Tunnel Segment Near-Surface Defect Detection Method Based on Hybrid Deep Learning Network

Abstract

The present invention provides a method for detecting near-surface defects of tunnel segments based on a hybrid deep learning network, which relates to the field of tunnel engineering detection and includes: S1, collecting the near-surface defect burial depth signals of tunnel segment specimens using the impact echo method; S2, inputting the defect burial depth signals into a first preset neural network and a second preset neural network to obtain a first predicted burial depth and a second predicted burial depth; S3, calculating the average error of the defect burial depth signals, shallow burial range signals, and deep burial range signals according to the average value of the differences between the first predicted burial depth and the second predicted burial depth and the actual burial depth; S4, calculating the weight ratios of the two neural networks for the defect burial depth signals, shallow burial range signals, and deep burial range signals, and then calculating the output of the final burial depth according to the adaptive multi-scale weighted algorithm. The present invention calculates the burial depth of defects through a hybrid neural network, realizes the detection of near-surface defects of tunnel segments, and improves the accuracy and efficiency of tunnel segment detection.

Description

Technical Field

[0001] The present invention relates to the field of tunnel engineering detection, and particularly to a method for detecting near-surface defects of tunnel segments based on a hybrid deep learning network. Background Technique

[0002] In recent years, with the continuous development of road traffic and the complexity of geological conditions in China, tunnel engineering has been widely used to meet the road construction needs in complex geological environments. Shield segments are the main assembled components in tunnel shield construction and the innermost barrier of tunnels. They bear the functions of resisting soil pressure, groundwater pressure and some special loads, and have a direct and important impact on the structural performance, durability and waterproof performance of tunnels, as referred to in Reference 1. However, in the process of segment manufacturing and construction, defects and quality problems are inevitable, such as cracks, cavities, pitted surfaces and uneven thickness, etc. These not only affect the bearing capacity and structural stability of tunnel segments, but may also lead to problems such as leakage, corrosion and damage, thus having an unpredictable impact on the operation safety and service life of the entire tunnel project, as referred to in Reference 2. Therefore, the timely detection and evaluation of potential internal defects of tunnel segments become very important.

[0003] The Impact Echo method (IE for short) was developed in the late 1980s and is a very effective detection means for internal defects of structures. IE instruments began to be marketed in the late 1990s. The American Society for Testing and Materials (ASTM) adopted this technology in 1998 and officially used it for the thickness measurement of concrete structures. Zhang Jingbin from Beijing Jiaotong University analyzed the characteristic spectrograms of different concrete specimens based on the Impact Echo method and established a basic calculation formula for the wave velocity and strength of concrete, so as to calculate the compressive strength value of concrete structures, as referred to in Reference 3. Dorafshan and Azari used the Impact Echo method to detect respectively the reinforced concrete bridge samples with artificial defects, the laboratory-made concrete samples with artificial underground defects and the covering systems made of cement and asphalt covering materials, and all obtained relatively accurate detection results, providing an effective method for detecting the quality of concrete structures, as referred to in References 4 and 5.

[0004] With the continuous development of Internet technology and the gradual upgrading of various computer parameters such as computing power and computing speed, the large-scale computing requirements of deep learning are met, and more and more algorithm models are applied to damage identification and detection, see Reference 6. Deep learning networks do not need to artificially set feature vectors according to image features and can be used for the detection and identification of complex tunnel damages. Zhu Hongchen et al. based on the Faster-RCNN object detection algorithm, trained and tested the constructed crack data set, and the results showed that the algorithm is applicable to the lining crack detection task in actual tunnel engineering, see Reference 7; Kumar et al. used the YOLO-v3 deep learning model for real-time detection of concrete damage in high-rise civil structures, and the proposed real-time damage detection system for high-rise structures based on unmanned aerial vehicles can be used to test concrete cracks of different shapes and sizes and the spalling of high-rise concrete structures, see Reference 8; Wu Hehe et al. proposed a tunnel crack detection method based on Faster R-CNN, verifying the possibility and accuracy of using deep learning methods to detect tunnel cracks, see Reference 9.

[0005] Segment lining is the permanent lining structure of shield tunnels. The quality of segment lining is directly related to the overall quality and safety of the tunnel, affecting the waterproof performance and durability of the tunnel. In engineering practice, when there are defects near the surface of tunnel segments, the existing detection methods can only detect the types of defects but cannot well locate the defect positions.

[0006] References:

[0007] [1] Wang Zhenxin. Durability of Shield Tunnels [J]. Underground Engineering and Tunnels, 2002(02): 2 - 5 + 49.

[0008] [2] Ye Zhi, Fu Anran, Liu Huabei. Influence of Seepage Erosion at the Crown of Shield Tunnels on Ground Settlement and Structural Deformation [J]. Journal of Hohai University (Natural Sciences), 2021, 49(03): 279 - 287.

[0009] [3] Zhang Jingbin. Application Research of Impact Echo Method in Nondestructive Testing of Prestressed Concrete Structures [D]. Beijing Jiaotong University, 2018.

[0010] [4] Dorafshan S, Azari H. Deep learning models for bridge deck evaluation using impact echo [J]. Construction and Building Materials, 2020, 263: 120109.

