Non-destructive identification and quantitative assessment method and system for defects in railway tunnel lining

Abstract

This application provides a non-destructive identification and quantitative assessment method and system for railway tunnel lining defects, specifically relating to the field of railway engineering inspection and intelligent identification technology. The method first acquires multi-source detection signals and performs adaptive preprocessing on them. Then, it performs multi-level fusion of the preprocessed multi-source detection signals under a unified spatial coordinate system to generate a high-precision dataset. Finally, the high-precision dataset is input into a multi-task lining defect assessment model to achieve refined identification and quantitative assessment of railway tunnel lining defects. This method significantly improves the automation and objectivity of lining defect detection, providing a reliable basis for tunnel structure maintenance decisions.

Description

Technical Field

[0001] This invention relates to the field of railway engineering inspection and intelligent identification technology, and more specifically, to a method and system for non-destructive identification and quantitative evaluation of defects in railway tunnel lining. Background Technology

[0002] As a crucial component of railway lines, railway tunnels are subjected to high humidity, high pressure, and complex geological environments for extended periods. This makes their linings prone to latent defects such as voids, cavities, cracks, and insufficient thickness. Failure to detect and address these issues promptly will directly impact the structural safety and operational stability of the tunnel. Therefore, conducting non-destructive testing and condition assessment of tunnel linings is of significant engineering importance. Current railway tunnel lining inspections primarily rely on single detection methods, such as ground-penetrating radar (GPR), ultrasonic or impact-echo imaging, and infrared thermal imaging. While each method has its advantages, single-source data is often limited by the complexity of the lining structure and environmental interference, making it difficult to comprehensively reflect the internal latent defects. For example, GPR results are easily affected by reinforcing steel mesh, electromagnetic wave attenuation, and reflected clutter, leading to complex signal interpretation. While impact-echo imaging can reflect lining thickness and density, its ability to identify the spatial distribution of voids or cavities is limited. Furthermore, current technologies for fusing multi-source detection information often rely on result overlay or manual comparison, lacking a quantitative registration and fusion mechanism within a unified coordinate space, resulting in insufficient fusion accuracy and low information utilization. Meanwhile, in the defect identification stage, manual experience-based interpretation or single-task classification models are often used, making it difficult to achieve automated and refined defect identification. For quantitative assessment of defects, current methods mostly rely on simple amplitude changes of signal features or estimation using empirical formulas, and a systematic quantitative analysis method based on fused data has not yet been developed.

[0003] In summary, existing technologies suffer from problems such as a lack of high-precision spatial registration and fusion mechanisms for multi-source detection data, limited recognition accuracy, and reliance on human experience for evaluation results. Summary of the Invention

[0004] The main objective of this invention is to provide a non-destructive identification and quantitative assessment method and system for defects in railway tunnel lining, so as to at least solve the problems of lack of high-precision spatial registration and fusion mechanism for multi-source detection data, limited identification accuracy, and reliance on human experience for assessment results. This enables an objective assessment of the lining condition and provides a reliable basis for tunnel structural health monitoring and maintenance decisions.

[0005] To achieve the above objectives, the present invention provides a method and system for non-destructive identification and quantitative evaluation of defects in railway tunnel lining.

[0006] In a first aspect, the present invention provides a non-destructive identification and quantitative assessment method for defects in railway tunnel lining, the method comprising:

[0007] A non-destructive testing system is used to collect multi-source detection signals that reflect the structural condition of railway tunnel lining. The multi-source detection signals include at least electromagnetic reflection signals and acoustic echo signals.

[0008] The multi-source detection signal is preprocessed, and the preprocessing includes at least adaptive noise reduction and feature enhancement based on noise characteristic analysis, and the preprocessed multi-source detection signal is output.

[0009] The preprocessed multi-source detection signals are mapped to a unified spatial coordinate system, and the mapped multi-source detection signals are fused at multiple levels to output a fused high-precision dataset.

[0010] Historical multi-source detection signals are acquired, and a multi-task lining defect assessment model for lining defect identification and quantitative assessment is established based on the historical multi-source detection signals, and the multi-task lining defect assessment model is trained.

[0011] The high-precision dataset is input into the trained multi-task lining defect assessment model, which outputs the defect types and quantitative indicators of the tunnel lining; and the lining defects are identified and assessed based on the defect types and quantitative indicators.

[0012] Specifically, preprocessing the electromagnetic reflection signal includes:

[0013] Noise characteristic analysis is performed on the electromagnetic reflection signal to obtain the signal spectrum characteristics and signal-to-noise ratio variation characteristics of the electromagnetic reflection signal. Filtering parameters are adaptively set based on the signal spectrum characteristics and the signal-to-noise ratio variation characteristics, wherein the filtering parameters include a first filtering parameter and a second filtering parameter.

[0014] The electromagnetic reflection signal is subjected to a first noise reduction process to reduce electromagnetic interference. The first noise reduction process includes at least wavelet denoising and a first bandpass filter. The wavelet denoising applies the first filtering parameters, and the first bandpass filter applies the second filtering parameters.

[0015] The denoised electromagnetic reflection signal is subjected to a first feature enhancement to generate a preprocessed electromagnetic reflection signal; the first feature enhancement includes at least envelope extraction and normalization processing.

[0016] Specifically, the preprocessing of the acoustic echo signal includes:

[0017] The acoustic echo signal is subjected to joint time-domain and frequency-domain analysis to obtain analysis results. Based on the analysis results, the energy distribution, amplitude variation, noise level, and frequency characteristics of the acoustic echo signal are statistically analyzed. Based on the statistical results, denoising parameters are adaptively determined, wherein the denoising parameters include a first denoising parameter and a second denoising parameter.

[0018] The acoustic echo signal is subjected to a second noise reduction process to suppress mechanical noise and multipath reflection interference. The second noise reduction process includes at least pulse denoising and a second bandpass filtering. The pulse denoising applies the first noise reduction parameter, and the second bandpass filtering applies the second noise reduction parameter.

[0019] The noise-reduced acoustic echo signal is subjected to a second feature enhancement to generate a preprocessed acoustic echo signal; the second feature enhancement includes at least Hilbert transform and short-time energy enhancement processing.

[0020] Specifically, the multi-level fusion includes at least signal-level fusion, which includes:

[0021] Calculate the local signal-to-noise ratio of the mapped multi-source detection signal at each spatial sampling point. :

[0022]

[0023] in, Indicates the signal source; Indicates the location of the spatial sampling point; The local dominant wave energy of the multi-source detection signal at the spatial sampling point location; The local background noise energy of the multi-source detection signal at the spatial sampling point location;

[0024] A confidence coefficient is assigned to the multi-source detection signal based on physical characteristics or historical experience. ;

[0025] By combining the local signal-to-noise ratio with the confidence coefficient, the weighting coefficients of the multi-source detection signals at the spatial sampling point locations are calculated:

[0026]

[0027] in, This indicates traversing all signal sources; Indicates the first The local signal-to-noise ratio of each signal source at the spatial sampling point location; Indicates the first The credibility coefficient of each signal source; It represents the sum of the products of the local signal-to-noise ratio and the confidence coefficient of all signal sources at the spatial sampling point location;

[0028] Weighting coefficients are applied to the spatial sampling points, and the multi-source detection signals are weighted and superimposed to obtain the signal-level fused signal of the spatial sampling points. :

[0029]

[0030] in, The value of the preprocessed multi-source detection signal at the spatial sampling point location is the corresponding signal value.

