Orthotropic steel decks (OSDs) have found extensive application in long span bridges, owing to their excellent mechanical performance and constructional convenience. However, the presence of fatigue cracks remains a persistent threat to long-term structural safety. It is thus imperative to develop highly efficient and accurate methods for fatigue crack identification and location. In this context, a single-sensor acoustic emission (AE) damage location framework is developed based on continuous wavelet transform (CWT) and convolutional neural network (CNN), which leverages the high sensitivity of AE and the strong learning ability of artificial intelligence. CWT helps to represent the all-around propagation characteristics of AE signals in the time–frequency domain. CNN model then extracts those location sensitive features from CWT maps and predicts the location of AE damage sources. A field pencil lead break (PLB) experiment was conducted on an OSD segment in the operating status of the long-span bridge to validate the performance of the developed method, where a series of CNN models, including AlexNet, VGG19, and ResNet18, were established and compared. Results show that the framework based on CWT and ResNet18 could achieve satisfactory location accuracy with a single AE sensor, demonstrating significant potential for large-scale engineering structures.
Steel wire ropes are critical load-bearing components in industrial conveying systems, and accurate quantification of broken-wire defects is essential for operational safety. However, in eddy current testing, lift-off disturbance significantly affects signals, causing nonlinear coupling between wire breakage severity and lift-off, which reduces evaluation accuracy and stability. To address this issue, a lift-off-resilient method for quantitative detection of broken-wire defects is proposed. First, a coaxial dual-receiver coil probe is designed to extract the minimum voltage values from the near and far receiver coils as dual features, thereby enhancing the system's capability to distinguish between lift-off variation and defect severity. Subsequently, a nonlinear relationship between voltage features and wire breakage severity is established based on finite element simulations under controlled lift-off conditions, and validated experimentally. Based on this, a forward model is constructed using natural neighbor interpolation, and a genetic algorithm is employed for global optimization, enabling the simultaneous inversion of wire breakage severity and lift-off value. The results demonstrate that the proposed method effectively mitigates the influence of lift-off variation on detection results, achieving a relative inversion error of less than 1
Automotive window film is a critical component for enhancing driving comfort and safety, as its infrared rejection rate directly affects the vehicle’s thermal insulation performance and energy consumption. Given the wide variety of brands and the considerable variation in product quality in the current market, there is an urgent need for a rapid, non-destructive, and brand-independent universal detection technology to regulate the market, protect consumer rights, and strengthen quality control during production. This study aims to develop a quantitative prediction model for the infrared rejection rate of automotive window films based on hyperspectral imaging technology. A total of 765 window film samples from five major brands were collected, and their visible-near-infrared hyperspectral data were acquired. To optimize spectral quality, six preprocessing methods-none, Savitzky-Golay smoothing, standard normal variate (SNV), baseline correction, detrending, and Normalization-were applied. Feature wavelength selection was performed using three algorithms: successive projections algorithm (SPA), competitive adaptive reweighted sampling (CARS), and uninformative variable elimination (UVE). The optimal number of principal components was determined using ten-fold cross-validation, the one-standard-error (1-SE) rule, and a threshold of CV R2 ≥ 0.80 to avoid overfitting. Subsequently, a partial least squares regression (PLSR) model was established for quantitative prediction of the infrared rejection rate. The results indicated that SNV and Normalization preprocessing yielded the best performance, significantly improving prediction accuracy and model stability. The CARS algorithm achieved the best balance between dimensionality reduction and prediction accuracy. The final optimal detection scheme was determined as Normalization preprocessing combined with CARS feature extraction and PLSR modeling. A total of 140 feature wavelengths were selected, reducing the dimensionality by 65.1