[0011] [5] Dorafshan S, Azari H. Evaluation of bridge decks with overlays using impact echo, a deep learning approach[J]. Automation in Construction, 2020, 113: 103133.

[0012] [6] Xue Yadong, Li Yicheng. A method for identifying diseases of shield tunnel linings based on deep learning[J]. Journal of Hunan University (Natural Sciences), 2018, 45(03): 100 - 109. DOI: 10.16339 / j.cnki.hdxbzkb.2018.03.012.

[0013] [7] Zhu Hongchen, Liu Yuchu. Intelligent identification of tunnel lining cracks based on deep learning[J]. Smart City, 2024, 10(03): 12 - 14. DOI: 10.19301 / j.cnki.zncs.2024.03.004.

[0014] [8] P. Kumar, S. Batchu, N. Swamy S. and S. R. Kota, "Real - Time Concrete Damage Detection Using Deep Learning for High Rise Structures," in IEEE Access, vol. 9, pp. 112312 - 112331, 2021, doi: 10.1109 / ACCESS.2021.3102647.

[0015] [9] Wu Hehe, Wang Anhong, Wang Haidong. Crack detection in tunnel images based on Faster R - CNN[J]. Journal of Taiyuan University of Science and Technology, 2019, 40(03): 165 - 168.

[0016] The background description provided in this article is for the purpose of presenting the context of the present disclosure generally. Unless otherwise indicated herein, the materials described in this section are not prior art to the claims of this application and should not be admitted as prior art by including them in this section. Summary of the Invention

[0017] In order to overcome the deficiencies in the background technology, the present invention discloses a method for detecting near - surface defects of tunnel segments based on a hybrid deep learning network.

[0018] To achieve the above - mentioned invention purpose, the present invention adopts the following technical solutions:

[0019] A method for detecting near-surface defects of tunnel segments based on a hybrid deep learning network, comprising the following steps:

[0020] S1. Use the impact echo method to collect the signal of the buried depth of near-surface defects of the tunnel segment test block;

[0021] S2. Input the buried depth signal of the defect into the first preset neural network and the second preset neural network to obtain the first predicted buried depth and the second predicted buried depth; the buried depth signal of the defect includes a shallow buried range signal and a deep buried range signal; the first predicted buried depth includes a first shallow buried predicted buried depth and a first deep buried predicted buried depth; the second predicted buried depth includes a second shallow buried predicted buried depth and a second deep buried predicted buried depth; the first preset neural network and the second preset neural network are neural networks of different types;

[0022] S3. Calculate the average error of the buried depth signal of the defect, the shallow buried range signal and the deep buried range signal in the first preset neural network and the second preset neural network according to the average value of the differences between the first predicted buried depth and the second predicted buried depth and the actual buried depth;

[0023] S4. Respectively perform reciprocal normalization on the average errors of the buried depth signal of the defect, the shallow buried range signal and the deep buried range signal in the first preset neural network and the second preset neural network to obtain their weight ratios in the two neural networks, and then calculate the output of the final buried depth according to the adaptive multi-scale weighted algorithm.

[0024] Specifically, step S3 specifically includes the following steps:

[0025] S31. Calculate the differences between the first predicted buried depth y1 and the second predicted buried depth y2 and the actual buried depth respectively, and then average the differences respectively to obtain the first average error and the second average error

[0026] S32. Calculate the differences between the first shallow buried predicted buried depth y_l1 and the second shallow buried predicted buried depth y_l2 and the actual buried depth respectively, and then average the differences respectively to obtain the first average shallow buried error and the second average shallow buried error

[0027] S33. Calculate the differences between the first deep buried predicted buried depth y_h1 and the second deep buried predicted buried depth y_h2 and the actual buried depth respectively, and then average the differences respectively to obtain the first average deep buried error and the second average deep buried error

[0028] Specifically, step S4 specifically includes the following steps:

[0029] S41. Reciprocally normalize the average errors of the defect burial depth signal in the first preset neural network and the second preset neural network to obtain the first burial depth weight ratio Q1 and the second burial depth weight Q2;

[0030] S42. Reciprocally normalize the average errors of the shallow burial range signal in the first preset neural network and the second preset neural network to obtain the first shallow burial weight ratio Q_l1 and the second shallow burial weight Q_l2;

[0031] S43. Reciprocally normalize the average errors of the deep burial range signal in the first preset neural network and the second preset neural network to obtain the first deep burial weight ratio Q_h1 and the second deep burial weight Q_h2;

[0032] S44. Calculate the final burial depth according to the adaptive multi-scale weighting algorithm; the adaptive multi-scale weighting algorithm is as follows:

[0033]

[0034] where y is the final burial depth; thr is the preset burial depth threshold, thr = (d max -d min )÷2 + d min , d min is the minimum burial depth; d max is the maximum burial depth.

[0035] Specifically, step S1 is as follows: Conduct impact echo detection on the near surface of the tunnel segment specimen, and then collect acoustic wave data through an air-coupled probe to obtain the defect burial depth signal.

[0036] Specifically, the tunnel segment specimen in step S1 is buried with defects of the same type but different burial depths; the defects are filled with hard foam materials.

[0037] Specifically, in step S2, the first preset neural network obtains the corresponding first predicted burial depth according to the input defect burial depth signal; the first preset neural network is trained by obtaining the first time-frequency image data according to the defect burial depth signal through short-time Fourier transform.