[0031] Specifically, the multi-level fusion further includes at least feature-level fusion, which includes:

[0032] The tunnel lining is spatially divided into several detection units, each consisting of multiple spatial sampling points; the multi-source detection signals and the signal-level fused signals of the detection units are used to extract feature vectors of the same type using a unified feature template.

[0033] For each fusion feature in the feature vector of the multi-source detection signal and the signal-level fusion signal, calculate the quality index of each fusion feature, assign corresponding common-source weights to each fusion feature according to the quality index of each fusion feature, and combine the fusion features according to the common-source weights to form a compressed representation vector of the multi-source detection signal and the signal-level fusion signal of the detection unit.

[0034] For the multi-source detection signal and the signal-level fused signal, calculate the overall quality index of the multi-source detection signal and the signal-level fused signal, and calculate the cross-source weights based on the overall quality index of the multi-source detection signal and the signal-level fused signal. Then, weight the compressed representation vectors according to the cross-source weights to obtain the high-precision dataset of feature-level fusion of the detection unit.

[0035] Specifically, establishing and training a multi-task lining defect assessment model for lining defect identification and quantitative evaluation includes:

[0036] The network structure of the multi-task lining defect assessment model is designed based on the objectives of identifying and quantitatively assessing lining defects, so that the multi-task lining defect assessment model can simultaneously predict the defect type and the quantitative index of the tunnel lining.

[0037] The historical multi-source detection signals are preprocessed and fused at multiple levels to obtain a high-precision historical dataset, which is then divided into a training set and a validation set.

[0038] The training set is input into the multi-task lining defect assessment model to initialize the parameters and iteratively train the model. The parameters of the multi-task lining defect assessment model are continuously updated through iterative training. At the same time, the validation set is used to monitor the iterative training process and evaluate the performance of the multi-task lining defect assessment model until the multi-task lining defect assessment model converges on the training set and achieves the expected performance on the validation set, thus obtaining the trained multi-task lining defect assessment model.

[0039] Specifically, the step of inputting the high-precision dataset into the trained multi-task lining defect assessment model, outputting the defect type and quantitative index of the tunnel lining, and identifying and assessing the lining defects based on the defect type and the quantitative index includes:

[0040] The high-precision dataset is input into the trained multi-task lining defect assessment model;

[0041] The detection unit outputs the defect type to achieve refined identification of the lining defect;

[0042] The quantitative index of the detection unit is output, and the severity of the lining defect is graded or determined in combination with a preset threshold.

[0043] Specifically, the quantitative indicators include at least the void volume or area of ​​the tunnel lining, the void thickness or height of the tunnel lining, and the thickness loss of the tunnel lining.

[0044] In a second aspect, the present invention provides a non-destructive identification and quantitative assessment system for defects in railway tunnel lining, the system applying the method described in the first aspect, the system comprising:

[0045] A multi-source signal acquisition unit, wherein the multi-source signal acquisition unit is used to acquire the multi-source detection signals of the tunnel lining;

[0046] A signal preprocessing unit is connected to the multi-source signal acquisition unit; the signal preprocessing unit is used to perform adaptive preprocessing on the multi-source detection signal to reduce noise interference and enhance signal characteristics.

[0047] A signal fusion unit is connected to the signal preprocessing unit; the signal fusion unit is used to perform multi-level fusion on the preprocessed multi-source detection signals to output the fused high-precision dataset.

[0048] A lining defect identification and evaluation unit is provided, which is connected to the signal fusion unit. The lining defect identification and evaluation unit is used to input the high-precision dataset into the multi-task lining defect evaluation model to achieve refined identification and quantitative evaluation of the lining defects.

[0049] Specifically, the lining defect identification and assessment unit includes:

[0050] The model building and training module builds the multi-task lining defect assessment model for predicting the defect type and the quantitative index; and trains the multi-task lining defect assessment model based on the historical multi-source detection data.

[0051] The model output module inputs the high-precision dataset into the trained multi-task lining defect assessment model to perform inference and prediction, and outputs the defect type and the quantitative index.

[0052] The quantitative indicator evaluation module is used to classify or determine the severity of the lining defects based on the quantitative indicators combined with preset thresholds.

[0053] This application provides a method and system for non-destructive identification and quantitative assessment of defects in railway tunnel lining. The method first acquires multi-source detection signals through a non-destructive testing system and performs adaptive preprocessing on these signals. Then, it performs multi-level fusion of the preprocessed multi-source detection signals in the same spatial coordinate system to obtain a high-precision dataset. Next, the high-precision dataset is input into a trained multi-task lining defect assessment model for inference, thereby outputting defect types and quantitative indicators. Finally, the quantitative indicators are combined with preset thresholds to complete the quantitative determination of lining defects. This method solves the problems of lacking high-precision spatial registration and fusion mechanisms for multi-source detection data, limited identification accuracy, and reliance on human experience for assessment results. Thus, it achieves refined identification and quantitative assessment of railway tunnel lining defects, significantly improving the identification accuracy and assessment efficiency of lining defects. Attached Figure Description

[0054] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0055] Figure 1 A flowchart illustrating the non-destructive identification and quantitative assessment method for railway tunnel lining defects provided in this application;

[0056] Figure 2 A schematic diagram of the connection of the non-destructive identification and quantitative assessment system for railway tunnel lining defects provided in this application.

[0057] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0059] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein.

[0060] In this invention, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0061] The method and system for non-destructive identification and quantitative assessment of defects in railway tunnel lining provided in this application firstly acquire multi-source detection signals through a non-destructive testing system and then perform adaptive preprocessing on the multi-source detection signals; next, multi-level fusion of the preprocessed multi-source detection signals is performed in the same spatial coordinate system to obtain a high-precision dataset; then, the high-precision dataset is input into a trained multi-task lining defect assessment model for inference and outputs the defect type and quantitative index; finally, the fine identification of lining defects is achieved based on the defect type, and the quantitative judgment of lining defects is achieved by combining the quantitative index and the preset threshold.

[0062] The technical solution of this application and how it solves the above-mentioned technical problems will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0063] Figure 1A flowchart illustrating the non-destructive identification and quantitative assessment method for railway tunnel lining defects provided in this application is shown below. Figure 1 As shown, this embodiment provides a non-destructive identification and quantitative assessment method for defects in railway tunnel lining. The method includes:

[0064] S101: Use a non-destructive testing system to collect multi-source detection signals that reflect the structural condition of railway tunnel lining. The multi-source detection signals include at least electromagnetic reflection signals and acoustic echo signals.

[0065] The specific steps of implementation S101 include:

[0066] 1. Spatial sampling point layout and signal acquisition

[0067] In order to characterize the physical state of the lining, this embodiment selects at least two detection signals, electromagnetic reflection signal and acoustic echo signal, to reflect the physical structure of the lining;

[0068] Electromagnetic reflection signals are generated based on the differences in dielectric properties between different materials within the lining, thus they can sensitively characterize the interfaces, interlayer distribution, and moisture content changes of the lining materials. When there are delamination, voids, cavities, or water-bearing areas within the lining, its dielectric constant will change significantly compared to intact concrete, causing anomalies in the amplitude, phase, and arrival time of the reflected electromagnetic waves.

