Carbon fibre reinforced polymer (CFRP) lap-bolted joints are widely used in aerospace and mechanical engineering, however, early bolt loosening is difficult to detect using conventional linear ultrasonic methods. Nonlinear ultrasonic techniques offer high sensitivity to variations at the contact interface, but their weak nonlinear responses are easily masked, limiting detection reliability. This study proposes a successive computing methodology (SCM) that integrates finite element (FE) modelling, ultrasonic analysis, and modal identification to enhance bolt loosening evaluation in CFRP lap-bolted joints using a novel out-of-plane dominated resonant nonlinear ultrasonic (OPR-NLU) method. A three-dimensional FE model is developed to simulate bolt loosening and ultrasonic wave propagation, and linear perturbation analysis is employed to distinguish OPR modes, in-plane resonance (IPR) modes, and non-resonant (NR) frequencies. Experimental validation is performed using a laser scanning vibrometer system. The results demonstrate that resonant excitation amplifies nonlinear ultrasonic responses, with OPR providing a stronger enhancement than IPR. The sensitivity for bolt loosening evaluation follows the order 'OPR > IPR > NR'. The proposed method significantly improves the sensitivity and stability of early bolt loosening evaluation, contributing to a reliable approach for structural health monitoring of CFRP lap-bolted joints.

Accurate quantification of multiple crack parameters during asphalt mixture self-healing remains an ill-posed inverse problem due to the weak nonlinear ultrasonic response and insufficient interpretability of existing data-driven methods. This study proposes a novel hybrid approach that integrates second harmonic generation (SHG) air-coupled ultrasonic technique and bidirectional long short-term memory (BLSTM) to achieve highprecision inversion of crack length and width. A hybrid dataset combining numerical simulation and experimental data is constructed to alleviate overfitting and improve generalization ability. Discrete wavelet transform (DWT) is employed for signal decomposition and noise suppression, enabling the BLSTM to capture bidirectional temporal dependencies critical for ultrasonic wave propagation analysis. Two interpretability methods based on decomposed sequences and SHAP values are innovatively introduced to reveal that over 95% of the prediction contribution originates from fundamental and second harmonic signals, consistent with the physical mechanism of contact acoustic nonlinearity. The proposed method achieves superior prediction performance with a mean absolute error (MAE) of 0.111 mm for crack length and 0.057 mm for crack width. Robustness tests under imbalanced datasets and noise interference further validate the reliability of the model. This work provides a physically interpretable and high-precision solution to the multi-parameter ill-posed inverse problem in nonlinear ultrasonic characterization, which can be extended to damage evaluation and healing monitoring of asphalt pavement. For model replication and additional resources, refer to: https://github.com/Dalongna/Multiple_crack_parameters.

Automated crack segmentation can support steel-bridge inspection, but patch-level overlap alone does not demonstrate full-image reliability. We developed an auditable pre-deployment evidence framework integrating training-manifest hard-example feedback, validation-only operating-point selection, locked full-size testing, protocol-separated domain commissioning, and calibration-bounded geometry diagnostics. On 106 locked full-size images, three end-to-end runs achieved global F1/Dice of 0.766 ± 0.014, with image-level sensitivity of 0.961, crack-free specificity of 0.900, and a false-positive burden of 224.2 px/Mpixel. Performance was comparable to a repeated ConvNeXt-like U-Net, indicating a framework rather than backbone contribution. External-domain results showed strong commissioning dependence: strict zero-shot F1/Dice was 0.338, validation-only threshold selection reached 0.397, and repeated 10-shot adaptation reached 0.794 ± 0.020. The framework defines evidence boundaries for human-supervised localization, adaptation need, and diagnostic-only geometry.