[0038] Specifically, the first preset neural network in step S2 is a ResNeXt model.

[0039] Specifically, in step S2, the second preset neural network obtains the corresponding second predicted burial depth according to the input defect burial depth signal; the second preset neural network is trained by obtaining the second time-frequency image data according to the defect burial depth signal through S-transform.

[0040] Specifically, the second preset neural network in step S2 is a ConvNeXt V2 model.

[0041] Specifically, before training the first preset neural network and the second preset neural network, it further includes: converting the first time-frequency image data and the second time-frequency image data into a tensor format and performing normalization processing through formula (1), and formula (1) is as follows:

[0042]

[0043] Among them, normalizedpixel is the pixel value after image data normalization; pixel is the original pixel value in the image data; mean is the mean vector; std is the standard deviation vector.

[0044] The method for detecting near-surface defects of tunnel segments based on a hybrid deep learning network provided by the present invention includes the following steps: S1. Using the impact echo method to collect the near-surface defect burial depth signal of the tunnel segment test block; S2. Inputting the defect burial depth signal into the first preset neural network and the second preset neural network to obtain the first predicted burial depth and the second predicted burial depth; S3. Calculating the average error of the defect burial depth signal, the shallow burial range signal, and the deep burial range signal in the first preset neural network and the second preset neural network according to the average value of the differences between the first predicted burial depth and the second predicted burial depth and the actual burial depth; S4. Respectively performing reciprocal normalization on the average errors of the defect burial depth signal, the shallow burial range signal, and the deep burial range signal in the first preset neural network and the second preset neural network to obtain their weight ratios in the two neural networks, and then calculating and outputting the final burial depth according to the adaptive multi-scale weighting algorithm. The present invention mainly focuses on using the impact echo method based on an air-coupled ultrasonic array to generate and collect acoustic wave data, converting the acoustic wave data into image data through a data processing process, and calculating the burial depth of defects using a hybrid neural network to realize the detection of near-surface defects of tunnel segments, improving the accuracy and efficiency of tunnel segment detection.

[0045] In addition, the average value of the absolute value of the difference between the calculation result of the data collected by the present invention on the hybrid depth prediction model and the actual burial depth is 1.76 mm, which can well locate the position of the near-surface defects of the tunnel segment and has guiding significance for actual engineering practice;

[0046] In addition, the present invention performs a series of standard preprocessing steps on the tunnel segment defect data image before model training to ensure that each image received by the model has the same scale and distribution. Such processing helps the model to better learn and generalize, makes the training of the model smoother, and makes the model easier to converge.

[0047] In addition, the present invention provides a method for manufacturing a tunnel segment test block, which can simulate the actual situation of tunnel segments under controlled laboratory conditions, allowing researchers to precisely control the experimental conditions, making the data more controllable. Moreover, the production of laboratory test blocks can follow a standardized process, making the experimental results more comparable. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0049] Figure 1 is a schematic flow chart of a method for detecting near-surface defects of tunnel segments based on a hybrid deep learning network according to an embodiment of the present invention;

[0050] Figure 2 is a schematic diagram of the principle of the impact echo method according to an embodiment of the present invention;

[0051] Figure 3 is a schematic diagram of the structure of an air-coupled sensor array according to an embodiment of the present invention;

[0052] Figure 4 is a schematic diagram of a test block design according to an embodiment of the present invention;

[0053] Figure 5 is a schematic diagram of the basic framework and defect fixation of a test block according to an embodiment of the present invention;

[0054] Figure 6 is a schematic diagram of a cast test block according to an embodiment of the present invention;

[0055] Figure 7 is a schematic diagram of the structure of a hybrid deep learning network according to an embodiment of the present invention;

[0056] Figure 8 is a schematic diagram of the STFT transform according to an embodiment of the present invention;

[0057] Figure 9 is a schematic diagram of the representation form of the ResNext blcok according to an embodiment of the present invention; wherein Figure 9 (a) is the original representation form; Figure 9 (b) is the Concat form; Figure 9 (c) is the grouped convolution form;

[0058] Figure 10 is a schematic diagram of the S transform according to an embodiment of the present invention;

[0059] Figure 11 It is a schematic diagram of the curve of the ResNeXt buried depth prediction loss change provided according to the embodiments of the present invention. Specific implementation manners

[0060] The present invention can be explained in detail through the following embodiments. The purpose of disclosing the present invention is to protect all technical improvements within the scope of the present invention. In the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "front", "rear", "left", "right", etc. indicating the orientation or positional relationship, they are only corresponding to the drawings of the present application. For the convenience of describing the present invention, rather than indicating or implying that the device or element referred to must have a specific orientation.

[0061] Embodiment 1

[0062] Refer to Figure 1 , this embodiment provides a method for detecting near-surface defects of tunnel segments based on a hybrid deep learning network, which specifically includes the following steps:

[0063] S1. Use the impact echo method to collect the buried depth signals of near-surface defects of tunnel segment specimens;

[0064] Refer to Figure 2 , the impact echo method is a precise non-destructive testing technology. It ingeniously utilizes the propagation characteristics of elastic waves excited by transient impacts in the medium to generate acoustic wave data. Typical IE instruments include an impact source, a nearby receiver, and a data acquisition system. A hammer or other impact device is used on the surface of the structure to excite an instantaneous shock wave. These waves propagate inside the concrete and are reflected when encountering cracks, voids, or other defects; these reflected waves are captured by acoustic wave sensors installed on the surface of the structure, and the sensors transmit the captured signals to the data acquisition card; subsequently, the data acquisition system receives these signals, and the data analysis and processing system converts the time-domain waveform of the reflected wave into a frequency-domain signal.

