Laser powder bed fusion (LPBF) technology, a cutting-edge process in metal additive manufacturing, has been successfully applied in high-end manufacturing sectors like aerospace. However, strong multi-physical field coupling effects frequently lead to dynamic instability in the molten pool, causing widespread porosity defects within fabricated parts and severely impacting forming quality stability. Traditional monitoring methods face limitations such as high cost and deployment difficulties, struggling to meet industrial production demands. To address these challenges, this paper proposes an online monitoring and intelligent internal quality discrimination method based on acoustic emission (AE)-deep learning fusion. An AE sensor-based online monitoring system for the LPBF process was developed. By continously monitoring AE signals throughout the entire process, the mapping relationship between AE signal characteristics and forming quality was revealed, creating a molten pool AE dataset comprising over 80,000 samples. To tackle the difficulty of extracting weak fluctuation features from the molten pool, a frequency domain feature extraction network based on the adaptive Fourier neural operator (AFNO) and a high-dimensional feature mapping classifier based on the Kolmogorov-Arnold network (KAN) were constructed. This approach analyzes molten pool dynamic characteristics through a multi-scale time domain feature fusion mechanism. By precisely mapping high-dimensional features using high-dimensional manifolds, the method achieves enhanced characterization of weak fluctuation features in AE signals and high-precision quality discrimination. Experimental results demonstrate that developed monitoring system effectively captures the dynamic behavior of the molten pool, and the proposed method achieves a quality discrimination accuracy exceeding 97%.

Liquid rocket engine turbopumps operate under severe non-stationary conditions, making it challenging for traditional vibration signal analysis methods to effectively extract fault features. To address this challenge, the cyclostationary random signal model is extended and a generalized cyclostationary analysis framework is established. This framework preserves the advantages of cyclostationary methods in fault diagnosis while broadening their applicability to non-stationary operating regimes. Focusing on vibration signal modeling, fault feature extraction, and characterization, a comprehensive generalized cyclostationary analysis framework specifically for rocket turbopump fault diagnosis is proposed. The superiority and validity of the established theoreical system are demonstrated through a cryogenic bearing operation experiment on a rocket turbopump and a cavitation fault simulation test on a centrifugal pump. Results indicate that vibration signals from rotating machinery can be regarded as approximately cyclostationary processes subject to time warping, which can be further transformed into modulated cyclostationary signals. In the rocket turbopump cryogenic bearing operation experiment, fault feature signals are extracted using the proposed blind adaptive cyclostationary-nonstationary signal extraction method. Its order-frequency spectral correlation map clearly detects spectral lines corresponding to the fundamental train frequency (0.42 Hz) and the ball pass frequency outer race (5.08 Hz). In the centrifugal pump cavitation fault simulation experiment, the proposed high-precision reassigned spectral correlation estimation technique enhances the localization of the blade-pass frequency (197 Hz) in the spectral correlation map. Furthermore, it reliably identifies fault features even under severe noise induced by increasing cavitation levels.

The increasing complexity of intelligent equipment and evolving operation and maintenance demands within Industry 4.0 highlight the inadequate adaptability of traditional maintenance decision-making methods in dynamic environments. Reinforcement learning (RL)-based maintence decision-making technology offers a paradigm for intelligent equipment maintenance by enabling autonomous strategy optimization through environmental interaction. This paper systematically explores the integration of RL theory and maintenance decision-making, focuses on 76 peer-reviewed articles published between 1954 and 2024. Core RL algorithms, including SARSA, Q-Learning, and Actor-Critic, are thoroughly examined and analyzed. The current state of intelligent equipment maintenance decision-making technology is also analyzed in depth. Typical application scenarios for RL in equipment maintenance decision-making are comprehensively dissected across four key areas: industrial manufacturing, energy, aerospace, and transportation. The study also identifies and discusses the core challenges facing current technology, such as algorithm convergence speed, computational efficiency, model interpretability, and issues related to data acquisition and privacy. This research provides a theoretical reference for algorithm innovation and engineering implementation in the field of intelligent operation and maintenance, fostering the deeper application of RL in maintenance decision-making.