Timely and nondestructive monitoring of early-age mechanical evolution in reinforced concrete is essential for understanding stiffness development during curing. However, the continuous evolution of concrete properties and the coupled influence of reinforcement-related propagation effects make it challenging for conventional ultrasonic methods to capture the state transition of a monitored component. This study proposes a graph-learning-assisted framework based on multichannel nonlinear ultrasonic modulation (NLUM) features for continuous monitoring of early-age reinforced concrete. Ultrasonic signals were collected through three propagation paths, including concrete–concrete (C–C), steel–steel (S–S), and steel–concrete (S–C) paths, and transformed into frequency-domain amplitude features involving carrier components, modulation sidebands, pump-frequency-related components, and harmonic responses. A k-nearest-neighbor graph was constructed according to feature similarity among NLUM measurements to represent local correlations within the monitoring process, and a graph convolutional network (GCN) was developed to predict the dynamic elastic modulus. The proposed framework was evaluated using repeated measurements collected from a reinforced concrete component during 30 days of curing, with comparisons against graph attention network (GAT) and Transformer models. The results show that the GCN model achieves superior prediction performance under the C–C, S–S, and S–C propagation paths. The C–C path mainly reflects the stiffness evolution associated with concrete matrix development, while the S–S and S–C paths provide complementary ultrasonic responses influenced by reinforcement propagation and steel–concrete coupling effects. The proposed method demonstrates the feasibility of integrating NLUM features and graph-based learning for continuous nondestructive monitoring of early-age reinforced concrete components.

Deep learning with improved interpretability and generalization capability has been employed to detect defects in carbon fiber reinforced polymers (CFRPs). However, these models have not yet been effectively utilized to enhance nonlinear ultrasonic techniques. To address this limitation, a nonlinear ultrasonic-guided deep learning (NLU-DL) approach is proposed, in which linear and NLU features are balanced through a weighting function. The feature weights and the original data are fed into the network, thereby forming a DL model with dual-channel feature fusion. The proposed NLU-DL is verified and validated using a 3D finite element (FE) model and a laser scanning Doppler vibrometry (LSDV)-based system, with the datasets being numerically and experimentally collected, enabling a high-efficiency “one-transmitter-multiple-receiver” scheme. For more effective feature extraction, a principal component analysis-integrated convolutional neural network (NLU-PCA-CNN) is designed for 1D time-series signals, while a squeeze-and-excitation module-integrated convolutional neural network (NLU-SE-CNN) is constructed for 2D time–frequency images. The results demonstrate that the NLU-DL significantly outperforms nonlinear ultrasonics in imaging precision, and the NLU-SE-CNN achieves higher imaging contrast and stability for single-, irregular-, and multi-delaminations. Furthermore, compared with purely data-driven models, the NLU-DL offers enhanced interpretability, detection accuracy, and generalization capability, thereby presenting a novel intelligent solution for damage imaging in composites.

CFST structures are widely used in high-rise buildings, but debonding and void can easily occur between the steel tubular and concrete because of poor grouting and long-term loading effect. In recent years, the deep learning method of impact echo (IE) signals has provided an intelligent approach for detecting internal defects in various structures. However, most deep learning networks are data-driven and require annotated data of both intact and damaged areas of structures for the supervised training. The large amount of annotation is usually costly and the supervised models often have generalizability problem in real scenarios. To address the above issues, this paper proposes a method for subsurface defect detection of CFST structure based on a two-stage unsupervised learning algorithm of IE signals. This method only requires measuring the IE signal of the intact area, which is then input to a stacked autoencoder with physical constraints (PC-SAE) for the reconstruction. The damage-sensitive features between the original and reconstructed signals are the input data to one-class support vector machine (OCSVM), and the offset distance output by the OC-SVM is the damage assessment index. Finally, the two-stage trained model can be used to locate defects and assess the damage severity of CFST structures. The experiments show that the accuracy, recall, precision, and F1-score of the proposed method for subsurface defect detection in CFST reaches 99 %, 73.3 %, 100 %, and 84.6 %, respectively.

Nonlinear ultrasonic detection of carbon fibre reinforced polymers (CFRP) is recognised as effective for barely visible impact damage (BVID), but its sensitivity is often compromised by the masking of weak harmonic responses by noise. In this study, local defect resonance (LDR) spectroscopy is employed to improve BVID localisation under noise interference. A novel successive computing methodology (SCM) is proposed, integrating a validated 3D finite element (FE) model (calibrated via impact force-time history and delamination characteristics) with experimental scanning laser Doppler vibrometry. Critically, linear perturbation analysis performed directly on the impact-induced BVID model determines LDR frequency without requiring a priori defect location knowledge, simplified geometric assumptions, or cumbersome sweeping tests and signal reconstruction procedures. The identified LDR is validated using a proposed damage index, the maximum response amplitude (MRA) of harmonics, confirming that harmonic MRAs are amplified when the excitation frequency matches the LDR. The determined LDR frequency is then introduced into noisy nonlinear ultrasonic tests, resulting in significantly enhanced BVID imaging resolution despite noise disturbance. Quantitative assessment using the Michelson contrast metric confirms that high imaging contrast is maintained by LDR excitation even under significant noise interference, demonstrating its robustness for practical BVID detection.

