CN117805247A

CN117805247A - Ultrasonic detection method and system for concrete defects

Info

Publication number: CN117805247A
Application number: CN202311863624.1A
Authority: CN
Inventors: 罗健; 卢晋荣; 邱先炎; 杨华翰; 黄富军
Original assignee: Guangdong Rongchuang High Tech Testing And Identification Co ltd
Current assignee: Guangdong Rongchuang High Tech Testing And Identification Co ltd
Priority date: 2023-12-29
Filing date: 2023-12-29
Publication date: 2024-04-02
Anticipated expiration: 2043-12-29
Also published as: CN117805247B

Abstract

The invention discloses a method and a system for ultrasonically detecting concrete defects, comprising the following steps: providing an ultrasonic phased array on the surface of the concrete to disperse the concrete into points; acquiring echo signals through an ultrasonic phased array, wherein the ultrasonic phased array comprises the steps of transmitting an excitation signal coded by a Huffman code through a transmitting end excitation transducer, receiving a corresponding impulse echo through a receiving end excitation transducer, and carrying out pulse compression on the impulse echo according to matched filtering to generate echo signals; and (3) completing noise reduction of wavelet decomposition of the echo signals, carrying out phased focusing imaging on each point in the discrete concrete defect by combining an array element plane to obtain images and position information of each point in the defect, and finally predicting the concrete strength by combining a preset strength prediction model so as to match unused use strategies according to different strengths. The method can rapidly and accurately detect the defects of the concrete, scientifically guide the use of the concrete according to the predicted strength of the defects, and improve the detection precision of products and the utilization rate of resources.

Description

Ultrasonic detection method and system for concrete defects

Technical Field

The invention relates to the technical field of concrete ultrasonic detection, in particular to a concrete defect ultrasonic detection method and system.

Background

Nondestructive testing is widely used in industrial materials as a method of detecting no damage to the material. The presence of defects in concrete can pose a potentially enormous threat to engineering. Common methods of ultrasonic non-destructive testing include ultrasonic rebound, pulse echo, and probe radar techniques. Although the ground penetrating radar can be used for single-sided test, electromagnetic waves are emitted by the ground penetrating radar, and the test result is greatly influenced by internal accessories. The pulse echo method is a traditional detection method, the principle is simple and the operation is simple and convenient, but the detection precision is generally not guaranteed. Ultrasonic rebound mainly combines an ultrasonic instrument and a rebound instrument, and reflects the state of a concrete surface layer and the condition of an internal structure by distinguishing an acoustic value and a rebound value. However, the method has high requirements on equipment and operation, is low in efficiency, is easy to be interfered by the outside to generate larger errors during testing, and has the problem of low measurement precision.

Disclosure of Invention

In order to solve at least one technical problem set forth above, the invention provides a method and a system for ultrasonic detection of concrete defects.

In a first aspect, the invention provides a method for ultrasonically detecting a concrete defect, the method comprising:

Setting an ultrasonic phased array on the surface of the concrete, and dispersing the concrete into points based on the ultrasonic phased array; the ultrasonic phased array comprises a plurality of transmitting-end excitation transducer array elements and receiving-end excitation transducer array elements;

acquiring echo signals through the ultrasonic phased array, wherein the ultrasonic phased array comprises the steps of transmitting excitation signals coded by Huffman codes through the transmitting-end excitation transducer, receiving corresponding impulse echoes through the receiving-end excitation transducer, and carrying out pulse compression on the impulse echoes according to matched filtering to generate echo signals;

performing wavelet decomposition on the echo signals, applying a threshold value to wavelet decomposition coefficients, removing high-frequency noise components, and then performing wavelet signal reconstruction to obtain noise-reduced echo signals, wherein the method comprises the steps of selecting proper wavelet bases, determining wavelet decomposition layers, performing wavelet decomposition on the signals, and obtaining corresponding wavelet decomposition coefficients; selecting proper threshold values and threshold value functions to perform threshold value quantization processing on the high-frequency coefficients under each decomposition scale to obtain new high-frequency coefficients; carrying out wavelet reconstruction of one-dimensional signals according to the lowest-layer low-frequency coefficient of wavelet decomposition and each layer high-frequency coefficient subjected to threshold quantization processing to obtain echo signals after noise reduction;

Performing phased focusing imaging on each point in the discrete concrete defect according to the echo signal after noise reduction and the selected array element plane to obtain images and position information of each point in the defect;

inputting the images and the position information of each point in the defect into a preset strength prediction model, and outputting a strength prediction result of the concrete; wherein training the intensity prediction model comprises:

obtaining images and position information of points in a plurality of defects in a concrete historical defect identification result, and taking the images and the position information as a training set;

preprocessing a training set to generate input data;

inputting the input data into a convolution layer of a CNN network to extract a feature vector;

inputting the feature vectors to an Attention mechanism layer of a CNN (computer numerical network), calculating probability weights of different feature vectors, and outputting weighted results of the feature vectors and the probability weights;

inputting the weighted results of the feature vectors and the probability weights to a full-connection layer of a CNN network, mapping the weighted feature vectors, and outputting predicted intensity;

verifying the precision of the predicted intensity by using a robustness cross verification method until the precision meets the preset requirement, and generating an intensity prediction model;

And matching the use strategy of the concrete according to the strength prediction result.

