration of the low-and high-resolution features. To ensure feature alignment between the two branches, pixel-level registration is performed on the images from both detectors before super-resolution reconstruction, guaranteeing a spatial displacement error of less than one pixel. A loss function incorporating physical constraints is introduced to ensure the physical accuracy of reconstructed angular parameters, such as the angle of polarization and ellipticity of polarization. This process achieves high-quality reconstruction of full polarization parameters, including intensity, degree of polarization, angle of polarization, and ellipticity of polarization. On a real-world dataset, the peak signal-to-noise ratios of the four reconstructed polarization feature images improved by 0.106 dB, 0.302 dB, 0.117 dB, and 1.085 dB, respectively, while the structural similarity index measure improved by 0.002, 0.008, 0.006, and 0.014. In the field exploration experiments, this system effectively identified the unmanned aerial vehicle targets in complex scenes, verifying its significant advantages in target detection.

Abstract:Stepped-frequency ground penetrating radar (GPR) serves as a key non-destructive testing technology for road structure defect detection, where imaging resolution directly affects the reliability and accuracy of defect detection and identification. To address the problems of low imaging resolution in existing stepped-frequency GPR systems for road defect detection, as well as the difficulties in regularization parameter selection and heavy reliance on manual experience when applying compressive sensing methods, a one-dimensional high-resolution imaging method based on the alternating direction method of multipliers (ADMM) network is proposed. The proposed method unrolls the iterative process of the ADMM algorithm into a physically interpretable deep network structure, constructing an end-to-end learning framework consisting of a reconstruction layer, a nonlinear transformation layer, and a multiplier update layer. The reconstruction layer performs backpropagation calculation of signals, the nonlinear transformation layer imposes sparse constraints via a soft-thresholding function, and the multiplier update layer completes the iterative update of Lagrange multipliers. The collaborative work of these three layers enables the network to adaptively learn the optimal parameter combination through training. After obtaining the optimal sequence of reflection coefficients output by the network, it is convolved with a Ricker wavelet to finally generate a high-resolution one-dimensional image. To validate the feasibility of the method, simulation data for three scenarios were collected using gprMax electromagnetic wave propagation simulation software and measured data were collected using the team′s self-developed radar prototype. Simulation and experimental results demonstrate that the proposed method achieves excellent noise immunity while maintaining high resolution. Compared with the improved orthogonal matching pursuit (OMP) algorithm, it improves accuracy by 2% and enhances resolution by approximately two times. When compared to the standard ADMM algorithm, it achieves about a 1% improvement in resolution. These results fully validate the feasibility of the proposed method.

Abstract:The mechanical properties of polymer materials are intricately linked to their curing process. This relationship highlights the need for full-field monitoring techniques to provide rich and reliable experimental data. Due to the irreversible nature of curing, current methods cannot repeatedly measure a unidirectional curing process to get comprehensive, multi-dimensional information. To overcome this, we propose a novel multimodal full-field curing monitoring method that combines fluorescence imaging and tomographic interferometry. This approach leverages the high-sensitivity, full-field measurement capabilities of fluorescence digital image correlation and phase-sensitive optical coherence tomography to simultaneously estimate full-field strain on both the polymer's surface and internal cross-section during curing. We validated this by building a multimodal monitoring system with parallel channels for near-ultraviolet fluorescence excitation, blue fluorescence speckle imaging, and near-infrared tomographic imaging. Using backside illumination, we comprehensively monitored the curing of CharmFil Flow. During data acquisition, we calculated image correlations from fluorescence speckle to monitor full-field surface strain in the x-y plane. We then applied image correlation to tomographic results for internal transverse full-field strain along the x-direction in the x-z plane and used differential phase analysis for internal z-axis longitudinal full-field strain in the same plane. The consistency of our multimodal curing monitoring results was further confirmed by quantitatively characterizing time-domain shrinkage deformation across multiple dimensions. In essence, our method effectively achieves simultaneous multimodal full-field monitoring of both surface and internal polymer curing, offering a comprehensive and reliable measurement tool for in-depth studies of polymer curing dynamics and optimization of curing parameters.