[0069] Acoustic echo signals reflect the structural condition by varying acoustic impedance and propagation velocity between different regions of the lining, exhibiting high sensitivity to changes in mechanical properties such as lining thickness, internal damage, and underlying compactness. When cracks, voids, or cavities exist within the lining, the acoustic propagation path and echo energy will be distorted, such as echo advance, abnormal amplitude attenuation, or enhanced scattering. Simultaneously, the echo delay at the bottom interface can be used to measure the lining thickness and its changes.

[0070] Therefore, this embodiment selects at least two signals, electromagnetic reflection signal and acoustic echo signal, to complement each other and jointly deconstruct the physical structure of the tunnel lining. In specific implementation, one or more other detection signals that can characterize the physical state of the lining may be added as appropriate.

[0071] In summary, a non-destructive testing system for acquiring signals should be equipped with at least electromagnetic detection and acoustic detection functions, and a vehicle-mounted non-destructive testing system is preferred for implementation.

[0072] In practice, spatial sampling points are set up along the tunnel lining at preset spatial intervals. The vehicle-mounted non-destructive testing system travels through the tunnel lining, and the testing system is triggered at each spatial sampling point to perform synchronous data collection during the travel.

[0073] 2. Record the detection signal

[0074] At each spatial sampling point, the multi-source time-domain waveform data acquired by the vehicle-mounted non-destructive testing system is recorded. The multi-source time-domain waveform data is the multi-source detection signal.

[0075] 3. Construct multi-source detection signals for spatial sampling points

[0076] The multi-source detection signals corresponding to each spatial sampling point are organized in spatial order to form a multi-source detection signal covering the tunnel lining, which is used for subsequent preprocessing and multi-level fusion.

[0077] This step utilizes an onboard non-destructive testing system to simultaneously acquire detection signals from multiple sources, including electromagnetic reflection and acoustic echo sensors. This allows for the simultaneous characterization of the electromagnetic and mechanical properties of the tunnel lining along a unified travel path, achieving continuous coverage of spatial sampling points. The combined acquisition of multiple signals significantly improves the comprehensiveness and reliability of lining condition information: electromagnetic signals are highly sensitive to changes in the material's dielectric properties, which is beneficial for revealing voids, cavities, and water-containing anomalies; acoustic signals are sensitive to thickness, density, and internal damage, supplementing mechanical property information. The multiple signals complement each other in structural characterization, reducing the blind spots of single detection methods and providing a rich, complementary, and high signal-to-noise ratio input data foundation for subsequent preprocessing and fusion.

[0078] S102: Preprocess the multi-source detection signal, the preprocessing including at least adaptive noise reduction and feature enhancement based on noise characteristic analysis, and output the preprocessed multi-source detection signal;

[0079] Specifically, preprocessing the electromagnetic reflection signal includes:

[0080] Noise characteristic analysis is performed on the electromagnetic reflection signal to obtain the signal spectrum characteristics and signal-to-noise ratio variation characteristics of the electromagnetic reflection signal. Filtering parameters are adaptively set based on the signal spectrum characteristics and the signal-to-noise ratio variation characteristics, wherein the filtering parameters include a first filtering parameter and a second filtering parameter.

[0081] The electromagnetic reflection signal is subjected to a first noise reduction process to reduce electromagnetic interference. The first noise reduction process includes at least wavelet denoising and a first bandpass filter. The wavelet denoising applies the first filtering parameters, and the first bandpass filter applies the second filtering parameters.

[0082] The denoised electromagnetic reflection signal is subjected to a first feature enhancement to generate a preprocessed electromagnetic reflection signal; the first feature enhancement includes at least envelope extraction and normalization processing.

[0083] Specifically, the preprocessing of the acoustic echo signal includes:

[0084] The acoustic echo signal is subjected to joint time-domain and frequency-domain analysis to obtain analysis results. Based on the analysis results, the energy distribution, amplitude variation, noise level, and frequency characteristics of the acoustic echo signal are statistically analyzed. Based on the statistical results, denoising parameters are adaptively determined, wherein the denoising parameters include a first denoising parameter and a second denoising parameter.

[0085] The acoustic echo signal is subjected to a second noise reduction process to suppress mechanical noise and multipath reflection interference. The second noise reduction process includes at least pulse denoising and a second bandpass filtering. The pulse denoising applies the first noise reduction parameter, and the second bandpass filtering applies the second noise reduction parameter.

[0086] The noise-reduced acoustic echo signal is subjected to a second feature enhancement to generate a preprocessed acoustic echo signal; the second feature enhancement includes at least Hilbert transform and short-time energy enhancement processing.

[0087] The specific steps of implementation S102 include:

[0088] This embodiment uses the preprocessing of electromagnetic reflection signals and acoustic echo signals in multi-source detection signals as an example for illustration:

[0089] 1. Preprocessing of electromagnetic reflection signals

[0090] 1.1 Analysis of Electromagnetic Signal Noise Characteristics

[0091] Noise characteristics analysis is performed on the original electromagnetic reflection signal, including spectral distribution, energy concentration area, main noise frequency band and signal-to-noise ratio variation trend, and the main interference types (structural noise, electromagnetic coupling noise, environmental noise, etc.) are identified.

[0092] The filtering parameters are adaptively determined based on the analysis results, where:

[0093] The first filtering parameters (as wavelet denoising parameters) include the waveform decomposition level and the wavelet threshold;

[0094] The second filtering parameter (as a bandpass filter parameter) includes the lower bandpass filter frequency and the upper bandpass filter frequency;

[0095] 1.2 Wavelet Denoising and Bandpass Filtering

[0096] 1.2.1 Wavelet Denoising

[0097] Wavelet denoising is applied to the electromagnetic reflection signal. First, discrete wavelet decomposition is performed on the original electromagnetic reflection signal to decompose the signal into approximation coefficients and detail coefficients at different scales, where noise is mainly concentrated in the high-frequency detail coefficients. Then, based on adaptively determined threshold parameters, thresholding is performed on the detail coefficients at each level to reduce high-frequency noise components and retain effective reflection information. Finally, inverse wavelet transform is performed on the processed detail coefficients and the unprocessed approximation coefficients to reconstruct the signal, thereby obtaining an electromagnetic reflection signal that preserves the main reflection characteristics and significantly suppresses noise.

[0098] Wavelet decomposition of the original signal:

[0099] Wavelet decomposition of the original electromagnetic reflection signal:

[0100] ;

[0101] in, This represents the original electromagnetic reflection signal; Indicates wavelet transform; Indicates the number of waveform decomposition layers, where ; Indicates the first Approximation coefficients of the layer; Indicates the first The detail factor of the layer;

[0102] After wavelet decomposition, approximate coefficients are obtained. and detail coefficient .