Current research on fault diagnosis for aircraft complex motion mechanisms primarily focuses on system functional failure analysis, neglecting a comprehensive understanding of the correlation between motion characteristics and actual faults. This study investigates fault diagnosis methods for complex motion mechanisms and proposes a three-tiered framework encompassing data generation, feature processing and data analysis to address this limitation. The framework utilizes dynamic modeling and a fault parameter system to generate a dataset of time-series signals representing typical fault conditions. One-dimensional time-series data are mapped using two-dimensional image conversion methods, constructing multidimensional tensors through feature-level fusion based on sensor types and feature extraction methods of the complex motion mechanisms. A deep learning-based fault diagnosis model is employed for precise fault identification of complex motion mechanisms. This framework further incorporates collaborative feature transformations using Gramian angular fields and Markov transition fields, as well as residual network models with channel and spatial attention mechanisms. Experimental validation using a landing gear lower strut lock mechanism demonstrates high accuracy, exceeding 0.9566 at a 95% confidence level, thus validating the feasibility of this approach for fault diagnosis in aircraft complex motion mechanisms. Ablation experiments confirm the effectiveness of each component, highlighting the overall superiority of the proposed framework.

To effectively detect and localize vibration faults such as clamp loosening in aero engine accessory pipes, this paper proposes a novel diagnosis method based on dynamic responses of accessory pipes and transmissibility functions. The dynamic model and equation of the faulty pipe are established based on the principle of structural similarity and dynamic similarity, where the impact of the fault is simulated as an additional nonlinear load acting on related position of the pipe. Through the analysis of structural dynamic characteristics, the relationship between dynamic responses and fault occurrence and location is derived and analyzed, a fault diagnosis method based on transmissibility function is then proposed, and the operating process of the method is summarized as well. The accuracy and practicality of the proposed fault diagnosis method are verified through multiple experimental examples. Theoretical analysis and testing results show that the diagnostic method proposed in this paper can accurately detect and localize the existence and position of the clamp loosening fault. Meanwhile, this diagnostic method is suitable for single and multiple clamp loosening faults in aero engine accessory pipes.

As a core component of an aero-engine, the structural integrity of a blade directly determines the engine’s performance and flight safety. Under extreme working conditions such as high temperature, high pressure, and high-speed rotation, blades are prone to generating micro-cracks under the action of complex stress fields. Once cracks propagate and cause blade fracture, they will trigger chain damage, posing significant safety hazards. Based on the damage tolerance concept, the critical duration during which a blade can still operate safely after crack initiation is defined as the remaining useful life (RUL).To address this, this study proposes a mechanism-data dual-driven RUL prediction method integrating the Paris crack propagation law and physics-informed neural networks (PINN). By constructing a loss function that incorporates physical constraints, this method regularizes and constrains the gradients of the neural network. It enables inverse identification of crack propagation parameters while effectively improving the model’s prediction accuracy under limited monitoring data. For aero-engine blades and CT (compact tension) specimens, compared with traditional physical models and data-driven methods, the proposed method dynamically updates characteristic parameters to adapt to system changes, significantly reducing prediction errors under limited sample conditions. Additionally, the PINN model developed in this study features lightweight architecture and fast inference capabilities, meeting the requirements of online monitoring and predictive maintenance. This method provides a new technical pathway for health management and intelligent operation and maintenance of aero-engines.

