Method and device for diagnosing hollow defects of concrete-filled steel tube structures based on sound signals

Abstract

The present invention relates to a method and device for diagnosing hollow defects of steel tube concrete structures based on sound signals, the method comprising: 1) obtaining sound signals of steel tube concrete structures in different hollow states; 2) reading information in the sound signals, using CMOR wavelet basis functions for continuous wavelet transform, converting one-dimensional sound signals into two-dimensional wavelet time-frequency diagrams as image features for hollow defect recognition; 3) preprocessing the acquired image features for hollow defect recognition of steel tube concrete structures; 4) dividing the image features for hollow defect recognition into training sets, verification sets, and test sets according to a set ratio, and training a convolutional neural network model for hollow defect recognition of steel tube concrete structures; 5) inputting the image features of the steel tube concrete components to be detected into the trained convolutional neural network model to obtain detection results. The method and device are conducive to accurate and efficient automatic diagnosis of hollow defects of steel tube concrete structures.

Description

Technical Field

[0001] The invention belongs to the technical field of structure detection, and in particular relates to a method and a device for diagnosing hollow defects of a steel tube concrete structure based on sound signals. Background Art

[0002] Steel tube concrete structures have excellent mechanical properties because they can play the "combined effect" of steel tubes and core concrete. They are widely used in large-scale projects at home and abroad. However, due to objective factors such as construction technology, material properties, and service environment, there will be hollow defects inside steel tube concrete structures. Since hollow defects will weaken or even destroy the "combined effect", the ultimate bearing capacity, stiffness, ductility and other mechanical performance indicators of steel tube concrete structures will be reduced, thereby causing structural damage and diseases. In severe cases, it may cause structural collapse, resulting in significant economic losses and serious casualties.

[0003] The premise for scientific control of void defects is accurate detection and identification. The existing steel tube concrete void defect detection technologies, such as manual tapping method, core sampling method, ultrasonic method, piezoelectric ceramic method, etc., have shortcomings such as large errors based on experience, need to be damaged, low efficiency, and need to embed sensors. In particular, the manual tapping detection method commonly used in the engineering community uses human ears to screen and judge whether the structure has voids by tapping the vibration sound of steel tube concrete. There is a problem of misjudgment or missed judgment due to human factors such as lack of experience. At the same time, for void defects inside the steel pipe, the sound signal generated by tapping the outer steel pipe may be a reflection process of steel pipe-air-concrete-steel pipe, or it may be directly collected by the sensor after diffraction along the steel pipe. Therefore, the sound signal is very complex. How to accurately extract the characteristics of such sound signals is the key to achieving accurate diagnosis. Summary of the invention

[0004] The purpose of the present invention is to provide a method and device for diagnosing hollow defects in steel tube concrete structures based on sound signals, which method and device are conducive to accurate and efficient automatic diagnosis of hollow defects in steel tube concrete structures.

[0005] To achieve the above object, the technical solution adopted by the present invention is: a method for diagnosing hollow defects of concrete-filled steel tube structures based on sound signals, comprising the following steps:

[0006] 1) Obtain the sound signals of the steel tube concrete structure under different voiding states;

[0007] 2) Read the information in the sound signal, use the CMOR wavelet basis function to perform continuous wavelet transform, and convert the one-dimensional sound signal into a two-dimensional wavelet time-frequency diagram as the image feature for hollow defect recognition;

[0008] 3) Preprocessing the image features obtained for identifying hollow defects in steel tube concrete structures;

[0009] 4) Dividing the image features for hollow defect recognition into a training set, a validation set, and a test set according to a set ratio, and training a convolutional neural network model for hollow defect recognition in steel tube concrete structures;

[0010] 5) Input the image features of the steel tube concrete component to be detected into the trained convolutional neural network model to obtain the detection results.

[0011] Furthermore, in the step 1), an acoustic sensor is arranged at a set distance above the surface of the steel tube concrete structure, and the surface of the steel tube concrete is knocked, and the collection device collects the sound signal through the acoustic sensor; the sound signal includes the sound signal collected through the reflection process of steel tube-air-concrete-steel tube and the sound signal collected after direct diffraction along the steel tube.