[0103] Thresholding is applied to the detail coefficients layer by layer:

[0104] Thresholding is applied to the detail coefficients of each layer:

[0105] ;

[0106] in, This represents the detail coefficients after thresholding. Indicates the wavelet threshold. Represents the threshold function;

[0107] Thresholding will reduce the detail factor Medium less than the threshold Partial (noise) suppression, greater than the threshold Some (signals) are preserved.

[0108] Inverse wavelet transform reconstructs the signal:

[0109] The electromagnetic reflection signal after wavelet denoising is obtained by reconstructing it using inverse wavelet transform:

[0110] ;

[0111] in, This represents the electromagnetic reflection signal after wavelet denoising. This represents the inverse wavelet transform.

[0112] 1.2.2 Bandpass Filtering

[0113] Bandpass filtering for electromagnetic reflection signals adaptively sets the lower and upper bandpass filter frequencies to preserve the effective frequency band containing reflections from the lining structure interface and defect echoes, while suppressing low-frequency components such as probe temperature drift and background magnetic field disturbances, as well as high-frequency electromagnetic noise. In implementation, firstly, based on noise characteristic analysis of the electromagnetic reflection signal, the concentrated frequency band of its main reflection energy is determined, and the optimal filtering bandwidth is adaptively estimated. Then, a digital bandpass filter is constructed to filter the electromagnetic signal, ensuring that the filtered electromagnetic reflection signal retains the interface reflection characteristics and the main energy components of the defect echoes.

[0114] The electromagnetic reflection signal after wavelet denoising is bandpass filtered to obtain the bandpass-filtered electromagnetic reflection signal:

[0115] ,

[0116] ;

[0117] in, This represents the electromagnetic reflection signal after bandpass filtering; This represents the frequency response function of a bandpass filter;

[0118] This indicates the lower limit frequency of the bandpass filter. Indicates the upper limit frequency of the bandpass filter; This indicates that multiplication is performed in the frequency domain to achieve filtering; This indicates that an inverse Fourier transform is performed on the filtered frequency domain signal.

[0119] 1.3 Feature Enhancement

[0120] The Hilbert envelope of the bandpass-filtered electromagnetic reflection signal is extracted and normalized to obtain the preprocessed electromagnetic reflection signal:

[0121] ,

[0122] ;

[0123] in, This represents the electromagnetic reflection signal after envelope extraction; This indicates that the normalized electromagnetic reflection signal is the preprocessed electromagnetic reflection signal.

[0124] 2. Preprocessing of acoustic echo signals

[0125] 2.1 Noise Characteristics Analysis of Acoustic Echo Signals

[0126] Time-domain and frequency-domain analyses were performed on the acoustic echo signal to obtain the following statistical results: energy distribution, amplitude variation, noise level, and frequency characteristics.

[0127] Based on the above statistical results, the denoising parameters are adaptively determined:

[0128] First denoising parameters: pulse detection threshold and pulse denoising window width;

[0129] The second noise reduction parameters are the lower limit frequency of the bandpass filter and the upper limit frequency of the bandpass filter.

[0130] 2.2 Pulse Denoising and Bandpass Filtering

[0131] 2.2.1 Pulse Denoising

[0132] Pulse denoising is used to remove anomalous amplitude changes in acoustic echo signals at spatial sampling points. These abrupt changes are typically caused by sensor jitter, mechanical shock, or transient coupling problems. This method first constructs a local window at each sampling point of the signal, calculating the local mean or median of the signal within the window as the "normal reference level" for that point. When the deviation between the amplitude of a sampling point and the local reference value exceeds a preset pulse detection threshold, that point is considered a pulse noise point, and the median or mean within the window is used to replace the original value, thus effectively eliminating pulse noise. This method can significantly suppress transient spike interference while maintaining the main characteristics of the echo at the lining interface without distortion, providing a smooth and stable input signal for subsequent bandpass filtering and feature enhancement.

[0133] For acoustic echo signals Pulse detection and denoising:

[0134] ;

[0135] in, This represents the signal after pulse denoising. This indicates a median filtering operation, used to replace impulse noise points; Indicates the width of the pulse denoising window;

[0136] when At that time, detection For the impulse noise point, a median filter operation is performed to replace the original acoustic echo signal, thereby completing the impulse denoising of the acoustic echo signal;

[0137] in, This represents the local mean or local median of a window near a spatial sampling point. This indicates the pulse detection threshold.

[0138] 2.2.2 Bandpass Filtering

[0139] The effective information of the acoustic echo signal is mainly concentrated in a specific frequency range determined by the acoustic characteristics of the transmitting transducer and the lining material. The low-frequency part outside this range usually contains background noise such as mechanical vibration, while the high-frequency part is susceptible to sensor electronic noise and multipath reflection interference. In order to extract the main echo components related to the lining thickness boundary and defect response, bandpass filtering is applied to the acoustic echo signal on the spatial sampling point sequence. The bandpass filter selects the signal in the frequency domain according to the preset lower bandpass filter frequency and upper bandpass filter frequency, retaining the effective components with frequencies within the frequency domain range, while suppressing low-frequency noise below the lower bandpass filter frequency and high-frequency interference above the upper bandpass filter frequency. After bandpass filtering, the main echo energy of the acoustic echo signal is effectively highlighted, which helps to enhance the echo attenuation of the lining reflection interface, the void area, and the defect scattering characteristics, providing a higher quality and higher signal-to-noise ratio input signal for subsequent feature enhancement and quantitative evaluation.

[0140] The acoustic echo signal after pulse denoising is bandpass filtered to obtain the bandpass filtered acoustic echo signal:

[0141] ,

[0142] ;

[0143] in, This represents the acoustic echo signal after bandpass filtering; This represents the frequency response function of a bandpass filter;

[0144] This indicates the lower limit frequency of the bandpass filter. Indicates the upper limit frequency of the bandpass filter; This indicates that multiplication is performed in the frequency domain to achieve filtering; This indicates that an inverse Fourier transform is performed on the filtered frequency domain signal.

[0145] 2.3 Feature Enhancement

[0146] The noise-reduced acoustic echo signal is subjected to Hilbert transform and short-time energy enhancement to generate the preprocessed acoustic echo signal.

[0147] In specific implementation, it is preferable to perform normalization processing after short-time energy enhancement processing to generate the pre-processed acoustic echo signal;

[0148] After bandpass filtering, a Hilbert transform is performed on the acoustic echo signal to convert the real-valued signal into an analytic signal to obtain its instantaneous envelope. This allows the amplitude variation characteristics of the lining reflection interface and potential defect locations to be clearly presented in the envelope domain. Subsequently, to further enhance these structural features, short-time energy enhancement processing is applied to the envelope signal. This involves calculating and smoothly amplifying the local energy within a sliding window, significantly highlighting regions with energy concentration characteristics such as voids, cavities, or abrupt thickness changes, thus forming enhanced acoustic echo features. Finally, the enhanced acoustic echo features are normalized to obtain... This refers to the pre-processed acoustic echo signal.

[0149] 3. Output the preprocessed multi-source detection signal

[0150] The preprocessed electromagnetic reflection signal, acoustic echo signal, and other signals characterizing the lining structure were organized into a single file. and output This is for subsequent multi-level fusion processing.