Traditional bearing fault diagnosis methods often suffer from low accuracy and weak model generalization under varying working conditions due to diverse sample distributions, scarcity of fault samples, and limited feature extraction capabilities of some few-shot learning algorithms. To address these challenges, this paper proposes a novel method for variable condition bearing fault diagnosis that combines a squeeze-and-excitation residual network (SE-ResNet) with meta-transfer learning (MTL). One-dimensional bearing vibration signals collected under different working conditions are converted into time-frequency images using continuous wavelet transform (CWT), thereby transforming the bearing fault diagnosis task into an image recognition problem. A squeeze-and-excitation (SE) attention mechanism is introduced to construct an SE-ResNet backbone network model. This focuses on more effective feature channels, thereby enhancing feature extraction and representation capabilities. Leveraging the advantages of transfer learning (which provides robust initial deep network parameters) and meta-learning (which enables rapid adaptation), the model undergoes sequential pre-training and meta-transfer training. This process yields a high-precision meta-transfer network that can be fine-tuned with only a small number of samples, ultimately achieving accurate bearing fault diagnosis under variable working conditions. The proposed method is validated using two benchmark datasets and a bearing fault simulation test bench developed in the laboratory. Comparative analysis with other methods demonstrates that the proposed method exhibits higher recognition accuracy and superior generalization performance for bearing fault diagnosis under both few-shot and variable working conditions.

Bearings are critical bogie components, making their early fault detection particularly important. This paper proposes an early fault detection method for bearings based on a relation network (RN). A health status detection relation network model is designed to effectively extract bearing condition features and measure the nonlinear distance between these features. In the offline modeling phase, normal samples from the bearing are collected for training, allowing the model to learn the nonlinear distances among the healthy state sample features. During the online monitoring phase, samples from the current operating state are acquired, and a relation score is obtained as a health indicator for the bearing condition. The 3σ criterion is then applied to determine the health indicator threshold for detecting the bearing health status and identifying faults promptly. Experiments were conducted on the XJTU-SY rolling bearing full-lifecycle dataset. Results show that, compared to methods like root mean square, kurtosis, and stacked autoencoders, the health indicator of the proposed method is more sensitive to early faults and exhibits better monotonicity and trend. Furthermore, in comparison with methods such as Isolation Forest, Support Vector Machine, and stacked autoencoders, the proposed method detects the first fault occurrence earlier, demonstrating considerable practical value.

To address the challenges of fully characterizing wheel information and accurately locating and quantifying wheel damage using trackside signals, this paper proposes a multi-sensor data fusion algorithm combined with an improved convolutional neural network (CNN) for wheel tread defect identification. A vehicle-track dynamics coupling model is established based on multi-body dynamics and finite element theory. By strategically arranging fewer sensors, multimodal features are extracted, and data fusion algorithms are optimized for parameters like wheel geometry and vehicle speed. An improved CNN model is then proposed, building upon both 1D-CNN and 2D-CNN architectures. Simutaneously, frequency domain features and image features are fused, leading to a new CNN algorithm model that incorporates these fusion features. Defect feature extraction is performed on the reconstructed signal, and the improved CNN, leveraging the fused data features, is used to achieve wheel damage identification. The effectiveness of the proposed method is validated using both simulation data and actual case studies, in conjunction with a proportional vehicle test rig. The damage identification performance of the proposed model is compared against CNN, BP neural network (BPNN), and support vector machine (SVM) under various signal test sets and data features. Results indicate that the proposed damage identification model can more effectively identify wheel tread defects, showing good consistency with measured results. Fusing data features from different dimensions can characterize defects under varying degrees of damage and significantly improve identification performance. This approach successfully addresses issues where trackside data alone cannot fully reconstruct wheel status, thereby providing crucial technical support for the online damage identification of wheel defects.

To address the limitations of deep neural networks in terms of interpretability and the inability of current interpretable networks to perform cross-domain diagnosis tasks, this paper proposes an interpretable triple feature extractor transfer network(ITFETN). For the interpretability challenge, a multi-layer sparse coding model is established, and its iterative solving algorithm is derived. By unrolling the fast iterative soft thresholding algorithm, an equivalent network form of the sparse coding model soving algorithm is obtained. This equivalent network then serves as a feature extractor, forming an interpretable algorithm-structure-equivalent network. To tackle the problem of cross-domain tranfer diagnosis, a triple feature extractor strategy is constructed. This strategy is designed to extract the shared features from the source and target domains, as well as their respective private features. Based on the concept of feature adversarial learning, a loss function for the transfer diagnosis task is designed for the effective training of ITFETN. This effectively extracts shared features with minimized distance between the source and target domains for cross-domain diagnosis, thereby achieving interpretable transfer diagnosis tasks. Experimental results demonstrate that ITFETN exhibits improved average accuracy and robustness in two case studies compared to benchmark methods. This confirms its effectiveness in achieving interpretable cross-domain diagnosis.

