Vehicle-implemented magnetic flux leakage (MFL) detection of rail head surface defects is often affected by complex multi-source interference, while dedicated methods for complete defect-related signal extraction are still lacking. An improved Savitzky-Golay (SG) filter with local adaptive adjustment and amplitude compensation is first employed to correct baseline drift while reducing distortion of defect-related features. A joint morphological and Wigner-Ville distribution (WVD) masking strategy is then employed for precise single-channel defect signal extraction, followed by inter-channel concatenation based on temporal overlap and channel continuity to reconstruct complete multi-channel defect responses. The proposed method is demonstrated using an artificially defective rail with defect depths ranging from 0.5 to 8 mm. Results obtained at different detection speeds show detection and extraction rates of 50% for defects with depths of 0.5 mm <= d < 1 mm and 100% for defects with depths of 1 mm <= d <= 8 mm. The method is validated using natural rail head surface defect data collected under vehicle-implemented MFL detection conditions. The extracted defect-related signals are consistent with manual inspection and rechecking results in terms of defect quantity, mileage position, and channel correspondence, confirming accurate and complete signal extraction by the proposed method.

Quantitative assessment of White Etching Layer (WEL) thickness in rails is essential for proactive railway maintenance. We develop an inversion model that correlates Magnetic Flux Leakage (MFL) signal features with WEL thickness by analyzing the physical relationship between martensite-induced permeability variations and characteristic bipolar MFL signals. To bridge the simulation-experiment gap, we implement a transfer learning strategy combining simulation-based pre-training with fine-tuning on limited experimental data. The method constructs a parametric sample library via finite element simulations, extracts key MFL features (peak-to-peak value, peak distance, waveform asymmetry), and fine-tunes the model using metallographically validated field samples. Field tests demonstrate a mean absolute error of 19.0 mu m and an 87.5% detection rate for WEL thickness prediction. This work provides a practical solution for quantitative microstructure assessment in nondestructive testing applications with scarce labeled data.

Magnetic flux leakage (MFL) technology is widely applied to detecting rail head surface damage. However, the leakage magnetic field generated by a surface damage is always so weak that the MFL signal is susceptible to interference, particularly when the probe vibrates vertically. This paper focuses on the coupling effects of magnetic yoke and sensor vertical vibrations on MFL signal integrity. Based on the vertical distribution characteristics of a damage leakage magnetic field, a sled-type bearing mechanism is designed. Furthermore, a vertical differential sensor array structure is developed to reduce vertical vibration interference, with an analysis of the optimal spacing between upper and lower sensor rows. An experimental platform is established to validate the method using both artificial and natural rail head surface damages. Results demonstrated that the proposed probe effectively suppresses vibration interference.

To overcome the destructive nature and online-monitoring limitations of traditional hardness testing, this paper proposes a non-destructive hardness evaluation method for ferromagnetic materials by fusing Magnetic Incremental Permeability (MIP) and Magnetic Barkhausen Noise (MBN). A dual-modal detection system was constructed for synchronous excitation and acquisition of MIP and MBN signals. Multi-dimensional electromagnetic features were extracted from the time domain, frequency domain, time-frequency domain, and butterfly plot geometry to characterize material hardness. To reduce feature redundancy, a Dynamic Weighted Recursive Feature Elimination (DW-RFE) algorithm was developed using multi-model ensemble learning, dynamic weight allocation, and stability evaluation. Experimental results show that the DW-RFE-based model achieves average relative errors of 2.30% on the complete dataset and 2.87% on the incomplete dataset, demonstrating good accuracy and generalisation ability for non-destructive hardness evaluation.