Aiming at the dynamic identification of fault states for automotive power steering pump rotors, this paper proposes a dual-threshold identification method based on multi-sensor acoustic emission(AE) signals and singular value decomposition(SVD) orthogonal modal centroid frequencies(CF). Firstly, rotors under normal(N), cracked(C), and reversed-blade(R) conditions are separately assembled into the same pump. Under identical operating conditions, 4-channel AE signals are collected near the inlet and outlet with a sampling frequency of 1 MHz. Secondly, single-cycle sub-signals are extracted from the acquired raw signals. After wavelet denoising(WD) and whitening processing, a feature matrix is constructed and decomposed by SVD to obtain four orthogonal modes. Subsequently, each orthogonal mode is processed using 3-layer wavelet packet decomposition(WPD), and the CF values of the first four nodes are calculated. Through preliminary screening based on the relative increase and dual-threshold verification combining the silhouette coefficient and coefficient of variation, third-layer WPD Node (3,1) is determined as the optimal sensitive feature node. Finally, discrimination thresholds are established based on the Shapiro–Wilk normality test and the 99
This study presents a real-time framework for diagnosing crack initiation in Aluminum 2024-T3 sharp notch samples by integrating acoustic emission (AE) processing with supervised machine learning. To overcome the technical limitations of traditional ex-situ analysis, a multi-threaded, concurrent processing architecture was developed in the C# programming language, enabling simultaneous data collection, feature extraction, and classification. A novel "initiation ratio" metric was implemented to identify the onset of cracking by quantifying the proportion of fracture-specific waveforms relative to other damage-related signals. Comparative analysis between contact-based piezoelectric (PZT) sensors and a non-contact laser vibrometer demonstrated that PZT sensors achieved 100
A feature extraction method combines anisotropic guided filtering with multimodal large-component adaptive segmentation. It addresses the local concealment features of aviation glass and enables accurate detection of small-target defects. The local concealment features of aviation glass are analyzed. Canny edge detection provides the edge foundation for the guided filtering. Local window size, regularization parameter, and anisotropic factor are dynamically adjusted to remove noise while preserving details. The adaptive threshold is dynamically determined according to the local gray-level characteristics of the image for binarization. An area threshold is set to filter and fuse potential target features, achieving accurate segmentation of local concealment feature images. The filtered image achieves a peak signal-to-noise ratio of 37.61, a structural similarity index of 0.91, and a feature extraction accuracy of 0.9962. The method effectively overcomes the interference of local concealment features in the precise segmentation and extraction of aviation glass feature images.
Detection of subsurface material loss in metallic pipelines is essential for ensuring structural integrity and operational safety yet remains challenging due to limited depth resolution and reduced sensitivity to deeper defects in conventional active infrared thermography techniques. In this work, a novel optimized 15-bit coded excitation sequence is proposed for the detection of material loss in mild steel pipelines using active infrared thermography. The excitation sequence is designed using a brute-force optimization framework to achieve favourable pulse-compression characteristics and an enhanced matched-filter response, enabling effective concentration of thermal energy within a narrow main lobe while suppressing sidelobe energy distribution. Experimental investigations are carried out on a mild steel pipeline specimen containing multiple artificial defects representing localized material loss. The acquired thermographic data are processed using frequency-domain phase, time-domain phase, and cross-correlation coefficient (CCC) based post-processing approaches to extract defect-related features and mitigate the effects of non-uniform heating. The performance of the proposed excitation scheme is quantitatively evaluated and compared with Linear Frequency Modulated Thermal Wave Imaging (LFMTWI) using spatial and temporal signal-to-noise ratio (SNR) as the primary figures of merit. The results demonstrate that the optimized 15-bit coded excitation, particularly when combined with CCC processing, provides significantly higher SNR, improved defect contrast, and superior detectability of deeper defects compared to LFMTWI and phase-based processing methods. These findings confirm that enhanced main-lobe energy concentration and effective sidelobe suppression are the key contributors to improved sensitivity, reliability, and depth-resolved inspection capability, establishing the proposed approach as a robust and practical solution for thermographic evaluation of material loss in mild steel pipeline structures.