[0065] Specifically, impact echo detection is performed on the surface of the tunnel segment specimen, and signals are collected through an air-coupled probe;

[0066] In this embodiment, an air-coupled probe is used for signal collection in the data collection stage of the impact echo method for tunnel segment specimens. The air-coupled probe has the advantages of non-contact, high sensitivity, can effectively detect and locate the defect area, and can avoid using a coupling agent between the probe and the material to be inspected. The air-coupled sensor array can also improve the detection speed and detection efficiency in engineering applications. Refer to Figure 3. In the inspection of tunnel segments, data can be generated by performing impact echo testing on the segment surface, which can accurately locate the position of cracks, evaluate their depth and severity. This method can not only provide an intuitive image of the cracks, but also reveal the shape and nature of the cracks by analyzing the time-frequency characteristics of the reflected waves. The application of this technology can greatly improve the efficiency and accuracy of tunnel segment inspection, providing a scientific basis for the maintenance and repair of tunnel projects.

[0067] Specifically, defects of the same type but different buried depths are buried in the tunnel segment test blocks.

[0068] This embodiment also provides a method for manufacturing a tunnel segment test block. In this embodiment, a reinforced concrete test block is cast to simulate the tunnel segment itself, and hard foam with different buried depths is buried in the reinforced concrete test block to simulate the "hollow" defect. During the experiment, the size of the test block designed in this paper is 1500mm * 1200mm * 200mm. The test block defects are filled with hard foam materials, and the sizes are all 100mm × 100mm. The burial depth is controlled as a single variable. The hard foam is buried in the range from 10mm deep to 170mm deep, and a burial depth is set every 5mm. The design drawing is as Figure 4 shown.

[0069] When manufacturing the reinforced concrete test block simulating the tunnel segment, basic frames with different depths are welded with steel bars to fix the positions of the buried materials, especially to ensure the accuracy of the defect burial depth, as Figure 5 shown. Fix the prepared buried materials in the set positions, and finally pour concrete into the steel bar framework to form a block, completing the production of the test block. As shown in 6, it is used for the test block tapping experiment after being completely solidified and formed.

[0070] Use the impact echo method based on air-coupled ultrasonic array to collect data. Prepare audio signal collection software on the computer, connect multiple air-coupled sound receiving units to form a specific geometric array, fix the device near the sound generation position, and use an automatic tapping device to tap the test block at a certain frequency to generate acoustic wave data. Since the tunnel segment test block contains defects with different buried depths, the acoustic wave data contains the buried depth signals of the near-surface defects of the tunnel segment test block. After being collected by the air-coupled sound receiving units, the computer displays the acoustic wave waveform for real-time observation of the data validity.

[0071] Since the designed defect positions are marked before signal collection, the actual buried depth corresponding to the collected defect buried depth signal can be known during signal collection.

[0072] S2. Input the defect burial depth signal into the first preset neural network and the second preset neural network to obtain the first predicted burial depth and the second predicted burial depth; the defect burial depth signal includes a shallow burial range signal and a deep burial range signal; the first predicted burial depth includes a first shallow burial predicted burial depth and a first deep burial predicted burial depth; the second predicted burial depth includes a second shallow burial predicted burial depth and a second deep burial predicted burial depth; the first preset neural network and the second preset neural network are neural networks of different types.

[0073] Reference Figure 7 , in the depth detection of near-surface defects of tunnel segments, due to different defect depths, different scale features may be involved. For example, shallow defects may correspond to higher-frequency features, while deep defects may correspond to lower-frequency features. To solve this problem, in this embodiment, a hybrid neural network is formed by using two neural networks, enabling them to capture different scale features and realizing the fusion of multi-scale information through weighted calculation of the results.

[0074] In this embodiment, based on the first preset neural network ResNeXt model and the second preset neural network ConvNeXt V2 model, multi-scale defect burial depth calculations are performed on the experimental data respectively. Considering the differences in the calculation results of the two models at different scales, an adaptive multi-scale weighted algorithm is proposed to combine the prediction results of the two:

[0075] First, divide the collected defect burial depth signal y into a shallow burial range signal y_l and a deep burial range signal y_h according to the burial depth range. The burial depth range can be divided and selected according to the designed thickness of the tunnel segment test block and the distribution of the pre-embedded defects.

[0076] Specifically, in this example, the defect burial depth signal with a burial depth range within [10, 65] mm is divided into the shallow burial range signal y_l, and the defect burial depth signal with a burial depth range within [70, 170] mm is divided into the deep burial range signal y_h.