The rapid and precise defect detection of rail surface defects is critical for railway operational safety. Conventional inspection methods, however, are often labor-intensive, resource-consuming, and vulnerable to noise. To address these challenges, this paper proposes a real-time defect segmentation and detection model called WT-YOLO12. Firstly, this paper introduces the MSCA mechanism to capture long-range contextual information, compensating for the limited receptive field of the YOLO12 backbone network. Secondly, this paper enhances high-frequency defect features using wavelet transforms, significantly alleviating the bottleneck in fine-grained track surface defect segmentation with YOLO12. Extensive experiments on the Type I and Type II track surface defect datasets validate the effectiveness of the proposed segmentation algorithm. On the Type II RSDD dataset, WT-YOLO12 improves Mask mAP by 11% over the baseline YOLO12s, while maintaining a processing speed of 71.2 FPS, which satisfies real-time application requirements. Furthermore, comparative experiments on the Railway dataset demonstrate the model's strong generalization, yielding a 3.9% increase in mean Average Precision (mAP) over the baseline YOLO12s.

With an increasing amount of space debris accumulating in Earth's orbit, spacecraft face a growing risk of debris-induced damage that may lead to leakage incidents. Acoustic methods demonstrate the potential for detecting and localizing spacecraft leaks in orbit. This study employs computational fluid dynamics to model spacecraft leakage scenarios, systematically investigating the jet flow characteristics generated by leakage holes with varying diameters, geometries, and spatial locations. Acoustic data under different experimental conditions are acquired via a vacuum leakage sound field testing system and analyzed in combination with the flow field characteristics obtained from the simulation. The results demonstrate that under one atmospheric pressure differential, leakage jets in atmospheric environments remain subsonic, whereas vacuum leakage jets achieve supersonic velocities. In the acoustic field distribution cloud map, there exists a quiet zone with a relatively lower signal amplitude. The obstruction has a significant impact on the amplitude of the vacuum leakage signal. Therefore, the angle of the leakage detection device needs to be adjusted to minimize the effect of the obstruction. The simulation and experimental results obtained in this study provide theoretical foundations and empirical data to support external leakage detection for on-orbit spacecraft, which directly guide the optimal deployment of detection devices. The results also provide technical support for improving the efficiency of leakage signal processing and advancing research on high-precision leakage localization algorithms.

In this paper, we address the challenge of detecting bottom cracks in railway tracks by proposing a signal enhancement technique designed to boost the signal-to-noise ratio of guided wave signals, thereby optimizing sensor performance. Utilizing Barker codes and binary phase shift keying technology, we modulated the original excitation signal to achieve spread spectrum processing and the subsequent despreading of received signals. This approach aims to overcome hardware and sensor limitations, enhancing the operational efficiency of the sensors used in ultrasonic guided wave detection systems. An experimental platform was established to test this technique on railway tracks with artificial bottom cracks of varying sizes. The results showed that the enhanced signals can travel further and detect cracks with a higher sensitivity than the original signals. This not only validates the effectiveness of our method in improving sensor capabilities but also supports the application of ultrasonic guided wave detection for bottom cracks in railway tracks.

To achieve non-destructive hardness testing of ferromagnetic materials, this study develops a bimodal detection system capable of synchronously exciting and acquiring magnetic incremental permeability (MIP) and magnetic Barkhausen noise (MBN) signals, from which a 38-dimensional multi-domain feature set is extracted, encompassing time-domain, frequency-domain, and butterfly diagram geometric features. To address the redundancy and noise problems inherent to high-dimensional features under small-sample conditions, a Dynamic Fusion Adaptive Recursive Feature Elimination (DFARFE) algorithm is proposed, which integrates a multi-criteria weighted evaluation framework, Bootstrap resampling-based dynamic weight allocation, and leave-one-out cross-validation. The resulting feature subset is highly correlated with hardness while exhibiting excellent interference resistance. Experimental results demonstrate that the multi-parameter regression model constructed upon this subset attains an average relative error of only 0.78% on the test set. Compared with conventional methods, the feature subset selected by the proposed approach possesses stronger interference resistance and generalization capability, thereby significantly enhancing both the accuracy and robustness of hardness evaluation for ferromagnetic materials.