Segmenting defects in X-ray digital radiography (DR) images of aluminum castings presents significant challenges due to the extremely low pixel coverage of minute defects, blurred boundaries, and weak contrast. The core issues lie in the fundamental trade-off between expanding the receptive field and preserving fine spatial details, as well as the background noise introduced by indiscriminate feature propagation during decoder fusion. To address these challenges, we propose a novel and improved U-Net architecture named DLE-UNet. First, we design a decoupled collaborative dual-path encoder. By structurally separating the contextual semantic stream from the spatial detail stream, it achieves synergistic optimization of large receptive fields and high-resolution details, thereby mitigating the loss of minute defect features in deep layers. Second, we introduce a cascaded boundary refinement decoder. In its initial stage, a Large-kernel Group Attention Gating module performs coarse feature screening to suppress background noise. Subsequently, a content-guided upsampling module conducts sub-pixel-level refinement of boundary features. Experiments on an aluminum casting defect dataset demonstrate that our method outperforms existing approaches on key metrics such as the F1-score (81.3
As the supporting structure connecting the upper and lower blade shell of wind turbine blades, the function of the blade web plate bears massive shear loads. The traditional sandwich structure of web plates, composed of glass fiber composite panels and polyvinyl chloride (PVC) foam cores, is prone to brittle fracture, exhibiting sudden catastrophic failure with limited energy absorption under long-term cyclic loading. Therefore, developing web plates with high toughness is a critical pathway to prevent brittle fracture during the operation of wind turbine blades. In this study, web plate based on polylactic acid carbon fiber (PLA + CF) material with a triply periodic minimal surfaces (TPMS) structure was designed. Static four-point bending test showed that, the web plates achieve an enhancement in toughness while maintaining sufficient strength compared with the traditional structure. To identify the damage modes of the novel web plate, acoustic emission damage identification method based on stress wave was applied in the static four-point bending test. The research results indicate that the novel web plate predominantly exhibits low frequency acoustic emission (AE) signals, corresponding to numerous matrix cracking and minor fiber/matrix delamination; in contrast, the traditional web plate is dominated by high-frequency signals, leading to through matrix cracking and significant fiber-matrix separation. The frequency difference reflects a transition from catastrophic brittle failure (in traditional web plate structures) to progressive damage (in the novel web plate structure).
Worldwide, about one billion prestressed concrete sleepers degenerate due to loads, chemical reactions, moisture or climatic conditions. Corrosion of the reinforcement or damage to the material structure can occur in the sleeper without any visible external signs. Therefore, non-destructive testing methods are required to reliably assess the condition. In this work, six prestressed concrete sleepers of type B70 are examined using the impact-echo method. Three different actuation methods are applied: a manually operated impact-hammer, an electromechanically mounted impact-cylinder and an air-coupled supersonic jet flow. All three actuation types show similar and consistent spectrograms. In intact rail sleepers, a dominant frequency above 10 kHz is always visible. In contrast, damaged sleepers show a dominant frequency significantly below 10 kHz. This simple threshold analysis can already be used for a condition assessment of the sleepers. Since the supersonic jet flow is continuous, an air-coupled inspection during travel at up to 80 km/h is theoretically possible. Since the nozzles are inexpensive, robust and virtually maintenance-free, a multi-channel actuation could be realised with little effort. In the future, the entire sleeper could be inspected non-destructively in a single high-speed pass, including a real-time condition assessment.
This work presents an experimental study that evaluates the influence of the frequency response of the Barkhausen Magnetic Noise (RMB) measurement system, as well as the intensity of the applied excitation magnetic field, on the sensitivity of the technique for measuring bending-induced stresses. The experiments were carried out on two samples made of different materials: AISI 1005 steel and AISI 430 steel. The MBN probe was equipped with five pickup coils with different numbers of turns, this being one of the variables that most strongly influences the frequency response of the sensor. The excitation magnetic field was applied at two intensity levels. The acquired signals were analyzed using different approaches, including frequency response profiles, spectrograms, MBN time-domain envelopes, number of voltage pulses in the MBN signal, and RMS values. The results indicate that coils with more than 2000 turns exhibit similar sensitivity levels. Below this threshold, sensitivity is reduced mainly due to a lower signal-to-noise ratio and, consequently, greater susceptibility to electromagnetic interference. Additionally, it was observed that, for AISI 430 steel, the variation in excitation magnetic field intensity between the two applied levels did not influence the sensitivity of the technique for stress measurement. In contrast, for AISI 1005 steel, the sensitivity was higher when a lower excitation magnetic field intensity was used. Finally, the results show how the number of turns installed in the MBN pickup coil significantly influences the performance of the measurement system and provide fundamental information that enables users of the technique to design sensors better suited for a given application.