[0012] Furthermore, in step 2), the sound signal x(t)∈L 2 (R) The calculation formula for continuous wavelet transform is as follows:

[0013]

[0014] Where x(t) is the collected sound signal; a is the scale factor, which represents the frequency-related expansion and contraction; The family function that represents the displacement and expansion of the base wavelet ψ(t) is called the wavelet basis function. The steps are as follows: first determine the wavelet basis and scale, and use cwt() to calculate the wavelet coefficient coefs, then use scal2frq() to convert the scale sequence into the actual frequency sequence f, and finally combine the time series t, imagesc(t,f,abs(coefs)) to draw the wavelet time-frequency diagram;

[0015] The CMOR wavelet is used as the basic wavelet and is defined as:

[0016]

[0017] Where t represents time, i is an imaginary number, Fb is the bandwidth factor, and Fc is the center frequency factor.

[0018] Furthermore, in the step 3), the preprocessing includes: removing the peripheral coordinates, text and energy bars of the time-frequency diagram of the image features, and normalizing and compressing the grid into a pixel format.

[0019] Furthermore, in step 4), the two-dimensional wavelet time-frequency diagram and the corresponding void defect label are taken as a sample pair, and all samples are divided into a training set, a validation set and a test set according to a set ratio.

[0020] Furthermore, in the step 4), a convolutional neural network model for identifying hollow defects in steel tube concrete structures is trained, the activation function adopts the ReLU function, the loss function adopts the cross entropy function, and the output layer adopts the softmax function to complete the model classification. The convolutional neural network model sequentially includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a fourth convolutional layer, a fourth pooling layer, a first fully connected layer, a second fully connected layer and a third fully connected layer.

[0021] Furthermore, the first convolution layer uses 64 3×3 convolution kernels with a step size of 1 to convolve the 128×128×3 pixel image to obtain a 128×128×64 pixel feature map, performs ReLU transformation on the data after the convolution operation, and then performs pooling, and the feature map size is halved to an image of 64×64×64 pixels;

[0022] The second convolutional layer uses 128 3×3 convolution kernels to convolve the pooled 64×64×64 pixel feature map to obtain a 64×64×128 pixel feature map. The convolution operation data is subjected to ReLU transformation, and then after pooling, the feature map size is halved to an image of 32×32×128 pixels.

[0023] The third convolutional layer uses 256 3×3 convolution kernels to convolve the pooled 32×32×128 pixel feature map to obtain a 32×32×256 pixel feature map, performs ReLU transformation on the convolution data, and then performs pooling, so that the feature map size is halved to an image of 16×16×256 pixels;

[0024] The fourth convolutional layer uses 512 3×3 convolution kernels to convolve the pooled 16×16×256 pixel feature map to obtain a 16×16×512 pixel feature map, performs ReLU transformation on the data after the convolution operation, and then after pooling, the feature map size is halved to an image of 8×8×512 pixels;

[0025] The number of neurons in the first fully connected layer is 1024, the number of neurons in the second fully connected layer is 512, and the number of neurons in the third fully connected layer is 11.

[0026] The present invention also provides a sound signal-based steel tube concrete structure hollow defect diagnosis device for implementing the above method, comprising:

[0027] Sound perception module, used to obtain sound signals of steel tube concrete structures in different void states;

[0028] The feature acquisition module is used to read the information in the sound signal, use the CMOR wavelet basis function to perform continuous wavelet transform, and convert the one-dimensional sound signal into a two-dimensional wavelet time-frequency diagram as the image feature for void defect recognition; it is also used to pre-process the image features obtained for void defect recognition of steel tube concrete structure;

[0029] A training module is used to divide the image features for void defect recognition into a training set, a validation set, and a test set according to a set ratio, and train a convolutional neural network model for void defect recognition in steel tube concrete structures;

[0030] A diagnosis module, which is used to input image features of the steel tube concrete component to be detected, and perform detection through the trained convolutional neural network model to obtain a detection result; and

[0031] The display module is used to display the detection results, that is, the probability of hollow defects in the steel tube concrete structure.