Under the background of industrial big data and intelligent manufacturing, data-driven intelligent fault diagnosis technology has become a crucial enabling technology. It ensures the safe and reliable operation of high-end equipment, facilitates equipment health management, and supports intelligent operation and maintenance. Existing intelligent fault diagnosis models often fail to simultaneously achieve superior diagnostic accuracy, strong noise immunity, high computational efficiency, and robust hyperparameter performance. To address these limitations, this paper proposes a novel spectral ensemble sparse representation classification model-driven super-robust intelligent diagnostic method. The proposed method designs a vibration data augmentation strategy based on cascade segmentation operators, aiming to enhance both the quantity and quality of vibration data samples. It utilizes the spectral features of vibration signals for dictionary atom design and constructs a spectral ensemble dictionary design strategy that incorporates spectral feature fusion. This improves the reconstruction capability of the spectral sparse representation dictionary. The method develops an intelligent recognition strategy based on the spectral sparse approximation error minimization criterion to achieve intelligent diagnosis of test samples health status. The proposed method is validated on a planetary gear transmission fault dataset. Results demonstrate that the intelligent diagnosis method can integrate the advantages of superior diagnostic accuracy, strong noise immunity, high computational efficiency, and robust hyperparameter selection. Its diagnosis results surpass existing advanced methods, showcasing significant application for data-driven intelligent fault diagnosis of industrial equipment.

Key components of industrial robots are prone to early-stage performance degradation under complex operating conditions, characterized by strongly non-stationary responses and significant heterogeneity across sensing channels. Traditional diagnostic methods struggle with robust and interpretable fusion of multi-source information, limiting their practical deployment. This paper proposes a dual-channel intelligent diagnostic method for robotic transmission mechanisms, integrating physics-driven sensitivity weighting and residual uncertainty compensation (RUC). Specifically, vibration and torque signals, representing structural response and driving excitation respectively, are selected due to their distinct temporal scales and complementary physical characteristics. A three-layer mapping (fault type-dynamic response characteristic-sensing channel) is constructed to quantify channel dominance for different fault modes. Then, a multi-scale sensitivity evaluation mechanism based on signal-to-noise ratio (SNR), modulation index (MI), and kurtosis guides adaptive weight allocation, while the RUC strategy enhances the expression of features from weakly dominant channels, improving fusion stability. Finally, a physically interpretable and lightweight diagnostic framework is established. Experiments conducted on a public gearbox dataset validate that the proposed method provides superior diagnostic accuracy, interpretability, and deployment potential, demonstrating significant promise for physically consistent multi-source fusion diagnosis in robotic transmission systems.

Vibration and temperature signals from bearings contain rich fault characteristic information, making them crucial for condition monitoring and fault diagnosis of traction motor bearings in heavy-haul locomotives. This study establishes a thermo-vibration coupling dynamics model for heavy-haul locomotive traction motor bearings, based on vehicle-track coupled dynamics. The model considers the nonlinear normal contact and tangential friction effects between the roller, raceway, as well as their defect areas. The influence of raceway defects in motor bearings on the thermo-vibration coupling characteristics of the traction motor is investigated. Additionally, the mapping relationship between defect width and both vibration response and bearing temperature rise is constructed. Results indicate that when the defect width reaches 1 mm, distinct fault characteristic frequencies appear in the vibration signal spectrum. The root mean square frequency, a frequency-domain statistical indicator, shows an increasing trend across the entire defect width range, while the frequency standard deviation is more sensitive to early defects. Time-domain statistical indicators of vibration signals, such as root mean square (RMS) and kurtosis values, are relatively sensitive to outer raceway defects. Conversely, inner raceway defects lead to a rapid temperature increase in the traction motor bearing, which is prone to triggering temperature rise alarms.