Rail internal defects pose significant threats to railway operational safety, and ultrasonic B-scan imaging provides an effective nondestructive testing approach for visualizing subsurface rail conditions. However, manual interpretation of B-scan images is time-consuming and depends heavily on operator experience. To overcome this problem, a Multi-scale Dense Attention Network (MSDA-Net) is proposed for automatic classification of rail ultrasonic B-scan images. The proposed network integrates dilated convolution, multi-scale dense feature extraction, and attention-based feature recalibration to capture defect features at different spatial scales while enhancing discriminative defect-related regions. A dataset containing 536 ultrasonic B-scan images is constructed for model training and evaluation, and the B-scan images are collected from multiple railway sections. Experimental results show that MSDA-Net achieved an accuracy of 0.909, with a macro Precision of 0.867, macro Recall of 0.845, and macro F1-score of 0.848. Comparative and ablation experiments further demonstrate the effectiveness of the proposed network structure for rail defect classification.

With the oil and gas field development entering a new intelligent stage， accurate and efficient condition diagnosis of pumping wells has become a core measure for ensuring stable production， optimizing maintenance decisions， and achieving cost reduction and efficiency improvement. Traditional diagnosis methods rely on the matching of personnel experience with fixed rules， making them suffer from limitations such as low diagnosis efficiency， strong subjectivity， and poor scalability. Although machine learning algorithms represented by support vector machine （SVM） and decision tree （DT） have been increasingly applied in intelligent diagnosis research， existing models still face common engineering bottlenecks： （1） insufficient model interpretability， making it difficult to gain the trust of domain experts； （2） limited generalization ability， being sensitive to changes in data distribution and class imbalance issues； and （3） fragmented technical processes， relying on manual feature engineering and complex tuning， while struggling to integrate deeply with production management systems. These constrain large-scale application of the models. This paper constructs an intelligent dynamometer card diagnosis system for pumping wells with eXtreme Gradient Boosting （XGBoost） as its core， covering a complete workflow from data acquisition and feature extraction to intelligent diagnosis and visualization. This system adopts a B/S architecture and a “1+N” distributed design， enabling real-time integration and unified management of multi-source heterogeneous data. By introducing XGBoost as the core classifier and combining it with multi-dimensional feature extraction techniques and the shapley additive explanation （SHAP） interpretability analysis framework， this system enhances classification accuracy while improving the transparency of the diagnosis process and expert credibility. Field test results show that the system achieves a diagnosis accuracy of over 90% for seven typical working conditions， reduces single⁃well diagnosis time from 30 minutes to less than 2 minutes， and attains an early warning alignment rate of 85.7%. Compared with traditional diagnosis methods， the proposed system significantly improves diagnosis efficiency and result interpretability while ensuring high diagnosis accuracy， providing a promotable technical solution for the intelligent diagnosis of pumping wells.

The detection of weak defects in eddy current pulsed thermography is hampered by nonuniform heating and thermal diffusion, which cause low signal-to-noise ratio (SNR). While conventional principal component thermography (PCT) is widely used for data reduction, its nature as a “blind” statistical tool often fails to separate these subtle defect signatures from the dominant background noise. To address this, we adapt conventional PCT by introducing a saliency-based spatial weighting strategy. The core innovation is a weighting scheme where a spatial saliency map, derived from time-integrated thermal energy, is nonlinearly enhanced via a gamma transformation. Integrating this high-contrast weight map into the PCT algorithm transforms it from a blind statistical tool into a targeted, guided analysis. Validated on six diverse metallic specimens, the proposed spatially weighted PCT (SW-PCT) significantly outperforms traditional PCT and other representative methods in SNR, especially for enhancing weak defects. An optimized vectorized implementation ensures high computational efficiency, establishing SW-PCT as a robust solution that balances detection accuracy and processing speed for industrial nondestructive testing applications.