[0032] Compared with the prior art, the present invention has the following beneficial effects: the present invention aims at the sound signal characteristics of the steel tube concrete hollowing defect, and based on the perceived hollowing sound signal, uses the continuous wavelet transform method to convert the one-dimensional time domain signal into a time-frequency diagram consisting of the time domain and the frequency domain, and extracts the image features to reduce the complexity of the noise and signal components of the sound signal; the time-frequency diagram is input into the steel tube concrete hollowing defect diagnosis model obtained by pre-training through the convolutional neural network, and finally the hollowing defect working condition classification is completed by using the softmax classifier to obtain the steel tube concrete hollowing defect recognition result, which solves the problem that the prior art uses the human ear to screen and distinguish the steel tube concrete knocking vibration sound, which is affected by human factors and leads to misjudgment or missed judgment. The present invention uses acoustic collection and acoustic diagnosis to collect and analyze the steel tube concrete knocking vibration sound, so as to realize the accurate and efficient automatic recognition and judgment of the hollowing defect of the steel tube concrete structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 is a flow chart of a method implementation of an embodiment of the present invention;

[0034] Figure 2 is a diagram of a steel tube concrete component according to an embodiment of the present invention;

[0035] Figure 3 is a sound signal diagram of non-empty and empty in an embodiment of the present invention;

[0036] Figure 4 is a schematic diagram of the pretreatment process in an embodiment of the present invention;

[0037] Figure 5 Schematic diagram of a convolutional neural network model in an embodiment of the present invention;

[0038] Figure 6 It is a wavelet time-frequency diagram of non-empty and empty in the embodiment of the present invention. DETAILED DESCRIPTION

[0039] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0040] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0041] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0042] like Figure 1 As shown, this embodiment provides a method for diagnosing hollow defects in steel tube concrete structures based on sound signals, comprising the following steps:

[0043] 1) Obtain the sound signals of steel tube concrete structures under different voiding states.

[0044] In the step 1), an acoustic sensor is arranged at a set distance (20 mm in this embodiment) above the surface of the steel tube concrete structure, and the surface of the steel tube concrete is knocked, and the collection device collects sound signals through the acoustic sensor. Among them, for the hollow defect inside the steel tube, the sound signal generated by knocking outside the steel tube may be collected through the reflection process of steel tube-air-concrete-steel tube, or it may be directly collected by the sensor after diffraction along the steel tube, so the sound signal is very complex.

[0045] In this embodiment, one of the steel tube concrete components is as follows Figure 2 As shown. Among them, a total of 4 steel tube concrete components were tested, including 2 square steel tube concrete components and 2 circular steel tube concrete specimens, of which 1 was a non-hollow component and 1 was a crowned hollow component. The debonding positions were all local, the total length of the component L = 1200mm, the diameter of the circular steel tube concrete component D = 150mm, the side length of the square steel tube concrete B = 150mm, the local hollow range was 80mm, the hollow value of the circular steel tube concrete component ds = 8mm, and the hollow value of the square steel tube concrete component dt = 8mm.

[0046] In this embodiment, a total of 6 different working condition data are collected. The sound signal diagram of non-empty and empty in this embodiment is as follows Figure 3 shown.

[0047] 2) Read the information in the sound signal, use the CMOR wavelet basis function to perform continuous wavelet transform, and convert the one-dimensional sound signal into a two-dimensional wavelet time-frequency diagram as the image feature for hollow defect recognition.

[0048] Due to the complexity of sound signals, accurately extracting the features of such sound signals is the key to achieving accurate diagnosis. Based on a large number of experiments, various feature extraction methods, such as Mel frequency feature matrix, zero crossing rate, etc., were tried and compared, and finally the wavelet theory was adopted.