[0077] Input the defect burial depth signal y into the hybrid neural network model. The defect burial depth signal y obtains the corresponding first pre-burial depth y1 through the first preset neural network, and the defect burial depth signal y obtains the corresponding second pre-burial depth y2 through the second preset neural network. It can be understood that since the defect burial depth signal y is composed of the shallow burial range signal y_l and the deep burial range signal y_h, when obtaining the first predicted burial depth, the first shallow burial predicted burial depth y_l1 and the first deep burial predicted burial depth y_h1 can be obtained at the same time. When obtaining the second predicted burial depth, the second shallow burial predicted burial depth y_l2 and the second deep burial predicted burial depth y_h2 can be obtained at the same time.

[0078] Specifically, the first preset neural network obtains the corresponding predicted buried depth according to the input defect buried depth signal; the first preset neural network is trained by obtaining first time-frequency image data through short-time Fourier transform of the defect buried depth signal;

[0079] The short-time Fourier transform (STFT) is achieved by sliding a time window with a fixed length over the signal and performing Fourier transform on the signal segment within the window. The core idea of this method is to divide the signal into a series of short-time segments and then perform frequency analysis on these segments separately. Since each segment is extracted from the original signal, they can capture the frequency components of the signal near a specific time point.

[0080] After the defect buried depth signal undergoes short-time Fourier transform, a series of time-frequency spectrum images will be obtained. As Figure 8 shown, in the time-frequency spectrum diagram, the horizontal axis usually represents time, the vertical axis represents frequency, and the color or brightness represents the frequency intensity or energy at a specific time point. These frequency diagrams show the frequency components of the signal in different time periods, which are called the first time-frequency diagrams here;

[0081] According to the test block design of the simulated tunnel segment, the true collected data of the defect buried depth ranges from 10 mm to 170 mm, the defect buried depth moves down 5 mm each time, and 100 data are taken for STFT transformation for each depth of the defect signal, and 80% of the data are randomly selected for training and 20% of the data are used for verification.

[0082] The first preset neural network is a ResNeXt model;

[0083] Generally, to improve the accuracy of the network model, methods such as deepening or widening the network are adopted. However, as the number of hyperparameters in the network structure increases (such as the number of channels, filter size, etc.), the difficulty of network design and the computational overhead will also increase. ResNeXt is an improved deep learning architecture that can improve the accuracy without increasing the parameter complexity and at the same time reduce the number of hyperparameters. It uses the split-transform-merge strategy of Inception, while keeping in mind the concept of repeating blocks with the same structure in VGG and cross-layer branches in ResNet. The difference between ResNeXt and ResNet-50 lies only in the block among them, including BN and ReLU. The ResNeXt block structure is as Figure 9 shown.

[0084] The ResNeXt block has various forms of expression, including the original form, the Concat form, and the grouped convolution form. Figure 9(a) shows the original representation of the ResNeXt block, which uses the split-transform-merge strategy of Inception to construct the corresponding block: reduce the number of channels, run 3X3 convolution, resize the width, and add the results of each branch, where the number of branches is called cardinality, which is 32 in the figure above. Cardinality is regarded as another dimension of the neural network (ResNeXt) in addition to depth and width. Figure 9 (b) Similar to the original representation, only the merging method is converted to Concat form. Figure 9 (c) is the simplest form of the ResNeXt block and is also the ResNeXt form used in this embodiment. The model starts with a 1X 1 convolution, then concats the number of channels, and finally applies a 1X 1 convolution to reduce the input 256 channels to 128 channels. Although these three ResNeXt blocks differ in form, they are equivalent.

[0085] In this embodiment, the first preset neural network model adopts the ResNeXt model with a 3*3 convolution kernel. This hyperparameter directly affects the performance of the model, the size of the receptive field, the computational complexity, and the ability to extract features. ResNeXt uses a small convolution kernel, which can better capture local details and texture information and easily extract small changes in the image;

[0086] The first preset neural network is trained using the first time-frequency image data obtained in advance, and the training process is as follows: input the first time-frequency image, normalize the time-frequency image, input the normalized image data into the first preset neural network, and output the corresponding first predicted burial depth.

[0087] Specifically, the second preset neural network obtains the corresponding second predicted burial depth according to the input defect burial depth signal; the second preset neural network is trained by obtaining the second time-frequency image data according to the defect burial depth signal according to S transformation;

[0088] S transform, also known as Stockwell transform, is an extension of the idea of ​​continuous wavelet transform and can also be regarded as a wavelet transform after phase correction. It overcomes the defect of fixed window function width of short-time Fourier transform and uses a movable, scalable and frequency-varying Gaussian window function. Figure 10 .

[0089] The S transform is defined as:

[0090]

[0091] where: w(τ - t, f) is the Gauss window; τ is the parameter controlling the position of the Gauss window on the time t axis; f is the frequency; x(t) is the audio frequency data collected in the experiment; j is the imaginary unit, and its value is e -j2πft is the complex exponential function of frequency f; dt represents the integration of the entire signal x(t) in the time domain;

[0092] where in the formula, σ is the abbreviation of, and σ(f) is the scale factor of the Gauss window function;

[0093] According to the design of the test block of the simulated tunnel segment, the buried depth of the defects in the truly collected data ranges from 10 mm to 170 mm, and the buried depth of the defects moves down by 5 mm each time. For each depth of the defect signal, 100 pieces of data are taken to perform the S transform, and 80% of the data is randomly selected for training, and 20% of the data is used for verification.