Intelligent maintenance of roads and highways requires accurate deterioration evaluation and performance prediction of asphalt pavement. To this end, we develop a time series long short-term memory (LSTM) model to predict key performance indicators (PIs) of pavement, namely the international roughness index (IRI) and rutting depth (RD). Subsequently, we propose a comprehensive performance indicator for the pavement quality index (PQI), which leverages the highway performance assessment standard method, entropy weight method, and fuzzy comprehensive evaluation method. This indicator can evaluate the overall performance condition of the pavement. The data used for the model development and analysis are extracted from tests on two full-scale accelerated test tracks, called MnRoad and RIOHTrack. Six variables are used as predictors, including temperature, precipitation, total traffic volume, asphalt surface layer thickness, pavement age, and maintenance condition. Furthermore, wavelet denoising is performed to analyze the impact of missing or abnormal data on the LSTM model accuracy. In comparison to a traditional autoregressive integrated moving average (ARIMAX) model, the proposed LSTM model performs better in terms of PI prediction and resiliency to noise. Finally, the overall prediction accuracy of our proposed performance indicator PQI is 93.8

In recent times, remote sensing image super-resolution reconstruction technology based on deep learning has experienced rapid development. However, most algorithms in this domain concentrate solely on enhancing the super-resolution network’s performance while neglecting the equally crucial aspect of inference speed. In this study, we propose a method for lightweight super-resolution reconstruction of remote sensing images, termed SRRepViT. This approach reduces model parameters and floating-point operations during inference through parameter equivalent transformation. Using the RSSOD remote sensing dataset as our benchmark dataset, we compared the reconstruction performance, inference time, and model size of SRRepViT with other classical methods. Compared to the lightweight model ECBSR, SRRepViT exhibits slightly improved reconstruction performance while reducing inference time by 16% and model parameters by 34%, respectively. Moreover, compared to other classical super-resolution reconstruction methods, the SRRepViT model achieves similar reconstruction performance while reducing model parameters by 98% and increasing inference speed by 90% for a single remote sensing image.

The application of nonlinear ultrasonic techniques for the near-end damage characterization of carbon fiber reinforced polymers (CFRP) has proven to be successful and extensive. However, the nonlinear ultrasonic response is weak and may be overwhelmed by the linear component of signals if the damage is located at the far-end. In this paper, the scaling subtraction method (SSM) is adopted to filter out the nonlinear ultrasonic signature aimed at the better characterization of far-end delaminations in CFRP laminates. A three-dimensional finite element model is constructed first to validate the feasibility of separation effect of SSM. Then the experimental study is conducted for a delaminated CFRP plate using laser scanning Doppler vibrometer (LSDV) response measurements. Results show that the far-end delamination can be detected and localized clearly since the nonlinear features can be accurately separated from the surface vibration response by using SSM. Moreover, the far-end delamination size is qualitatively analyzed by using the maximum response amplitude (MRA) of the higher harmonics of SSM signals, and a positively correlated trend is shown as the delamination size increases. This work demonstrates the applicability and potential of SSM technique for the comparative assessment of far-end defect information in composites.

Detecting concrete internal defects through deep learning analysis of impact echo signals faces two challenges: (1) the traditional signal processing method such as wavelet transform (WT) fails to reflect data-sensitive damage characteristics due to the uncertainty principle and (2) the limited labeled data acquired from real structures impedes network training. To address the first challenge, this paper proposes the WT-based synchrosqueezing transform (WT-SST) for the conversion of time-series data to the time-frequency spectrogram, which can provide effective features for the network in time and frequency domains simultaneously. To overcome the second challenge, numerical simulation data are supplemented for the augment of labeled data. To minimize the effect of data variance between experiments and simulations, this paper uses an unsupervised domain adaptation (DA) network for the transfer training of labeled simulation data (original domain) and unlabeled experimental data (target domain). The DA network extracts domain-invariant features by maximizing the domain recognition error and minimizing the probability distribution distance. The damage probability is calculated by the trained model to produce a 2D defect contour image of concrete specimens, and the three-dimensional visualization of internal defects by estimating the defect depth based on the defect area of contour image. Finally, the recognition precision, recall, F1-score, and accuracy of the model of unsupervised DA network trained by a hybrid dataset reaches 89.4%, 88.4%, 88.9%, and 94.7%, respectively.