[0151] This step, through adaptive preprocessing, effectively suppresses the effects of environmental noise, electromagnetic interference, and multipath reflection, while highlighting key features related to lining defects. This significantly improves the signal-to-noise ratio and reliability of the signal, providing a high-quality data foundation for subsequent multi-level fusion and refined identification and quantitative assessment of lining defects.

[0152] S103: Map the preprocessed multi-source detection signals to a unified spatial coordinate system, perform multi-level fusion on the mapped multi-source detection signals, and output the fused high-precision dataset.

[0153] Specifically, the multi-level fusion includes at least signal-level fusion, which includes:

[0154] Calculate the local signal-to-noise ratio of the mapped multi-source detection signal at each spatial sampling point. :

[0155] ;

[0156] in, Indicates the signal source; Indicates the location of the spatial sampling point; The local dominant wave energy of the multi-source detection signal at the spatial sampling point location; The local background noise energy of the multi-source detection signal at the spatial sampling point location;

[0157] A confidence coefficient is assigned to the multi-source detection signal based on physical characteristics or historical experience. ;

[0158] By combining the local signal-to-noise ratio with the confidence coefficient, the weighting coefficients of the multi-source detection signals at the spatial sampling point locations are calculated:

[0159] ;

[0160] in, This indicates traversing all signal sources; Indicates the first The local signal-to-noise ratio of each signal source at the spatial sampling point location; Indicates the first The credibility coefficient of each signal source; It represents the sum of the products of the local signal-to-noise ratio and the confidence coefficient of all signal sources at the spatial sampling point location;

[0161] Weighting coefficients are applied to the spatial sampling points, and the multi-source detection signals are weighted and superimposed to obtain the signal-level fused signal of the spatial sampling points. :

[0162] ;

[0163] in, The value of the preprocessed multi-source detection signal at the spatial sampling point location is the corresponding signal value.

[0164] Specifically, the feature-level fusion includes:

[0165] The tunnel lining is spatially divided into several detection units, each consisting of multiple spatial sampling points; the multi-source detection signals and the signal-level fused signals of the detection units are used to extract feature vectors of the same type using a unified feature template.

[0166] For each fusion feature in the feature vector of the multi-source detection signal and the signal-level fusion signal, calculate the quality index of each fusion feature, assign corresponding common-source weights to each fusion feature according to the quality index of each fusion feature, and combine the fusion features according to the common-source weights to form a compressed representation vector of the multi-source detection signal and the signal-level fusion signal of the detection unit.

[0167] For the multi-source detection signal and the signal-level fused signal, calculate the overall quality index of the multi-source detection signal and the signal-level fused signal, and calculate the cross-source weights based on the overall quality index of the multi-source detection signal and the signal-level fused signal. Then, weight the compressed representation vectors according to the cross-source weights to obtain the high-precision dataset of feature-level fusion of the detection unit.

[0168] The specific steps in step S103 during implementation include:

[0169] 1. Spatial mapping

[0170] The preprocessed multi-source detection signals are mapped to a unified spatial coordinate system to align the multi-source detection signals in space, preparing for subsequent multi-level fusion (signal-level fusion and feature-level fusion).

[0171] 2. Signal-level fusion

[0172] In practice, the weighting coefficients of the multi-source detection signals at each spatial sampling point are determined based on the source detection signals. These weighting coefficients are then applied to each spatial sampling point, and the multi-source detection signals are weighted and superimposed to obtain the signal-level fused signal for each spatial sampling point. This provides a data foundation for subsequent feature-level fusion.

[0173] 3. Feature-level fusion

[0174] 3.1 Unified Feature Extraction

[0175] For each detection unit Each signal source (Including multi-source detection signals and signal-level fusion signals) Extract similar features using a unified feature template to obtain feature vectors. :

[0176] ;

[0177] Among them, preset extraction A fusion feature.

[0178] 3.2 Calculation and Weighted Aggregation of Homologous Feature Quality Indicators

[0179] For each source The Each feature in the unit Calculate quality indicators (For example, feature signal-to-noise ratio or stability), normalizing the quality metrics yields the homogeneous weights:

[0180] ,

[0181] ;

[0182] in, ; Indicates the weights of similar origins;

[0183] The compressed representation vector of the signal source is obtained by weighting and aggregating the fusion features based on their common origin weights.

[0184] ;

[0185] in, This represents a compressed representation vector.

[0186] 3.3 Calculation and Weighted Fusion of Cross-Source Quality Indicators

[0187] For each source In unit Calculate the overall quality index (For example, the average signal-to-noise ratio of a cell or the stability of a characterization), normalizing the overall quality metric yields the cross-source weights:

[0188] ,

[0189] ;

[0190] There are t preset signal sources. Indicates cross-source weights;

[0191] The final fused feature vector is synthesized by weighting the compressed representation vectors of each source according to cross-source weights.

[0192] ;

[0193] This is the feature-level fusion output of the detection unit, which forms the sample input of the high-precision dataset; if necessary, [further details needed]. Perform global normalization or dimensionality reduction (PCA, autoencoder) to obtain a compact representation.

[0194] This step, through sequential signal-level fusion and feature-level fusion, fully leverages the complementary advantages of multi-source nondestructive testing signals, significantly improving the accuracy and stability of lining structure state characterization. First, in signal-level fusion, the preprocessed signals from multiple sources at the same spatial sampling point are aligned, synchronized, and weighted for synthesis. This effectively reduces random noise and measurement bias in single-source signals, resulting in a fused signal with a higher signal-to-noise ratio, more complete reflection / echo morphology, and stronger defect response continuity in both the time and frequency domains. Subsequently, feature-level fusion is performed at the detection unit scale. By extracting features from the multi-source signals and the fused signal using a unified template, and weighting and compressing them based on feature quality indices and cross-source quality indices, features with high information contribution and strong stability dominate the fusion process, resulting in a more compact, accurate, and interference-resistant high-precision characterization vector. Through this multi-level fusion strategy, the final dataset significantly improves the input quality of subsequent multi-task defect identification and quantitative evaluation models, enhancing the model's generalization ability and evaluation reliability.

[0195] S104: Acquire historical multi-source detection signals, and establish a multi-task lining defect assessment model for lining defect identification and quantitative assessment based on the historical multi-source detection signals, and train the multi-task lining defect assessment model.

[0196] Specifically, establishing and training a multi-task lining defect assessment model for lining defect identification and quantitative evaluation includes:

[0197] The network structure of the multi-task lining defect assessment model is designed based on the objectives of identifying and quantitatively assessing lining defects, so that the multi-task lining defect assessment model can simultaneously predict the defect type and the quantitative index of the tunnel lining.

[0198] The historical multi-source detection signals are preprocessed and fused at multiple levels to obtain a high-precision historical dataset, which is then divided into a training set and a validation set.

[0199] The training set is input into the multi-task lining defect assessment model to initialize the parameters and iteratively train the model. The parameters of the multi-task lining defect assessment model are continuously updated through iterative training. At the same time, the validation set is used to monitor the iterative training process and evaluate the performance of the multi-task lining defect assessment model until the multi-task lining defect assessment model converges on the training set and achieves the expected performance on the validation set, thus obtaining the trained multi-task lining defect assessment model.