The turnout switch rail is a type of variable-cross-section rail, and its irregular structural characteristics result in complex ultrasonic guided wave detection signals. When employing the reflection method to detect cracks in the rail base, the amplitude of the echo signal cannot represent the size of the crack. To quantitatively analyze the crack signals, a method that combines deep learning and ultrasonic guided wave technology is employed to quantitatively assess the depth of cracks in the rail base of the turnout switch rail. By applying wavelet transform to obtain wavelet time-frequency diagrams, four deep learning models-GoogLeNet, Mobilenetv1, Mobilenetv2, and Mobilenetv3-are utilized to classify the depth of cracks in the rail base, and the performance of these models is assessed using experimental data. The experimental results show that the combination of the Mobilenetv3 deep learning model and ultrasonic guided wave technology achieves a 95% recognition accuracy for the quantitative detection of cracks in the rail base of turnout switch rails. This research work provides a foundation for the feasibility and reliability of combining deep learning models with ultrasonic guided wave technology for the quantitative detection of crack depths in turnout switch rails.

This study addresses the critical challenge of insufficient classification accuracy for different defect signals in rail magnetic flux leakage (MFL) detection by proposing an enhanced intelligent classification framework based on particle swarm optimized radial basis function neural network (PSO-RBF). Three key innovations drive this research: (1) A dynamic PSO algorithm incorporating adaptive learning factors and nonlinear inertia weight for precise RBF parameter optimization; (2) A hierarchical feature processing strategy combining mutual information selection with correlation-based dimensionality reduction; (3) Adaptive model architecture adjustment for small-sample scenarios. Experimental validation shows breakthrough performance: 87.5% accuracy on artificial defects (17.5% absolute improvement over conventional RBF), with macro-F1 = 0.817 and MCC = 0.733. For real-world limited samples (100 sets), adaptive optimization achieved 80% accuracy while boosting minority class (“spalling”) F1-score by 0.25 with 50% false alarm reduction. The optimized PSO-RBF demonstrates superior capability in extracting MFL signal patterns, particularly for discriminating abrasions, spalling, indentations, and shelling defects, setting a new benchmark for industrial rail inspection.

Due to the rolling friction, it is easy to produce harmful scratches, cracks, blocks and damage on the rail surface. In order to detect and evaluate the damage accurately and efficiently, this paper proposed an evaluation method based on magnetic flux leakage (MFL) detection, where the defect threshold is calculated by using the adaptive threshold method and the impact of noise on the results reduced. With the study on the relationship between the depth, detection speed and signal peak-to-peak value of the manual sample, quantitative statistics is made on the depth of the defect. By calculating the average severity of the damage within a length of the rail, the condition of the section is assessed. During the study, the feasibility of the method is verified by finite element simulation analysis for the three-section damage of the main line of the high-speed railway. The test results show that the method can evaluate the rail surface damage quickly and effectively.

The existing techniques for measuring longitudinal stress in rails are primarily limited to the shallow surface area of the rail head, hindering accurate assessment of the overall stress state across the rail's cross section. To overcome this limitation, this study presents a novel methodology for rail stress measurement based on electromagnetic ultrasonic birefringence. This method utilizes an electromagnetic acoustic transducer (EMAT) to generate two types of shear waves with polarization directions parallel and perpendicular to the stress direction, which propagate through the entire rail cross section. By calibrating the acoustoelastic birefringence of these shear waves, a model for longitudinal stress measurement was established, and its performance was quantitatively analyzed using the probability of detection (POD). Test results indicate a strong linear relationship between acoustoelastic birefringence and longitudinal compressive stress in the rail. The absolute errors of stress measurements for all rails are within 3.60 MPa. The key POD parameters ( a(50) and a(90/95 )) for CHN60, CHN50, and CHN43 rails are -3.11/-7.46 MPa, -3.56/-9.79 MPa, and -5.31/-10.93 MPa, respectively. These data demonstrate that the electromagnetic ultrasonic birefringence stress measurement method provides high measurement accuracy for rail longitudinal stress, offering valuable insights for developing rail stress measurement equipment.