Second harmonic generation (SHG) technique using Rayleigh surface wave has been successfully attempted for the non-destruction evaluation of asphalt mixture during the self-healing process. The air-coupled transducer placement is desirable for the convenient implementation of SHG for the large-scale detection. Nevertheless, the pronounced attenuation of ultrasonic wave by air-coupled propagation will cause the second harmonic induced by microcracks to be barely detectable, particularly under the influence of inevitable noise. In this study, a novel approach by combining Duffing oscillator and van der pol equation is developed to identify and quantify the second harmonic in noise-contained signals based on the phase trajectory and periodic trajectory area exponent. The multi-physics finite element model is first established to simulate the air-asphalt propagation of Rayleigh waves, and proof-of-concept numerical simulations are carried out to estimate the second harmonic amplitude of noise-contaminated signals. The air-coupled SHG experiments are conducted, and the approach of the Duffing-van der pol oscillator is applied to extract the second harmonic in experimental data, and the self-healing behavior of asphalt mixture is analyzed through the variation of the nonlinear parameter of SHG measurements. The results indicate that the contactless SHG method combined with the Duffing-van der pol oscillator can effectively realize the convenient and reliable estimation of the nonlinear ultrasonic signal during the cycle of asphalt deterioration and healing, which is important for the property characterization of asphalt mixture considering its high attenuation to the ultrasonic waves.

The softening point and gradient healing have significant engineering significance for the self-healing of asphalt mixtures. This study introduces a novel approach utilizing a contactless ultrasonic system in conjunction with the Second Harmonic Generation (SHG) technique to determine the softening point and monitor the gradient healing process of asphalt mixtures. Three distinct fracture patterns resulting from temperature variations are accurately characterized through simulation using the Discrete Element Method (DEM). The ultrasonic nonlinear parameter β, reflecting different crack interfaces, proves to be a highly effective alternative for characterizing asphalt fractures, outperforming traditional methods such as bearing force and wave velocity. In experimental analysis, variations in the healing index highlight the significant influence of asphalt mixture fracture patterns on healing efficiency. The observed decrease in ultrasound amplitude signifies the transition of the asphalt binder from a solid to a liquefied state, accompanied by an increase in the nonlinear parameter indicating enhanced plasticity. The softening point is precisely determined when the Rayleigh surface wave disappears. The cumulative healing process is visualized through the continuous decrease in the nonlinear parameter. To further enhance the understanding of the asphalt self-healing process, the internal temperature gradient field is characterized using the proposed energy difference method. A correlation between the nonlinear parameter and the depth of the temperature interval is established, providing a non-destructive method for evaluating the distribution of the temperature interval. This innovative approach significantly accelerates the development of non-destructive testing in the field of asphalt self-healing.

Locating and imaging internal defects in concrete is a challenging task. The traditional impactecho (IE) method offers only qualitative analysis of defects and requires operation by skilled technicians, often leading to misjudgments and significant consumption of resources. To address these limitations, this study develops an automatic rapid detection system for concrete subsurface defects based on a deep learning model. The system is embedded with the unsupervised domain adaptation (DA) network in which the time-frequency images of IE signals are input, producing a two-dimensional (2D) defect contour map based on the damage probability value. Three types of subsurface defects including delamination, voids and pipes are expected to be detected in the concrete slabs. The system can effectively acquire and process IE signals on the concrete slab along a preset path, and the recognition accuracy, precision, recall, and f1-score reached 98.1 %, 92 %, 79.2 %, and 85.1 %, respectively. This study also establishes a semi-analytical formula linking natural frequency (fI), area-to-depth squared ratio (A/h2), and concrete property variables to estimate the defect depth. The mean square error (MSE) between the estimated value and actual value of the defect depth is only 3.64x10- 4. Overall, the proposed detection system provides quantitative evaluation of concrete subsurface defects with excellent accuracy and efficiency, aiding in the fast repair and maintenance strategies of concrete infrastructure.