[0200] The specific steps in step S104 during implementation include:

[0201] 1. Model Structure Design

[0202] Based on the dual objectives of lining defect identification (classification task) and defect quantification assessment (regression task), a multi-task network structure with a shared encoder and task branches is constructed. Information complementarity between different tasks is achieved through a shared feature representation layer, and targeted loss functions are designed in the classification and regression branches respectively, so that the multi-task lining defect assessment model can simultaneously output defect type and quantification index.

[0203] 2. Construction of historical high-precision dataset

[0204] Historical multi-source detection signals are subjected to adaptive noise reduction, feature enhancement, and multi-level fusion according to the input preprocessing strategy consistent with the online process to generate a high-precision representation dataset. Subsequently, the historical high-precision dataset is divided into training set and validation set according to the proportion to ensure the independence and reliability of training and performance evaluation of the multi-task lining defect assessment model.

[0205] 3. Model Training and Performance Monitoring

[0206] The training set is input into the multi-task lining defect assessment model to perform parameter initialization and iterative optimization. During training, the classification loss and regression loss are jointly minimized, and the parameters of the shared layer and task branches are continuously updated through backpropagation. At the same time, the validation set is used as independent monitoring data to evaluate and adjust the overfitting trend, task balance and overall performance in real time during the training process.

[0207] 4. Model convergence determination and final model acquisition

[0208] When the model reaches a stable convergence state on the training set, that is, when the loss functions of both tasks tend to be stable and the performance indicators obtained on the validation set meet the expected requirements, the multi-task lining defect assessment model is considered to have completed training, and the final model that can be used for actual reasoning is obtained, namely the trained multi-task lining defect assessment model.

[0209] This step involves constructing and training a multi-task lining defect assessment model based on historical multi-source detection data, achieving a unified modeling capability for identifying and quantifying lining defects. On one hand, training with a preprocessed and multi-layered fused historical high-precision dataset enables the model to learn the deep correlation features between electromagnetic reflection signals and acoustic echo signals under different lining conditions, significantly improving the model's representation and generalization capabilities for complex lining structures. On the other hand, the multi-task network structure simultaneously optimizes defect type classification and quantitative index prediction, enabling synergistic promotion between tasks in a shared feature space, improving the overall accuracy of classification and regression tasks. Furthermore, the separation of training and validation sets, along with continuous monitoring of model performance during training, helps suppress overfitting and ensures the model's stability and reliability on unseen data, ultimately resulting in a high-performance multi-task lining defect assessment model that can be directly deployed in practical engineering projects.

[0210] S105: Input the high-precision dataset into the trained multi-task lining defect evaluation model, output the defect type and quantitative index of the tunnel lining; and identify and evaluate the lining defects based on the defect type and the quantitative index.

[0211] Specifically, the step of inputting the high-precision dataset into the trained multi-task lining defect assessment model, outputting the defect type and quantitative index of the tunnel lining, and identifying and assessing the lining defects based on the defect type and the quantitative index includes:

[0212] The high-precision dataset is input into the trained multi-task lining defect assessment model;

[0213] The detection unit outputs the defect type to achieve refined identification of the lining defect;

[0214] The quantitative index of the detection unit is output, and the severity of the lining defect is graded or determined in combination with a preset threshold.

[0215] The quantitative indicators include at least the volume or area of ​​the voids in the tunnel lining, the thickness or height of the voids in the tunnel lining, and the thickness loss of the tunnel lining.

[0216] The specific steps in step S105 during implementation include:

[0217] 1. Data Input

[0218] The high-precision dataset obtained by preprocessing and multi-level fusion of multi-source detection data is input into the trained multi-task lining defect assessment model.

[0219] 2. Defect type identification

[0220] For each detection unit, the model outputs the type of lining defect, enabling refined identification of tunnel lining defects.

[0221] 3. Quantitative Indicator Prediction

[0222] For each detection unit, the model outputs corresponding quantitative indicators, including the volume or area of ​​voids in the tunnel lining, the thickness or height of voids in the tunnel lining, and the thickness loss of the tunnel lining.

[0223] 4. Quantitative assessment of lining defects

[0224] 4.1 Determination of Void Volume or Area

[0225] The void volume or area of ​​each detection unit is compared with a preset threshold; if the index exceeds the threshold, the unit is determined to have a significant void defect; if it is below the threshold, the void defect is determined to be negligible or minor.

[0226] 4.2 Determination of Void Thickness or Height

[0227] The void thickness or height of each detection unit is compared with the corresponding preset threshold; if it exceeds the threshold, the unit is determined to have a void problem; if it is below the threshold, the void effect can be considered acceptable.

[0228] 4.3 Determination of Lining Thickness Loss

[0229] The thickness loss of each detection unit is compared with the design allowable loss or threshold; if it exceeds the threshold, it is determined that there is a serious thickness loss in the lining; if it does not exceed the threshold, it is determined that the thickness loss is within the safe range.

[0230] 4.4 Comprehensive Defect Level Assessment

[0231] The results of the determination of voids, cavitation, and thickness loss are comprehensively analyzed;

[0232] Based on the number and severity of indicators exceeding the threshold, each detection unit is assigned a defect level (such as minor, moderate, or severe), generating an overall assessment result for lining defects.

[0233] This step achieves simultaneous, refined identification and quantitative assessment of defects in railway tunnel linings by inputting high-precision multi-source detection data into a trained multi-task lining defect assessment model. The model accurately determines the defect type of each detection unit and outputs quantitative indicators such as void volume or area, void thickness or height, and lining thickness loss. It also classifies the defect level based on preset thresholds. This method makes the assessment results more objective, quantitative, and repeatable, effectively reflecting the true condition of the lining structure. It provides a scientific basis for engineering maintenance, structural reinforcement, and safety decisions, significantly improving the accuracy, reliability, and operability of tunnel lining monitoring.

[0234] This embodiment provides a non-destructive testing (NDT) method for identifying and quantitatively assessing defects in railway tunnel linings. The method first acquires multi-source detection signals using a NDT system and performs adaptive preprocessing on these signals. Then, it fuses the preprocessed signals at multiple levels within the same spatial coordinate system to obtain a high-precision dataset. This high-precision dataset is then input into a trained multi-task lining defect assessment model for inference, resulting in the output of defect types and quantitative indicators. Finally, the quantitative indicators are combined with preset thresholds to quantitatively determine the lining defects. This method effectively solves the problems of lacking high-precision spatial registration and fusion mechanisms for multi-source detection data, limited identification accuracy, and reliance on human experience for assessment results. Therefore, it achieves refined identification and quantitative assessment of railway tunnel lining defects, significantly improving the accuracy and efficiency of lining defect identification and assessment.

[0235] Figure 2 This is a connection diagram of the non-destructive identification and quantitative assessment system for railway tunnel lining defects provided in this application, as shown in the figure. Figure 2 As shown, this embodiment provides a non-destructive identification and quantitative assessment system for railway tunnel lining defects. This system applies... Figure 1 The non-destructive identification and quantitative assessment system for railway tunnel lining defects described in the embodiment includes:

[0236] A multi-source signal acquisition unit, wherein the multi-source signal acquisition unit is used to acquire the multi-source detection signals of the tunnel lining;

[0237] A signal preprocessing unit is connected to the multi-source signal acquisition unit; the signal preprocessing unit is used to perform adaptive preprocessing on the multi-source detection signal to reduce noise interference and enhance signal characteristics.