[0094] ConvNeXt V2 is a new type of convolutional neural network architecture. It integrates a fully convolutional mask autoencoder framework and a global response normalization (GRN) layer in the model architecture, and uses self-supervised learning technology to improve the generalization ability and efficiency of the model.

[0095] In this embodiment, the second preset neural network model is the ConvNeXt V2 model. The ConvNeXt V2 model uses a 7*7 convolutional kernel, and this hyperparameter directly affects the performance of the model, the size of the receptive field, the computational complexity, and the feature extraction ability. ConvNeXt V2 uses a large convolutional kernel, which can capture global features better, cover a larger receptive field, and help understand the overall structure and context information of the image.

[0096] The second preset neural network is the ConvNeXt V2 model;

[0097] The second preset neural network is trained using the pre-processed second time-frequency image data, and its training process is as follows: input the second time-frequency image, normalize the time-frequency image, input the normalized image data into the second preset neural network, and output the corresponding second predicted buried depth;

[0098] Before training the first preset neural network and the second preset neural network, it also includes: converting the first time-frequency image data and the second time-frequency image data into a tensor format and performing standardization processing through formula (1).

[0099] A series of standard preprocessing steps were performed on the input processed tunnel segment defect data images, namely the first time-frequency image data and the second time-frequency image data, to ensure that each image received by the model has the same scale and distribution. Such processing helps the model to better learn and generalize, making the training of the model smoother and easier for the model to converge. This includes converting the image data into a tensor format and normalizing it using the following formula (1):

[0100]

[0101] where normalizedpixel is the pixel value after normalizing the image data; pixel is the original pixel value in the image data; mean is the mean vector (0.485, 0.456, 0.406); std is the standard deviation vector (0.229, 0.224, 0.225).

[0102] The values of mean and std are from the current largest image dataset ImageNet, and this set of mean and std is applicable to most images.

[0103] S3. Calculate the average error of the defect burial depth signal, shallow burial range signal, and deep burial range signal in the first preset neural network and the second preset neural network according to the average of the differences between the first predicted burial depth, the second predicted burial depth, and the actual burial depth;

[0104] Specifically, it includes the following steps:

[0105] S31. Calculate the differences between the first predicted burial depth y1 and the second predicted burial depth y2 and the actual burial depth respectively, and then average the differences to obtain the first average error and the second average error

[0106] S32. Calculate the differences between the first shallow burial predicted burial depth y_l1 and the second shallow burial predicted burial depth y_l2 and the actual burial depth respectively, and then average the differences to obtain the first average shallow burial error and the second average shallow burial error

[0107] S33. Calculate the differences between the first deep burial predicted burial depth y_h1 and the second deep burial predicted burial depth y_h2 and the actual burial depth respectively, and then average the differences to obtain the first average deep burial error and the second average deep burial error

[0108] This embodiment also provides a specific implementation manner for calculating the difference between the predicted burial depth and the actual burial depth of the model,

[0109] Specifically, taking the ResNeXt model as an example, before using the ResNeXt model for burial depth detection, the data needs to be preprocessed. First, for the burial depth corresponding to the image defects, the experiment performed max-min normalization on the predicted burial depth using formula (2), mapping the burial depth value between 0 and 1. Formula (2) is as follows:

[0110]

[0111] where d normalized is the normalized burial depth; d min is the minimum burial depth; d max is the maximum burial depth;

[0112] In this experiment, the minimum burial depth d min of the collected data was set to 10, and the maximum burial depth d max was set to 170. The mean squared error (MSE) loss was selected as the loss function during model training. It measures the average squared difference between the predicted burial depth value and the actual burial depth value, as shown in formula (3):

[0113]

[0114] where MSE is the mean squared error; is the predicted burial depth value; y i is the actual burial depth value; n is the number of data;

[0115] Each model training lasted for 100 epochs, and an early stopping strategy was adopted. When the loss on the validation set did not further decrease for 8 consecutive epochs, the training would stop early. This helped to avoid overtraining the model and saved the model state with the optimal validation performance at the appropriate time. After each training epoch, the performance of the model on the validation set was evaluated by calculating the root mean squared error (RMSE), and then it was denormalized to the range of the original burial depth value to obtain the actual error of the model in predicting the burial depth, as shown in formula (4):

[0116] RMSE original = RMSE normalized ×(d max - d min ) (4)

[0117] where, RMSE original represents the root mean squared error after denormalization, that is, the root mean squared error on the original dataset; RMSE normalized is the root mean squared error obtained after normalizing the data;

[0118] Figure 11 is the curve graph of the burial depth prediction loss change of the ResNeXt model.

[0119] It is understandable that in the ConvNeXt V2 model, the above method can also be used to predict the change in buried depth difference and the change in buried depth prediction loss.

[0120] S4. Respectively, take the reciprocal normalization of the average errors of the defect buried depth signal, shallow buried range signal, and deep buried range signal in the first preset neural network and the second preset neural network to obtain their weight ratios in the two neural networks, and then calculate and output the final buried depth according to the adaptive multi-scale weighting algorithm.