New fault modes will continuously emerge in the long-term operation and service process of machinery equipment, which poses higher requirement of the continual learning and lifelong diagnosis capability for intelligent diagnostic models. Lifelong intelligent diagnosis technology driven by class-incremental learning provides new approaches to ensure the full lifecycle safe operation of high-end equipment. However, existing class-incremental learning methods cannot address the problem of efficient incremental transfer diagnosis under the circumstance of cross-operating conditions. To this end, this paper proposes a cross-domain lifelong intelligent diagnostic method driven by meta-class-incremental transfer learning. An enhanced feature extractor is developed via integrating deep residual networks with a convolutional block attention feature fusion module to achieve deep feature extraction and fusion across channel and spatial dimensions. A multi-level knowledge distillation strategy is constructed through combining feature-level and decision-level knowledge distillation mechanisms to effectively address catastrophic forgetting issues in incremental transfer diagnostic scenarios. A meta-class-incremental parameter learning mechanism is proposed by innovatively incorporating the idea of meta-learning into class-incremental learning framework, thus improving the model generalization ability for incremental transfer diagnosis. Experiment validations were conducted on subway train transmission system test rig. Results show that the proposed method achieves an average diagnostic accuracy of 94.96% and an average forgetting rate of 3.85% across different incremental transfer diagnostic scenarios, and outperforms state-of-the-art class-incremental learning methods, offering insights for achieving lifelong intelligent fault diagnosis in full lifecycle health management of high-end equipment.

In planetary needle roller bearings, slip ratio and raceway surface quality mutually influence each other. The propogation of early localized raceway defects can significantly impact the cage slip and service performance. To address this problem, a slip dynamic model for planetary needle roller bearings is established, specifically considering localized raceway defects. This model is used to analyze the influence of localized defect size on the slip rate of the bearing. Based on the morphological characteristics of raceway localized defects, the displacement and friction coefficient of the rolling element passing through the localized defect area on the raceway surface are represented by piecewise functions. These piecewise displacement excitation and friction coefficient models are then integrated into the slip dynamic model to investigate the effects of localized defect width and depth on the cage slip. Results indicate that when localized defects occur on the inner and outer raceways, the friction coefficients at the corresponding positions significantly increase. The influence of inner and outer raceway localized defects on the cage slip varies considerably. As the localized defect width on both inner and outer raceways increases, the cage slip rises. However, the localized defect depth has little effect on the cage slip.

The Hilbert resonance demodulation (HRD) technique is widely applied in bearing fault diagnosis. Current studies typically evaluate noise reduction methods based on the frequency components of envelope signals. However, there is a lack of quantitative analysis regarding the generation process of these frequency components. This might lead to misattributing HRD-induced effects to noise reduction methods during evaluation. In engineering, the empirical selection of test conditions and analysis parameters often results in missed diagnoses due to improper choices. To address these issues, this study constructs a fault signal model for vibration responses of rolling bearing outer race local faults. Through quantitative analysis of the Hilbert resonance demodulation process, the mapping relationship between signal parameters and envelope amplitude spectra is revealed. By investigating the influence of system physical parameters—such as natural frequency, damping ratio, and rotational speed—on signal characteristics, the correspondence between these parameters and the envelope amplitude spectrum distribution is established. This provides a clearer theoretical foundation for HRD applications. Simulated vibration signals processed by HRD demonstrate consistency with theoretical derivations, and experimental validation confirms the analytical conclusions based on the HRD demodulation mechanism.