[0049] The method of converting sound signals into wavelet time-frequency diagrams can accurately reflect the characteristics of sound signals of steel tube concrete void defects, and the wavelet threshold has a certain noise reduction effect. The data set constructed based on this can be effectively used for deep neural network discrimination.

[0050] In this embodiment, the implementation process of continuous wavelet transform is as follows:

[0051] S21. First, you need to select a suitable wavelet basis function and fix a scale factor.

[0052] In this embodiment, the basic wavelet adopts CMOR wavelet, which is the abbreviation of complex wavelet Morlet and is defined as:

[0053]

[0054] Where t represents time, i is an imaginary number, Fb is the bandwidth factor, and Fc is the center frequency factor. The CMOR wavelet can adjust the accuracy of time-frequency analysis by changing Fb and Fc. The time domain and frequency domain forms of the CMOR wavelet have certain properties of the Gaussian function, have good resolution, and reflect good time-frequency aggregation on the time-frequency diagram, and can adaptively decompose the target signal.

[0055] In this embodiment, a fixed scale sequence length is 256.

[0056] S22. Calculate the wavelet coefficients, which reflect the similarity between the wavelet at the current scale and the corresponding signal segment.

[0057] Among them, according to the determined wavelet basis function and scale sequence, the wavelet coefficient coefs is calculated using cwt(). Then scal2frq() is used to convert the scale sequence into the actual frequency sequence f, and finally combined with the time series t, imagesc(t,f,abs(coefs)) is used to draw the wavelet time-frequency diagram.

[0058] S23. Change the translation factor to make the wavelet produce a certain displacement along the time axis, repeat the above steps to complete a transformation.

[0059] S24. Increase the scale factor and repeat the above three steps to complete the second analysis.

[0060] S25. Execute the above steps repeatedly until the analysis requirements are met.

[0061] Specifically, after the wavelet basis function ψ(t) is shifted by b, it is inner-producted with the original sound signal x(t) at different scales a to complete the wavelet transform process, which can be expressed as follows:

[0062]

[0063] Wherein, a represents a scale factor and a>0, which is used to scale the wavelet basis function; b represents a translation factor, which reflects the displacement change of the wavelet basis function ψ(t), and its value can be positive or negative; in the embodiment of the present invention, a and b are both set as continuous variables, so it is called continuous wavelet transform (CWT).

[0064] 3) Preprocess the image features obtained for identifying void defects in steel tube concrete structures.

[0065] In step 3), the preprocessing includes: removing the peripheral coordinates, text and energy bars of the time-frequency graph from the image features, and normalizing and compressing the grid into a pixel format. Figure 4 shown.

[0066] 4) The image features for void defect recognition are divided into training set, validation set, and test set according to a certain ratio, and a convolutional neural network model for void defect recognition in steel tube concrete structure is trained.

[0067] Specifically, the two-dimensional wavelet time-frequency diagram and the corresponding void defect label are taken as a sample pair, and all samples are divided into a training set, a validation set, and a test set according to a set ratio.

[0068] In step 4), a convolutional neural network model for identifying hollow defects in steel tube concrete structures is trained, the activation function uses the ReLU function, the loss function uses the cross entropy function, and the output layer uses the softmax function to complete the model classification. Figure 5 As shown, the convolutional neural network model includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a fourth convolutional layer, a fourth pooling layer, a first fully connected layer, a second fully connected layer and a third fully connected layer in sequence.

[0069] The first convolutional layer uses 64 3×3 convolution kernels with a step size of 1 to convolve the 128×128×3 pixel image to obtain a 128×128×64 pixel feature map. The data after the convolution operation is ReLU transformed and then pooled. The size of the feature map is halved to an image of 64×64×64 pixels.

[0070] The second convolutional layer uses 128 3×3 convolution kernels to convolve the pooled 64×64×64 pixel feature map to obtain a 64×64×128 pixel feature map. The convolution operation data is ReLU transformed, and then after pooling, the feature map size is halved to an image of 32×32×128 pixels.