[0238] A signal fusion unit is connected to the signal preprocessing unit; the signal fusion unit is used to perform multi-level fusion on the preprocessed multi-source detection signals to output the fused high-precision dataset.

[0239] A lining defect identification and evaluation unit is provided, which is connected to the signal fusion unit. The lining defect identification and evaluation unit is used to input the high-precision dataset into the multi-task lining defect evaluation model to achieve refined identification and quantitative evaluation of the lining defects.

[0240] Specifically, the lining defect identification and assessment unit includes:

[0241] The model building and training module builds the multi-task lining defect assessment model for predicting the defect type and the quantitative index; and trains the multi-task lining defect assessment model based on the historical multi-source detection data.

[0242] The model output module inputs the high-precision dataset into the trained multi-task lining defect assessment model to perform inference and prediction, and outputs the defect type and the quantitative index.

[0243] The quantitative indicator evaluation module is used to classify or determine the severity of the lining defects based on the quantitative indicators combined with preset thresholds.

[0244] The non-destructive identification and quantitative assessment system for railway tunnel lining defects provided in this embodiment includes the following specific implementations:

[0245] 1. Structure and Connection of Multi-Source Signal Acquisition Unit

[0246] 1.1 Physical Connection

[0247] The multi-source signal acquisition unit is the front-end unit of the system, which is in direct contact with the railway tunnel environment and acquires lining structure data through the vehicle-mounted non-destructive testing system.

[0248] 1.2 Function Implementation

[0249] The system collects multi-source detection signals that reflect the condition of the tunnel lining, including at least electromagnetic reflection signals and acoustic echo signals, and can be expanded to collect other signals that can characterize the physical properties of the lining.

[0250] 1.3 Technical Effects

[0251] By acquiring high-precision, multi-source real-time data, complete information on the lining structure is captured, providing a reliable data foundation for subsequent signal preprocessing and fusion, and ensuring detection accuracy and coverage.

[0252] 2. Signal preprocessing unit structure and connection

[0253] 2.1 Physical Connection

[0254] The signal preprocessing unit is directly connected to the multi-source signal acquisition unit and is used to receive the acquired raw multi-source detection signals.

[0255] 2.2 Function Implementation

[0256] Adaptive preprocessing is performed on multi-source detection signals, including filtering and feature enhancement based on noise characteristic analysis, such as wavelet denoising, bandpass filtering, Hilbert transform, and short-time energy enhancement, to reduce noise interference and extract key features.

[0257] 2.3 Technical Effects

[0258] By performing refined preprocessing on the original signal, the signal-to-noise ratio is improved, the relevant characteristics of lining defects are highlighted, and high-quality signal input is provided for multi-level fusion, thereby improving the accuracy of subsequent identification and evaluation.

[0259] 3. Signal Fusion Unit Structure and Connection

[0260] 3.1 Physical Connection

[0261] The signal fusion unit is directly connected to the signal preprocessing unit and is used to receive preprocessed multi-source signals.

[0262] 3.2 Function Implementation

[0263] The preprocessed multi-source detection signals are fused at multiple levels, including signal-level fusion (based on weighted combination of sampling points) and feature-level fusion (based on weighted feature of detection units), and the fused high-precision dataset is output.

[0264] 3.3 Technical Effects

[0265] By integrating signal-level and feature-level signals, the complementary and redundant information of each source signal is fully utilized to improve the spatial accuracy and reliability of lining defect features, providing high-precision data support for the refined identification and quantitative evaluation of lining defects.

[0266] 4. Lining Defect Identification and Assessment Unit Structure and Connection

[0267] 4.1 Model Building and Training Module

[0268] Physical connection: This module is directly connected to the signal fusion unit and receives the fused high-precision historical dataset as input.

[0269] Functionality: Establish a multi-task lining defect assessment model. The model structure is designed to simultaneously classify defect types and predict quantitative indicators. The model is trained using historical multi-source detection data, including parameter initialization, iterative training and validation until the model converges.

[0270] Technical effect: The multi-task model obtained through training can accurately map the lining defect features with the corresponding defect types and quantitative indicators, providing a reliable computing tool for real-time identification and evaluation, and improving identification accuracy and quantitative evaluation capabilities.

[0271] 4.2 Model Output Module

[0272] Physical connection: This module and model are connected to the training module, and receive the trained multi-task lining defect assessment model and real-time high-precision dataset as input.

[0273] Functionality: The high-precision input dataset is imported into the trained multi-task model for inference and prediction, and the lining defect type and corresponding quantitative index of each detection unit are output.

[0274] Technical benefits: Enables automated identification and quantitative prediction of lining defects, combining qualitative identification and quantitative assessment to improve data processing efficiency and result consistency.

[0275] 4.3 Quantitative Indicator Evaluation Module

[0276] Physical connection: This module is connected to the model output module and receives the output quantization metrics as input.

[0277] Functionality: Based on the quantitative indicators of each detection unit and in combination with preset thresholds, the severity of lining defects is graded or determined, and the health status assessment results of the lining structure are generated.

[0278] Technical benefits: It enables objective and quantitative assessment of lining defects, providing a scientific basis for engineering decisions, maintenance and reinforcement, while reducing the bias caused by subjective human judgment and improving the reliability and repeatability of the assessment.

[0279] This system acquires multi-source detection signals of the tunnel lining through a multi-source signal acquisition unit, performs adaptive preprocessing of the multi-source detection signals through a signal preprocessing unit, performs multi-level fusion of the preprocessed multi-source detection signals to generate a high-precision dataset, and achieves refined identification and quantitative evaluation of lining defects through the input of the high-precision dataset and the output of model inference.

[0280] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the claims.