In this paper, a hybrid method combining the deep learning algorithm and bending mode analysis of acoustic vibration signals is proposed for the multi-type classification and 3D visualization of internal defects in concrete plates. A novel deep learning model termed fully convolutional network based on principal component analysis and attention-embedded long-short term memory (PCA-ALSTM-FCN) is established. The PCA-ALSTM-FCN model successfully classifies multi-type internal defects, and the two-dimensional (2D) defect contour maps are generated based on the predicted state by the trained model. The defect depths are calculated according to the analytical formula of bending vibration mode of acoustic vibration signals. Combining the 2D defect map and depth information, the three-dimensional (3D) visualization images of defects are obtained. The average recognition accuracy of PCA-ALSTM-FCN model for different types of defects reaches 94.8 %, and the defect depth calculation error range is approximately 10 %-20 %. The experimental results show that the proposed method in this paper can accurately distinguish different types of internal defects and effectively locate the 3D position of defects in concrete.

The performance evaluation of asphalt materials during low-temperature thermal cycles remains a challenge, particularly in quantifying the early tiny damage development and recovery processes. To address this issue, a high-sensitivity nonlinear ultrasonic method -the second harmonic generation (SHG) technique in the form of Rayleigh surface waves, is proposed to investigate these behaviors in asphalt mixture. The decrease in thermal contraction value and stiffness modulus with cycles verifies the damage behavior at the macro level. Turning points in crack length, crack density, and nonlinear parameter during the thermal cycle indicate the generation of cracks from -20 degrees C to -30 degrees C during the cooling process. The decrease of crack length and crack density in the heating phase demonstrates the recovery of asphalt. The nonlinear ultrasonic parameter beta'(c) caused by crack is successfully separated from the total nonlinear parameter beta', and the recovery index (RI)is defined to quantify the recovery behavior. Both of them are highly consistent with crack length and crack density. While the sensitivity index (SI) of nonlinear parameter is hundreds or thousands of times greater than other measurements of damage development. Furthermore, a qualitative correlation between the nonlinear parameters and plastic strain has been established. This breakthrough widens the scope of studying low-temperature weak damage in asphalt mixtures, moving beyond the confines of crack analysis.

Timely crack detection of pavement helps inspectors access road conditions and determine the maintenance strategy, which can reduce repair costs and safety risks. Deep learning has greatly advanced the development of automated crack detection, but there are still challenges that hinder the application of crack segmentation networks in engineering practice such as the bloated models, the class imbalance problem, and the high-performance device dependency. For efficient crack segmentation tasks, this paper proposes a novel high-performance lightweight network termed multi-path convolution feature fusion lightweight network (MCFF-L Net) and utilize the concept of knowledge distillation. The MCFF-L Net with only 1.18 M parameters achieves F1 score of 85.70% and intersection over union (IoU) of 78.22%, which surpasses the popular heavyweight networks and lightweight networks. The proposed network is further implemented on an embedded device of Jetson Xavier NX and the detection speed of pavement cracks can reach 9.71 frames per second (FPS). The combination of embedded device and proposed lightweight networks is suitable for application scenarios where the portable and efficient crack segmentation is needed and the reliable data transmission through network is not accessible.

In this work, the potential of using defect resonance-based vibro-acoustic modulation (VAM) spectroscopy to determine pumping and probing frequencies to maximize the nonlinear VAM response of carbon fiber reinforced polymer (CFRP) plates is investigated. An integrated 3D finite element (FE) model which implements the barely visible impact damage (BVID) due to impact force and the loading of ultrasonic signals is proposed. Based on this model, the scaling subtraction method (SSM) is used to determine the low-frequency global plate resonance that leading strongest defect excitation for pumping input (GPRpump) and high-frequency local defect resonance for probing input (LDRprobe). A scanning laser Doppler vibrometer (SLDV) is adopted to experimentally validated the proposed numerical methodology. Then, the maximum response amplitude (MRA) of nonlinear VAM response is proposed to evaluate the influence of defect resonance on the nonlinear VAM response. The numerical and experimental results are in good agreement and show that MRA of the VAM sideband can be improved when the GPRpump frequency is taken as the pumping input and the LDRprobe frequency is taken as the probing input, respectively. The best BVID detection effect is obtained when the pumping input of the VAM test is selected as the GPRpump frequency and the probing input is selected as sum or difference of the GPRpump frequency and LDRprobe frequency.