[0071] The third convolutional layer uses 256 3×3 convolution kernels to convolve the pooled 32×32×128 pixel feature map to obtain a 32×32×256 pixel feature map, performs ReLU transformation on the data after the convolution operation, and then after pooling, the feature map size is halved to an image of 16×16×256 pixels.

[0072] The fourth convolutional layer uses 512 3×3 convolution kernels to convolve the pooled 16×16×256 pixel feature map to obtain a 16×16×512 pixel feature map, performs ReLU transformation on the data after the convolution operation, and then after pooling, the feature map size is halved to an image of 8×8×512 pixels.

[0073] The number of neurons in the first fully connected layer is 1024, the number of neurons in the second fully connected layer is 512, and the number of neurons in the third fully connected layer is 11.

[0074] In this embodiment, each working condition includes 800 samples, and there are 4800 samples in total. The time-frequency graph data of the six working conditions are labeled and divided into a training set, a validation set, and a test set in a ratio of 7:2:1. Figure 6 It is the wavelet time-frequency diagram of non-empty and empty.

[0075] The training and validation samples are input into the network for forward propagation, and the predicted value is obtained after passing through the classification layer. The loss value is back-propagated through the optimization algorithm to update the network parameters. Among them, the hyperparameters of the model are set as follows: the learning rate is set to 0.002; the number of training rounds is 60; and the optimization algorithm selects the Adam optimization algorithm.

[0076] In this embodiment, the model is composed of four convolutional layers, four pooling layers and three fully connected layers. After the training and validation sets are used to adjust the network model, the model parameters have been iterated to the optimal level, and the feasibility of the entire model is verified by using the test set to remove defective samples.

[0077] The test set is input into the optimal model to obtain the diagnosis result.

[0078] 5) Input the image features of the steel tube concrete component to be detected into the trained convolutional neural network model to obtain the detection results.

[0079] This embodiment also provides a sound signal-based steel tube concrete structure void defect diagnosis device for implementing the above method, including: a sound perception module, a feature acquisition module, a training module, a diagnosis module and a display module.

[0080] The sound perception module is used to obtain the sound signals of the steel tube concrete structure in different emptying states;

[0081] The feature acquisition module is used to read the information in the sound signal, and uses the CMOR wavelet basis function to perform continuous wavelet transform to convert the one-dimensional sound signal into a two-dimensional wavelet time-frequency diagram as the image feature for void defect recognition; it is also used to pre-process the image features obtained for void defect recognition of steel tube concrete structure;

[0082] The training module is used to divide the image features of hollow defect recognition into a training set, a validation set, and a test set according to a set ratio, and train a convolutional neural network model for hollow defect recognition of steel tube concrete structure;

[0083] The diagnosis module is used to input the image features of the steel tube concrete component to be detected, and perform detection through the trained convolutional neural network model to obtain the detection result; and

[0084] The display module is used to display the detection results, that is, the probability of hollow defects in the steel tube concrete structure.