Eddy current pulsed thermography (ECPT) is an emerging nondestructive testing technique widely applied for surface defect detection in metallic materials. This article proposes a multistage synergistic optimization method to address key technical challenges, including background noise, low defect contrast, and complex thermal gradients. The proposed method comprises three stages: preprocessing, feature extraction, and postprocessing. In the preprocessing stage, the excitation peak frame is dynamically selected using image entropy difference, static background noise is suppressed through differential operations, and a geometric prior mask is generated via edge detection. The feature extraction stage uses the augmented Lagrangian multiplier (ALM) method to solve a differential image-based robust principal component analysis (DI-RPCA) model. This process reconstructs a sparse matrix that preserves the complete defect morphology while simultaneously eliminating high-frequency noise from the coil region. In the postprocessing stage, adaptive histogram equalization is applied to the sparse matrix to enhance indication edge details, and a local spatial consistency-based dual thresholding segmentation (LSC-DTS) method is designed for efficient anomaly-background separation. It combines adaptive thresholding based on image grayscale distribution with an area-based filtering mechanism for defect candidate regions. The experimental results on steel plates with natural fatigue cracks demonstrate that the proposed method offers significant advantages in quantitative detection performance over conventional methods.

Both microstructure and stress affect the structure and kinematic properties of magnetic domains. In fact, microstructural and stress variations often coexist. However, the coupling of microstructure and stress on magnetic domains is seldom considered in the evaluation of microstructural characteristics. In this investigation, Magnetic incremental permeability (MIP) and magnetic Barkhausen noise (MBN) techniques are used to study the coupling effect of characteristic microstructure and stress on the reversible and irreversible motions of magnetic domains, and the quantitative relationship between microstructure and magnetic domain characteristics is established. Considering the coupling effect of microstructure and stress on magnetic domains, a patterned characterization method of microstructure and stress is innovatively proposed. Pattern recognition based on the Multi-layer Perceptron (MLP) model is realized for microstructure and stress with an accuracy rate higher than 97%. The results show that the pattern recognition accuracy of magnetic domain features and micro-magnetic features simultaneously as input parameters is higher than that of micro-magnetic features alone as input parameters.

In this paper, Monte Carlo simulations are performed based on the two-dimensional Ising model with the objective of matching the simulated magnetic Barkhausen noise (MBN) signals with the measured MBN signals obtained from empirical research on bearing steel of different hardness levels. Firstly, the methods for obtaining simulated MBN signals based on the Ising model are studied. This paper suggests that simulated MBN signals obtained by applying a digital filter to the simulated magnetization curve, both in the time domain and frequency spectrum, are closer to the actual measured signals. Secondly, the influencing factors of the two-dimensional Ising model are studied, including lattice size ( N ), temperature ( T ), neighbor interaction ( J ), external magnetic field ( H ( t )), number of simulation points per period ( P_sim ) and Monte Carlo step ( MCS ). Furthermore, the simulated MBN signals and their feature diagrams under different temperatures and neighbor interactions are plotted. Finally, a method is proposed to match the simulated MBN signals with the actual measured MBN signals using scaling and shifting, reducing the relative error between the simulated and measured MBN signal features to within 7

Our research introduces a novel stochastic resonance (SR) model featuring a single potential well and develops a dedicated detection system designed to address the challenging problem of detecting impact signals within a highly noisy background. We begin by examining the limitations of conventional metrics, such as the cross-correlation coefficient and kurtosis index, in identifying nonperiodic impact signals, and subsequently introduce an improved metric. By harnessing parameter-adjusted SR, this innovative potential well model and metric is integrated to formulate an adaptive detection method for nonperiodic impact signals. This method automatically adjusts system parameters in response to the input signal. Subsequently, numerical simulations of the system is conducted so as to perform a comparative analysis with experimental results obtained from both asymmetric single potential well and periodic potential systems. Our findings conclusively demonstrate the enhanced effectiveness of our proposed method in detecting impact signals within a high-noise environment. Furthermore, the method provides more accurate estimates of both the intensity and precise location of the input impact signal from the output results.