Planet bearing kinematics involve spinning-revolution coupling, resulting in complex and weak fault vibration signals that pose a tremendous challenge to fault diagnosis. Under time-varying speed conditions, the frequency characteristics of gear mesh vibrations overlap with those of planet bearing faults, severely interfering with their fault diagnosis. To address this issue, this paper proposes an order-frequency spectral correlation analysis method for non-stationary signals. The method removes the time-varying low-frequency amplitude envelope of the vibration signal and performs angular domain resampling to stabilize the order characteristics of gear components. Discrete random separation in the order domain is applied to eliminate gear vibration while retaining residual random components. These random components are inverse angular domain resampled to restore the original amplitude envelope, and planet bearing fault features are extracted through their order-frequency spectral correlation or coherence. This method enhances planet bearing fault features and improves the diagnosis capability under time-varying speed conditions. The principle of the method is demonstrated through numerical simulation analysis. Its performance is validated experimentally by successfully diagnosing localized fault on the inner and outer race and rolling elements of planet bearings.

The optimized layout of sensor measuring points is of significant importance for the condition monitoring, fault diagnosis and health management of mechanical equipment. Focusing on gear transmission systems, this paper investigates a method for optimizing vibration sensor placement based on the effective independent analysis of gearbox modes and frequency response functions. This method uses the fused signals from the optimized measuring points to monitor the vibration of the gear transmission system. A dynamic model of the gearbox is established, and modal analysis is performed to obtain the mode shapes of the gearbox. These mode shapes are utilized for effective independent analysis to determine an initial sensor layout. Harmonic response analysis is conducted to obtain the frequency response functions of these initial measuring points relative to the main bearing seats. The measuring points are further optimized using principal component analysis and effective independent analysis. The optimized measuring points most sensitive to gear fault excitation are selected. Spectral weighted fusion is performed based on the importance weights of the optimized measuring points. The fused spectrum then enables vibration monitoring of the gear transmission. Analysis of measured data from a bevel gear transmission test rig demonstrates that, compared to the fused spectra from arbitrarily selected measuring point groups, the fused spectrum from the optimized measuring point group exhibits a larger overall response amplitude. This indicates a superior monitoring effect for the oprational status of gear transmission.

Maximum correlated kurtosis deconvolution (MCKD), which uses correlated kurtosis as its deconvolution target, effectively extracts both periodic and impulsive features of mechanical faults. This is a widely used method for solving rolling bearing fault diagnosis problems. However, the performance of MCKD heavily relies on accurate prior fault period information. Existing solution often only focus on period estimation during the iterative process, making them ineffective under low signal-to-noise ration (SNR) conditions. To address this limitation, a period-refined maximum corrlated kurtosis deconvolution (PRMCKD) method is proposed. This approach refines the iteration period using time synchronous averaging (TSA) for reconolution, enabling accurate extraction of subtle bearing fault features even in strong noise environments. The method operates by first utilizing a filter bank for preliminary localization of the resonance frequency band, thus defining the correct deconvolution direction. With correlated kurtosis as the objective function, and leveraging the period information refined by TSA technology, the optimal filter coefficients are iteratively solved. Rolling bearing fault localization is achieved through the fault features present in the filtered signal. Simulation and experimental analysis results demonstrate that the proposed PRMCKD method offers significant advantages over traditional deconvolution methods for extracting weak fault features in rolling bearings.

Motor current monitoring systems have garnered significant attention due to their non-invasive and cost-effective advantages. However, conventional current spectrum analysis is susceptible to inherent harmonic and installation errors, and the high amplitude of the fundamental frequency can obscure fault characteristics. To reveal the frequency modulation patterns in motor current caused by reduced meshing stiffness due to gear faults, a motor current model incorporating faulty gear meshing stiffness is established, and its instantaneous frequency expression is derived. Addressing the limitation of traditional time-frequency analysis methods, which often suffer from low instantaneous frequency estimation accuracy, this paper proposes an instantaneous frequency polar view method based on high-order synchrosqueezing transform (HSST) for extracting gear fault features. This method intuitively demonstrates gearbox faults by detecting frequency modulation characteristics that are synchronized with the meshing period of faulty teeth. The instantaneous frequency polar view effectively avoids interference from inherent harmonics and fundamental frequency, offering a unique representation of gear fault characteristics. Analysis of motor current signals from an RV gearbox test rig validates the accuracy of the proposed motor current model and the distribution patterns of fault characteristics. It also confirms the effectiveness of the instantaneous frequency polar view method based on HSST for gearbox faults diagnosis.