[0121] Specifically, it includes the following steps:

[0122] S41. Take the reciprocal normalization of the average errors of the defect buried depth signal in the first preset neural network and the second preset neural network to obtain the first buried depth weight ratio Q1 and the second buried depth weight Q2;

[0123] Calculate the first average error and the second average error reciprocal and normalize these reciprocals to obtain the weight ratios of the defect buried depth signal in the results of the two models, the first error weight the second error weight

[0124] S42. Take the reciprocal normalization of the average errors of the shallow buried range signal in the first preset neural network and the second preset neural network to obtain the first shallow buried weight ratio Q_l1 and the second shallow buried weight Q_l2;

[0125] Calculate the first average shallow buried error and the second average shallow buried error reciprocal and normalize these reciprocals to obtain the weight ratios of the shallow buried range signal in the results of the two models, the first shallow buried error weight the second shallow buried error weight

[0126] S43. Take the reciprocal normalization of the average errors of the deep buried range signal in the first preset neural network and the second preset neural network to obtain the first deep buried weight ratio Q_h1 and the second deep buried weight Q_h2;

[0127] Calculate the first average deep buried error and the second average deep buried error reciprocal and normalize these reciprocals to obtain the weight ratios of the deep buried range signal in the results of the two models, the first deep buried error weight, the second deep buried error weight

[0128] S44. Calculate the final burial depth according to the adaptive multi-scale weighting algorithm; the adaptive multi-scale weighting algorithm is as follows:

[0129]

[0130] where y is the final burial depth; thr is the preset burial depth threshold, thr = (d max -d min )÷2 + d min , thr is a relative number, determined according to the experimental design and having different values in different situations. Preferably, in this embodiment, thr is 90, with the unit of mm.

[0131] It can be seen from the results that the average value of the absolute value of the difference between the calculation results of all the collected data of the test blocks of the simulated tunnel segments on the ResNeXt depth prediction model and the actual burial depth is 4.782 mm. The average value of the absolute value of the difference between the calculation results of the y_l data on the ResNeXt depth prediction model and the actual burial depth is 3.863 mm. The average value of the absolute value of the difference between the calculation results of the y_h data on the ResNeXt depth prediction model and the actual burial depth is 5.701 mm. The average value of the absolute value of the difference between the calculation results of all the collected data of the test blocks on the ConvNeXt depth prediction model and the actual burial depth is 4.613 mm. The average value of the absolute value of the difference between the calculation results of the y_l data on the ConvNeXt depth prediction model and the actual burial depth is 5.131 mm. The average value of the absolute value of the difference between the calculation results of the y_h data on the ConvNeXt depth prediction model and the actual burial depth is 4.095 mm.

[0132] Table 1 Weight calculation

[0133] Data ResNeXt ConvNeXt Weight Q y_l 3.863 5.131 Q_l1 = 57.05%, Q_l2 = 42.95% Y_h 5.701 4.095 Q_h1 = 41.80%, Q_h2 = 58.20% All data 4.782 4.613 Q1 = 49.10%, Q2 = 50.90%

[0134] According to the setting of the hybrid network model, when testing the network subsequently, the weight ratios of ResNeXt and ConvNeXt are shown in Table 2. Randomly collect new data on the test blocks, process the data of the known defect depths that have not been used for model training, input them into the trained model for burial depth calculation, and then output the calculation results. It is obtained that most of the errors in the judgment of each burial depth are in the range of 1 mm to 4 mm, as shown in Table 2.

[0135] Table 2 Partial burial depth calculation errors

[0136]

[0137]

[0138] As can be seen from the data in the table, the re - collected depth data performs well on this hybrid model, with the average error being only 1.76 mm. Moreover, when compared with the prediction error values of each individual network model separately, the calculation results of the hybrid depth prediction model proposed in this paper are all better.

[0139] The method for detecting near - surface defects of tunnel segments based on a hybrid deep - learning network provided in this embodiment includes the following steps: S1. Use the impact - echo method to collect the near - surface defect burial depth signals of the tunnel segment test block; S2. Input the defect burial depth signals into the first preset neural network and the second preset neural network to obtain the first predicted burial depth and the second predicted burial depth; S3. Calculate the average error of the defect burial depth signals, shallow - burial range signals, and deep - burial range signals in the first preset neural network and the second preset neural network according to the average value of the differences between the first predicted burial depth, the second predicted burial depth and the actual burial depth; S4. Respectively perform reciprocal normalization on the average errors of the defect burial depth signals, shallow - burial range signals, and deep - burial range signals in the first preset neural network and the second preset neural network to obtain their weight ratios in the two neural networks, and then calculate and output the final burial depth according to the adaptive multi - scale weighting algorithm. This embodiment mainly focuses on using the impact - echo method based on an air - coupled ultrasonic array to generate and collect acoustic wave data, and converting the acoustic wave data into image data through a data - processing process, and using a hybrid neural network to calculate the burial depth of defects, so as to realize the detection of near - surface defects of tunnel segments and improve the accuracy and efficiency of tunnel segment detection.

[0140] In addition, the average value of the absolute value of the gap between the calculation result of the data collected in this embodiment on the hybrid depth prediction model and the actual burial depth is 1.76 mm, which can well locate the position of the near - surface defects of the tunnel segment and has guiding significance for actual engineering practice;

[0141] In addition, before model training in this embodiment, a series of standardized pre - processing steps were performed on the tunnel segment defect data images to ensure that each image received by the model has the same scale and distribution. Such processing helps the model to learn and generalize better, makes the training of the model smoother, and makes the model easier to converge.