Accurate detection of defects in steel ropes is a crucial prerequisite for monitoring their health status. However, the detection process is often affected by noise interference, resulting in a low signal-to-noise ratio (SNR) and making it difficult to detect small wire break defects. To address this issue, this study proposes a signal processing method for steel rope defect detection that effectively enhances the detection capability of small wire break defects. First, based on the spatial response characteristics between the defect and the sensor array, a strategy using relative peaks and waveform slopes is proposed to eliminate invalid channels containing strand and shaking noise, thereby improving the input signal quality before subsequent denoising. Second, a cascaded workflow integrating multi-stage filtering and nonlinear mapping is constructed to achieve seamless noise stripping and feature enhancement. Kalman filtering and dual median filtering are used for noise reduction to suppress strand and shaking noise, while gamma transformation enhances defect signal features and wavelet denoising extracts the optimized characteristics, facilitating the identification of small defect signals. The experimental results show that, compared with the six reported methods, the method achieves a relatively high SNR of 12.18 dB, the highest peak fidelity of 95.96
Fibre-reinforced polymer (FRP) composites are susceptible to impact damage during their service life. Vibration-based structural health monitoring (SHM) has emerged as a promising technique for reliably detecting and assessing such damage. This approach leverages the principle that damage induces local stiffness discontinuities, altering vibrational parameters such as natural frequencies, mode shapes, and damping. Research in this area has focused on using frequency shifts across multiple modes and applying inverse algorithms like Artificial Neural Networks (ANN) or Genetic Algorithms (GA) to predict the location and extent of damage. Training these machine learning algorithms (MLAs) requires extensive databases of frequency shifts corresponding to various damage parameters. To address the computational demands of database generation, numerical models offer an economical alternative. This study introduces a novel numerical methodology for evaluating frequency shifts in FRP composite laminates subjected to impact damage, facilitating machine learning-based non-destructive evaluation. The approach employs finite element modelling with restart analysis to efficiently simulate multiple damage scenarios without the need for complex parametric mappings. Simulations of composite plates under various impact energies demonstrate the method's effectiveness in characterizing cumulative damage through frequency shifts, particularly at higher energy levels, where it outperforms conventional techniques. The study also examines the influence of different damage criteria, including two-dimensional and three-dimensional Hashin models, on frequency shifts and mode shapes, showcasing the versatility of the proposed method. By streamlining the generation of extensive frequency shift databases, this approach significantly enhances the training of MLAs for inverse damage identification. The methodology has the potential to advance SHM techniques for composite structures across aerospace, automotive, and wind energy applications.
Barely visible impact damage (BVID) in carbon fiber-reinforced polymer (CFRP) laminates is difficult to assess reliably using a single nondestructive testing (NDT) technique. Therefore, this study proposes an in situ measurement strategy that combines pencil lead break acoustic emission (PLB-AE) with nonlinear ultrasonics to quantitatively assess impact damage in aluminum-lined CFRP laminates. Low-velocity impact tests were conducted at energies of 10, 20, 30, 45, and 60 J to induce BVID of different severity. After verifying waveform consistency using the skewness and kurtosis of the PLB-AE signals before and after impact, the damage index (DI) was developed using Mahalanobis distance (MD) based on PLB-AE characteristics (amplitude, energy, count, and rise time amplitude (RA)). The DI derived from PLB-AE characteristics can be correlated with damage and used to identify the transition in damage mode. Likewise, the acoustic nonlinearity coefficient obtained from nonlinear ultrasonic testing (NLUT) shows higher sensitivity to impact damage and reflects changes in damage modes. A pronounced shift in the trends of both the DI and the nonlinearity coefficient occurs around 30 J, indicating that a transition in damage mechanisms takes place near this impact energy in the tested laminated structures. The proposed PLB-NLUT method provides a reliable and practical strategy for evaluating BVID in composite structures.
Carbon fiber-reinforced polymers (CFRP) are particularly vulnerable to low-velocity impact loads that can lead to material degradation like cracks and delaminations. Due to the difficult visual detectability of these damages, ultrasonic testing has been established for many decades as the standard test method. Usually, the interpretation of the ultrasonic C-Scans is performed manually which requires a high level of operator qualification and experience. Therefore, this study compares different convolutional neural networks (CNN) for the classification of damaged and undamaged ultrasonic C-Scans of CFRP-specimen. For the required dataset, in-house ultrasonic C-Scans were compiled and categorized into three different classes: low-velocity impact damage, “non-impact” damage and undamaged CFRP-material. Before use in a CNN, the dataset was edited with an automated image processing to reduce inhomogeneities. The final dataset was first attached to various pretrained networks. Further it was used for the training of a self-generated model. The adapted and pretrained models outperformed the self-generated one with an accuracy of up to 92,5