Gear systems in precision machinery and aerospace applications are subjected to complex vibration problems due to mass eccentricity, time-varying backlash, and dynamic meshing parameter variations. A nonlinear dynamic model with six degrees of freedom is established, incorporating time-varying meshing stiffness, derived using the potential energy method and mass eccentricity. The Runge-Kutta method is employed to solve the system response under varying eccentricities and rotational speeds. Time-domain and frequency-domain analyses, phase portraits, and Poincaré maps are used to investigate the dynamic characteristics. The results indicate that mass eccentricity significantly influences system behavior, leading to the evolution from single-period to multi-period motions (e.g., 20-period cycles), and aggravates bifurcation and oscillation phenomena. The findings provide theoretical support for structural optimization and vibration control of gear transmission systems.

Gear transmission systems are critical for power delivery in major equipment, yet they are susceptible to diverse and unpredictable to diverse and unpredictable faults, compromising operational safety and service reliability. In practical engineering applications, the absence of fault samples and unknown fault locations pose significant challenges. To address these issues, this paper proposes an unsupervised hypersphere-based fault localization (UHFL) method empowered by mode component energy features. This method extracts mode component energy features that are not only related to the fault mechanisms of transmission components but also possess clear fault localization interpretability. These features then enable both unsupervised anomaly detection and interpretable fault localization. Specifically, an unsupervised data description model is constructed using the proposed features. An attribution explanation method is introduced to quantify the contribution of each feature to the anomaly detection result, thereby achieving interpretable fault localization of gear transmission system components under conditions lacking fault samples. The proposed method is validated through single-fault and compound-fault localization experiments conducted on a helicopter main reducer planetary stage test bench and an armored vehicle transmission system test bench. Experimental results demonstrate that the proposed UHFL method can accurately localize faults in transmission components without requiring any fault sample training. This method offers an effective solution for fault localization in gear transmission systems under data-scarce conditions, showcasing valuable engineering promotion potential and application prospects.

Planetary gear transmission systems are extensively used in industrial applications. Due to their compact and complex configurations, mechanical components are prone to failure during long-term operation. Compared to fixed-axis gear systems, planetary gear systems exhibit multiple excitation sources and time-varying signal transmission paths stemming from their intricate structural and kinematic characteristics. Consequently, condition monitoring techniques based on fixed vibration measurement points face significant challenges in compound fault diagnosis, especially when a planet gear fault is coupled with a bearing fault. To address these issues, this study proposes a non-contact torsional vibration monitoring and residual vibration analysis method utilizing laser doppler vibrometry (LDV). The laser beam is positioned on the low-speed shaft surface to directly acquire torsional vibration information from the gear system. To mitigate the impact of measurement noise on fault feature extraction, a hybrid denoising strategy combining cepstrum-based soft-threshold editing and median filtering is developed to suppress pseudo-vibration artifacts and random impulse noise, respectively. For different types of compound faults, a progressive residual vibration decomposition framework is established. This framework systematically peels off residual broadband response and residual meshing sideband components from the torsional vibration signal. Specifically, optimized filtering is applied to the broadband response obtained via cepstrum short-pass to extract the second-order cyclostationary features of bearing faults. Concurrently, a phase self-demodulation-based order domain resampling method is proposed to highlight gear fault features by reconstructing meshing sideband residual signals of different orders. Experiments involving tooth spalling on planetary gears and raceway spalling on the sun gear bearing demonstrate that the proposed method can effectively achieve non-contact compound fault diagnosis for planetary gear systems. Compared to conventional flexible synchronous averaging and accelerometer-based methods, the proposed approach exhibits superior performance in early-stage planet gear fault detection under varying speeds.