[0281] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for non-destructive identification and quantitative evaluation of defects in railway tunnel lining, characterized in that, The method comprises: acquiring multi-source detection signals reflecting the state of the railway tunnel lining structure by using a non-destructive testing system, wherein the multi-source detection signals at least include electromagnetic reflection signals and acoustic echo signals; preprocessing the multi-source detection signals, wherein the preprocessing at least includes adaptive noise reduction based on noise characteristic analysis and feature enhancement, and outputs the preprocessed multi-source detection signals; mapping the preprocessed multi-source detection signals to a unified spatial coordinate system, performing multi-level fusion on the mapped multi-source detection signals, and outputting a high-precision data set after fusion; acquiring historical multi-source detection signals, and establishing a multi-task lining defect evaluation model for lining defect identification and quantitative evaluation based on the historical multi-source detection signals and training the multi-task lining defect evaluation model; inputting the high-precision data set into the trained multi-task lining defect evaluation model, outputting the defect type and quantitative index of the tunnel lining, and identifying and evaluating the lining defects based on the defect type and the quantitative index; wherein the multi-level fusion at least includes signal level fusion, and the signal level fusion comprises: computing a local signal-to-noise ratio of the mapped multi-source detection signal at each spatial sampling point : wherein, represents a signal source; represents the spatial sampling point position; is the local dominant wave energy of the multi-source detection signal at the spatial sampling point position; is the local background noise energy of the multi-source detection signal at the spatial sampling point position; assigning a reliability coefficient to the multi-source detection signal according to physical characteristics or historical experience ; combining the local signal-to-noise ratio and the confidence coefficient to calculate the weighting coefficient of the multi-source detection signals at the spatial sampling point position: wherein represents traversing all signal sources; represents the local signal-to-noise ratio of the th signal source at the spatial sampling point position; represents the confidence coefficient of the th signal source; represents the sum of the product of the local signal-to-noise ratio and the confidence coefficient of all signal sources at the spatial sampling point position; applying a weighting coefficient to the spatial sampling points, and weighting and superimposing the multi-source detection signals to obtain a signal level fusion signal of the spatial sampling points : wherein is the corresponding signal value of the pre-processed multi-source detection signal at the spatial sampling point position; the multi-level fusion at least further includes feature level fusion, and the feature level fusion comprises: dividing the tunnel lining into a plurality of detection units according to space, wherein each detection unit is composed of a plurality of spatial sampling points; extracting feature vectors of the same type from the multi-source detection signals and the signal level fusion signals of the detection units according to a unified feature template; for each fusion feature in the feature vectors of the multi-source detection signals and the signal level fusion signals, calculating a quality index of each fusion feature, and assigning a corresponding homologous weight to each fusion feature according to the quality index of each fusion feature, and weighting and combining the fusion features into a compressed representation vector of the multi-source detection signals and the signal level fusion signals of the detection unit according to the homologous weight; for the multi-source detection signals and the signal level fusion signals, calculating an overall quality index of the multi-source detection signals and the signal level fusion signals, and calculating a cross-source weight according to the overall quality index of the multi-source detection signals and the signal level fusion signals, and weighting and combining the compressed representation vector according to the cross-source weight to obtain the high-precision data set of the feature level fusion of the detection unit.

2. The method for non-destructive identification and quantitative evaluation of railway tunnel lining defects according to claim 1, characterized in that, The preprocessing of the electromagnetic reflection signals comprises: performing noise characteristic analysis on the electromagnetic reflection signals to obtain signal spectrum characteristics and signal-to-noise ratio variation characteristics of the electromagnetic reflection signals, and adaptively setting filter parameters according to the signal spectrum characteristics and the signal-to-noise ratio variation characteristics, wherein the filter parameters include first filter parameters and second filter parameters; performing first noise reduction processing on the electromagnetic reflection signals to weaken electromagnetic interference, wherein the first noise reduction processing at least includes wavelet denoising and first band-pass filtering; the wavelet denoising applies the first filter parameters, and the first band-pass filtering applies the second filter parameters; The electromagnetic reflection signal after noise reduction is subjected to first feature enhancement to generate the preprocessed electromagnetic reflection signal; the first feature enhancement at least includes envelope extraction and normalization processing.

3. The method for non-destructive identification and quantitative evaluation of railway tunnel lining defects according to claim 1, characterized in that, The preprocessing of the acoustic wave echo signal includes: The time domain and frequency domain joint analysis of the acoustic wave echo signal is performed to obtain an analysis result, the energy distribution of the acoustic wave echo signal, the amplitude variation of the acoustic wave echo signal, the noise level of the acoustic wave echo signal and the frequency characteristics of the acoustic wave echo signal are statistically based on the analysis result, and the de-noising parameters are adaptively determined based on the statistical result, wherein the de-noising parameters include first de-noising parameters and second de-noising parameters; The second de-noising processing of the acoustic wave echo signal is performed to suppress mechanical noise and multi-path reflection interference, and the second de-noising processing at least includes pulse de-noising and second band-pass filtering; the pulse de-noising applies the first de-noising parameters, and the second band-pass filtering applies the second de-noising parameters; The second feature enhancement of the acoustic wave echo signal after noise reduction is performed to generate the preprocessed acoustic wave echo signal; the second feature enhancement at least includes Hilbert transform and short-time energy enhancement processing.

4. The method for non-destructive identification and quantitative evaluation of railway tunnel lining defects according to claim 1, characterized in that, The establishment of the multi-task lining defect evaluation model for lining defect identification and quantitative evaluation and the training of the multi-task lining defect evaluation model include: According to the identification and quantitative evaluation target of the lining defect, the network structure of the multi-task lining defect evaluation model is designed, so that the multi-task lining defect evaluation model can simultaneously complete the prediction of the defect type and the quantitative index of the tunnel lining; The historical multi-source detection signal is preprocessed and multi-level fused to obtain a historical high-precision data set, and the historical high-precision data set is divided into a training set and a verification set; The training set is input into the multi-task lining defect evaluation model for parameter initialization and iterative training of the multi-task lining defect evaluation model, the parameters of the multi-task lining defect evaluation model are continuously updated through the iterative training, and the process of the iterative training is monitored and the performance of the multi-task lining defect evaluation model is evaluated by using the verification set, until the multi-task lining defect evaluation model converges on the training set and reaches the expected performance on the verification set to obtain the trained multi-task lining defect evaluation model.

5. The method for non-destructive identification and quantitative evaluation of railway tunnel lining defects according to claim 1, characterized in that, The high-precision data set is input into the trained multi-task lining defect evaluation model, and the defect type and quantitative index of the tunnel lining are output; And identifying and evaluating the lining defect based on the defect type and the quantitative index include: The high-precision data set is input into the trained multi-task lining defect evaluation model; The defect type of the detection unit is output to realize fine identification of the lining defect; The quantitative index of the detection unit is output, and the severity of the lining defect is graded or determined in combination with a preset threshold.

6. The method for non-destructive identification and quantitative evaluation of railway tunnel lining defects according to claim 1, characterized in that, The quantitative index at least includes the cavity volume or area of the tunnel lining, the emptying thickness or height of the tunnel lining, and the thickness loss of the tunnel lining.

7. A system for non-destructive identification and quantitative evaluation of defects in railway tunnel lining, characterized by The system applies the method of any one of claims 1-6, and the system comprises: a multi-source signal acquisition unit, configured to acquire the multi-source detection signals of the tunnel lining; a signal preprocessing unit, connected with the multi-source signal acquisition unit, and configured to perform adaptive preprocessing on the multi-source detection signals to reduce noise interference and enhance signal features; a signal fusion unit, connected with the signal preprocessing unit, and configured to perform the multi-level fusion on the preprocessed multi-source detection signals to output the fused high-precision data set; a lining defect identification and evaluation unit, connected with the signal fusion unit, and configured to input the high-precision data set into the multi-task lining defect evaluation model to realize fine identification and quantitative evaluation of the lining defects.

8. The system for non-destructive identification and quantitative evaluation of railway tunnel lining defects according to claim 7, characterized in that, The lining defect identification and evaluation unit comprises: a model establishment and training module, configured to establish the multi-task lining defect evaluation model for predicting the defect types and the quantitative indicators, and to train the multi-task lining defect evaluation model based on the historical multi-source detection data; a model output module, configured to input the high-precision data set into the trained multi-task lining defect evaluation model for inference prediction to output the defect types and the quantitative indicators; a quantitative indicator evaluation module, configured to grade or determine the severity of the lining defects according to the quantitative indicators in combination with preset threshold values.

Images