[0085] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0086] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0087] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0088] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0089] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A method for diagnosing hollow defects in steel tube concrete structures based on sound signals, characterized in that: The following steps are involved: 1) Obtain the sound signals of the steel tube concrete structure under different voiding states; 2) Read the information in the sound signal, use the CMOR wavelet basis function to perform continuous wavelet transform, and convert the one-dimensional sound signal into a two-dimensional wavelet time-frequency diagram as the image feature for hollow defect recognition; 3) Preprocessing the image features obtained for identifying hollow defects in steel tube concrete structures; 4) Dividing the image features for hollow defect recognition into a training set, a validation set, and a test set according to a set ratio, and training a convolutional neural network model for hollow defect recognition in steel tube concrete structures; 5) Input the image features of the steel tube concrete component to be detected into the trained convolutional neural network model to obtain the detection results; In the step 4), a convolutional neural network model for identifying hollow defects in steel tube concrete structures is trained, the activation function adopts the ReLU function, the loss function adopts the cross entropy function, and the output layer adopts the softmax function to complete the model classification, and the convolutional neural network model sequentially includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a fourth convolutional layer, a fourth pooling layer, a first fully connected layer, a second fully connected layer and a third fully connected layer; The first convolutional layer uses 64 3×3 convolution kernels with a step size of 1 to convolve the 128×128×3 pixel image to obtain a 128×128×64 pixel feature map. The convolution operation data is subjected to ReLU transformation and then pooling. The feature map size is halved to a 64×64×64 pixel image. The second convolutional layer uses 128 3×3 convolution kernels to convolve the pooled 64×64×64 pixel feature map to obtain a 64×64×128 pixel feature map. The convolution operation data is subjected to ReLU transformation, and then after pooling, the feature map size is halved to an image of 32×32×128 pixels. The third convolutional layer uses 256 3×3 convolution kernels to convolve the pooled 32×32×128 pixel feature map to obtain a 32×32×256 pixel feature map, performs ReLU transformation on the convolution data, and then performs pooling, so that the feature map size is halved to an image of 16×16×256 pixels; The fourth convolutional layer uses 512 3×3 convolution kernels to convolve the pooled 16×16×256 pixel feature map to obtain a 16×16×512 pixel feature map, performs ReLU transformation on the data after the convolution operation, and then after pooling, the feature map size is halved to an image of 8×8×512 pixels; The number of neurons in the first fully connected layer is 1024, the number of neurons in the second fully connected layer is 512, and the number of neurons in the third fully connected layer is 11.

2. The method for diagnosing hollow defects in concrete-filled steel tube structures based on sound signals according to claim 1 is characterized in that: In the step 1), an acoustic sensor is arranged at a set distance above the surface of the steel tube concrete structure, and the surface of the steel tube concrete is knocked, and the collection device collects the sound signal through the acoustic sensor; the sound signal includes the sound signal collected through the reflection process of steel tube-air-concrete-steel tube and the sound signal collected after direct diffraction along the steel tube.

3. The method for diagnosing hollow defects in concrete-filled steel tube structures based on sound signals according to claim 1 is characterized in that: In step 2), the calculation formula for continuous wavelet transform of the sound signal x(t) is as follows: Among them, x(t) is the collected sound signal; a is the scale factor, which represents the frequency-related expansion and contraction; b is the translation factor, which reflects the displacement change of the wavelet basis function; The family function that represents the base wavelet ψ(t) after displacement and expansion is called the wavelet basis function; The CMOR wavelet is used as the basic wavelet and is defined as: Where t represents time, i is an imaginary number, Fb is the bandwidth factor, and Fc is the center frequency factor.

4. The method for diagnosing hollow defects in concrete-filled steel tube structures based on sound signals according to claim 1 is characterized in that: In the step 3), the preprocessing includes: removing the peripheral coordinates, text and energy bars of the time-frequency diagram of the image features, and normalizing and compressing the grid into a pixel format.

5. The method for diagnosing hollow defects in concrete-filled steel tube structures based on sound signals according to claim 1 is characterized in that: In step 4), the two-dimensional wavelet time-frequency diagram and the corresponding void defect label are taken as a sample pair, and all samples are divided into a training set, a validation set, and a test set according to a set ratio.

6. A device for diagnosing hollowing defects of concrete-filled steel tube structures based on sound signals for implementing the method according to any one of claims 1 to 5, characterized in that: include: Sound perception module, used to obtain sound signals of steel tube concrete structures in different void states; The feature acquisition module is used to read the information in the sound signal, use the CMOR wavelet basis function to perform continuous wavelet transform, and convert the one-dimensional sound signal into a two-dimensional wavelet time-frequency diagram as the image feature for hollow defect recognition; It is also used to pre-process the image features acquired for identifying hollow defects in steel tube concrete structures; A training module is used to divide the image features for void defect recognition into a training set, a validation set, and a test set according to a set ratio, and train a convolutional neural network model for void defect recognition in steel tube concrete structures; The diagnosis module is used to input the image features of the steel tube concrete component to be detected, and perform detection through the trained convolutional neural network model to obtain the detection result; as well as The display module is used to display the detection results, that is, the probability of hollow defects in the steel tube concrete structure.

Images