[0142] In addition, this embodiment provides a method for manufacturing a tunnel segment test block. This method can simulate the actual situation of tunnel segments under controlled laboratory conditions, and allows researchers to precisely control the experimental conditions, the data is more controllable, and the production of laboratory test blocks can follow a standardized process, making the experimental results more comparable.

[0143] This specification is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the specification. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to produce a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices produce a device for implementing the functions specified in the Figure 1 one or more flows and / or blocks Figure 1 one block or multiple blocks.

[0144] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including an instruction device that implements the functions specified in the Figure 1 one or more flows and / or blocks Figure 1 one block or multiple blocks.

[0145] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in the Figure 1 one or more flows and / or blocks Figure 1 one block or multiple blocks.

[0146] The parts not detailed in the present invention are the prior art. For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention; therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting, and are intended to cover all changes falling within the meaning and scope of the equivalent elements within the present invention.

Claims

1. A tunnel segment near-surface defect detection method based on a hybrid deep learning network, characterized in that: The following steps are involved: S1. Use the impact echo method to collect the buried depth signal of the near-surface defects of the tunnel segment test block; Step S1 specifically includes: performing impact echo detection on the surface of the tunnel segment test block, and then collecting acoustic wave data through an air-coupled probe to obtain a defect burial depth signal; S2, inputting the defect burial depth signal into the first preset neural network and the second preset neural network to obtain the first predicted burial depth and the second predicted burial depth; the defect burial depth signal includes a shallow burial range signal and a deep burial range signal; the first predicted burial depth includes a first shallow burial predicted burial depth and a first deep burial predicted burial depth; the second predicted burial depth includes a second shallow burial predicted burial depth and a second deep burial predicted burial depth; the first preset neural network and the second preset neural network are different types of neural networks; S3, calculating the average error of the defect burial depth signal, the shallow burial range signal and the deep burial range signal in the first preset neural network and the second preset neural network according to the average value of the difference between the first predicted burial depth and the second predicted burial depth and the actual burial depth; S4, respectively, reciprocally normalize the error averages of the defect burial depth signal, the shallow burial range signal, and the deep burial range signal in the first preset neural network and the second preset neural network to obtain their weight ratios in the two neural networks, and then calculate and output the final burial depth according to the adaptive multi-scale weighted algorithm; Step S4 specifically includes the following steps: S41, reciprocally normalizing the error averages of the defect depth signal in the first preset neural network and the second preset neural network to obtain a first depth weight ratio Q1 and a second depth weight ratio Q2; S42, reciprocally normalizing the error averages of the shallow burial range signal in the first preset neural network and the second preset neural network to obtain a first shallow burial weight ratio Q_l1 and a second shallow burial weight ratio Q_l2; S43, reciprocally normalizing the error averages of the deep burial range signal in the first preset neural network and the second preset neural network to obtain a first deep burial weight ratio Q_h1 and a second deep burial weight ratio Q_h2; S44, calculating and outputting the final burial depth according to an adaptive multi-scale weighted algorithm; the adaptive multi-scale weighted algorithm is as follows: Where y is the final burial depth; thr is the preset burial depth threshold, thr = (d max -d min )÷2+d min , d min is the minimum burial depth; d max is the maximum burial depth.

2. The method according to claim 1, characterized in that: Step S3 specifically includes the following steps: S31, respectively calculating the difference between the first predicted burial depth y1 and the second predicted burial depth y2 and the actual burial depth, and then averaging the differences to obtain a first error average value and the second error mean S32, respectively calculate the difference between the first shallow buried predicted burial depth y_l1 and the second shallow buried predicted burial depth y_l2 and the actual burial depth, and then average the differences to obtain the first shallow buried error average value and the second shallow buried error average S33, respectively calculating the difference between the first deep burial predicted depth y_h1 and the second deep burial predicted depth y_h2 and the actual burial depth, and then averaging the differences to obtain the first deep burial error average value and the average value of the second deep buried error 3. The method according to claim 1, characterized in that Step S1: Defects of the same type and different burial depths are buried in the tunnel segment test block; the defects are filled with hard foam materials.

4. The method according to claim 1, characterized in that In step S2, the first preset neural network obtains the corresponding first predicted burial depth according to the input defect burial depth signal; the first preset neural network is trained by obtaining the first time-frequency image data according to the short-time Fourier transform of the defect burial depth signal.

5. The method according to claim 1, characterized in that In step S2, the first preset neural network is a ResNeXt model.

6. The method according to claim 1, characterized in that In step S2, the second preset neural network obtains the corresponding second predicted burial depth according to the input defect burial depth signal; the second preset neural network is trained by obtaining the second time-frequency image data according to the S transform of the defect burial depth signal.

7. The method according to claim 1, characterized in that In step S2, the second preset neural network is a ConvNeXt V2 model.

8. The method according to claim 1, characterized in that: Before the first preset neural network and the second preset neural network are trained, the method further includes: converting the first time-frequency image data and the second time-frequency image data into a tensor format and performing standardization processing through formula (1), which is as follows: Among them, normalizedpixel is the pixel value after normalization of the image data; pixel is the original pixel value in the image data; mean is the mean vector; std is the standard deviation vector.

Images