To address the insufficient real-time capability and strong noise interference associated with conventional crack detection methods for oil and gas pipelines, this paper proposes a crack propagation pattern recognition method that integrates acoustic emission (AE) signals with intelligent algorithms. An improved complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) method is introduced to denoise AE signals. Wavelet packet energy spectrum analysis and kernel principal component analysis (KPCA) are then combined to extract the principal components of frequency-band energy, and multi-channel data are fused by a feature serial fusion strategy to construct comprehensive feature vectors. Furthermore, a convolutional neural network optimized by the sparrow search algorithm (SSA-CNN) is developed to accurately classify three stages of crack propagation: crack initiation, stable propagation, and unstable propagation. The experimental results show that the proposed method can effectively identify crack propagation patterns under both sufficient-sample and small-sample conditions. Compared with the conventional SVM and unoptimized CNN models, the recognition accuracy is significantly improved, reaching a maximum of 98.67
Gas cylinders are typical pressure vessels used for transporting and storing gases (including liquefied gases) and serve as essential infrastructure supporting a country’s economic and social development. Repeated filling and discharging in use may cause material fatigue, leading to plastic deformation and eventual failure. Permanent volumetric expansion is a key safety and quality indicator for cylinders. Traditional methods, whether direct expansion method or water jacket method, require dedicated equipment and sites, cannot realize in-situ inspection, and cannot capture localized plastic deformation. They are also inefficient and costly. To enable accurate in-situ measurement of permanent volumetric expansion, this paper proposes a method based on 3D point cloud modeling, combining laser scanning with theoretical analysis and experimental comparison. Using 3D laser scanning, point cloud data of five cylinders were collected before and after hydraulic testing. Triangular mesh models were constructed from the data, and permanent volumetric expansion was derived from volume change calculations. Experimental results compared with the water jacket method show that the proposed method uses fewer devices, is simple to operate, and yields small errors, confirming its feasibility. More importantly, unlike existing methods that only reflect overall deformation, this method accurately and intuitively reveals deformation at each local section of the cylinder. This is significant for precisely assessing cylinder safety. The method demonstrates clear advantages over similar techniques and offers engineering practicality. Future work may establish corresponding measurement and evaluation standards based on this method’s technical features.
A long in-service period of steel bridges can be achieved through a comprehensive approach that involves clarifying the corrosive environment, understanding the rate of deterioration caused by corrosion, and designing appropriate repairs. Therefore, monitoring systems to predict the deterioration of infrastructure structures and nondestructive testing methods to evaluate the degree of deterioration have been developed. This study introduces short-wavelength (SW)-infrared hyperspectral imaging as a method for detecting deteriorated areas in the top coats of anticorrosion systems with unknown infrared spectral properties. Deteriorated regions on in-service bridge coatings are identified by selecting specific wavelengths that exhibit significant differences in infrared reflection intensity within the SW-infrared spectrum. Moreover, measurements of SW-infrared hyperspectral imaging obtained using a maintenance vehicle were nearly as accurate as those from the motorized stage, indicating the imaging systemʼs potential to detect early-stage deterioration during routine maintenance. Additionally, the spectral lock-in method—which employs a lock-in process to extract synchronous components from a reference signal—is demonstrated as an effective technique for identifying top coat deterioration based on distinct variations in the spectral properties of the coatings.
Carbon Fibre Reinforced Polymer (CFRP), thanks to their excellent specific stiffness and strength compared to metals, composites are considered as suitable materials for liquid hydrogen LH2 tanks. However, the cryogenic operating conditions required for LH₂ storage (20 K) present significant challenges to the structural integrity of CFRP, especially due to matrix cracking induced by thermal cycling and mismatched thermal expansion coefficients between fibres and matrix. Such damage can compromise the leak tightness of a tank and lead to hydrogen release. This paper presents a method to automatically detect, count and measure matrix crack density, under static loading, using X-ray micro-computed tomography (µCT) in CFRP laminates through digital image processing. By considering the image dimensions, the method calculates crack density along the different plies of a specimen. The proposed method was used for inspecting the progression of transverse matrix cracks in cross-ply CFRP laminates. The results demonstrate that µCT is an effective non-destructive evaluation (NDE) technique for characterising early-stage matrix cracking in CFRP laminates. The proposed method achieved a minimum detectable crack density of 0.05 cracks/mm.
This paper establishes the fundamental form of a dielectric model based on Maxwell's equation of electromagnetic wave theory by studying the relationship between the dielectric constant and constituents according to the dielectric properties of the asphalt mixture. Based on an analysis of the impact of frequency and temperature on dielectric properties, the dielectric model of the asphalt mixture is constructed by taking frequency and temperature into full consideration. combined with the test data, the dielectric model is verified. The test results show that: the dielectric constant decreases linearly with the increase of frequency; the dielectric constant increases slightly with the increase of temperature; after considering the influence of temperature and frequency, the calculation accuracy of the dielectric constant is improved, and the model is more suitable for describing the dielectric properties of asphalt mixture.