Preferably, the disposing an ultrasonic phased array on the concrete surface comprises:

establishing a loss function, wherein parameters to be optimized in the loss function at least comprise position parameters of each array element in an ultrasonic phased array, and the value of the loss function is obtained by fusing side lobe grades corresponding to a plurality of beam directions of the ultrasonic phased array;

determining a target to-be-optimized parameter that minimizes the value of the loss function;

and carrying out array element layout on the ultrasonic phased array by utilizing the target parameters to be optimized.

Preferably, acquiring echo signals by the ultrasonic phased array further comprises:

dividing the concrete area to generate a plurality of subareas;

setting an ultrasonic phased array in a sub-area, in the ultrasonic phased array, taking a central ring array element positioned in the middle of an outer ring array element and an inner ring array element as a transmitting end, taking the outer ring array element and the inner ring array element as a receiving end, transmitting an excitation signal coded by Huffman codes by the transmitting end, diffusing the excitation signal to the outer ring array element and the inner ring array element, and collecting echo signals of the receiving end in the current sub-area;

Echo signals of other sub-regions are acquired by translating the ultrasound phased array.

Preferably, before acquiring echo signals of other sub-regions by translating the ultrasound phased array, determining an optimal excitation frequency of the excitation signal includes:

acquiring diffusion coefficients and dissipation coefficients under different wave bands based on echo signals of receiving ends in the current subarea;

normalizing the diffusion coefficient and the dissipation coefficient, and fitting the diffusion coefficient and the dissipation coefficient respectively to generate a diffusion coefficient curve and a dissipation coefficient curve;

and obtaining a first frequency corresponding to the maximum slope of the diffusion coefficient curve and a second frequency corresponding to the maximum slope of the dissipation coefficient curve, and obtaining the optimal excitation frequency by weighted average of the first frequency and the second frequency.

Preferably, a 6 th order Daubechies wavelet function is used as a basis function of wavelet decomposition, and an AdaRound algorithm is used to determine a threshold of the adaptive scale.

Preferably, the preprocessing the training set to generate input data includes:

performing feature dimension reduction on the training set by using an SAE algorithm;

fitting the dimension-reduced training set by using a random forest algorithm and a self-adaptive integration algorithm;

Screening out the feature quantity with the maximum correlation according to the fitted comprehensive index;

and fusing the characteristic quantity with the images and the position information of each point in the defect to serve as input data of the CNN network.

Preferably, the inputting the feature vector to the Attention mechanism layer of the CNN network calculates probability weights of different feature vectors, including:

training by adopting a plurality of parallel base classifiers in an attribute mechanism layer to obtain classification errors of the base classifiers;

updating the weight of each base classifier according to the classification error of the base classifier;

and determining probability weights of different feature vectors through the updated base classifier weights.

In a second aspect, the present invention also provides a system for ultrasonic detection of defects in concrete, said system comprising:

the excitation source distribution control unit is used for setting an ultrasonic phased array on the surface of the concrete and dispersing the concrete into points based on the ultrasonic phased array; the ultrasonic phased array comprises a plurality of transmitting-end excitation transducer array elements and receiving-end excitation transducer array elements;

the echo signal acquisition unit is used for acquiring echo signals through the ultrasonic phased array, and comprises the steps of transmitting excitation signals coded by Huffman codes through the transmitting-end excitation transducer, receiving corresponding impulse echoes through the receiving-end excitation transducer, and carrying out pulse compression on the impulse echoes according to matched filtering to generate echo signals;

The wavelet decomposition unit is used for carrying out wavelet decomposition on the echo signals and acting a threshold on wavelet decomposition coefficients, carrying out wavelet signal reconstruction after removing high-frequency noise components to obtain echo signals after noise reduction, and carrying out wavelet decomposition on the signals after selecting proper wavelet bases and determining the layers of the wavelet decomposition to obtain corresponding wavelet decomposition coefficients; selecting proper threshold values and threshold value functions to perform threshold value quantization processing on the high-frequency coefficients under each decomposition scale to obtain new high-frequency coefficients; carrying out wavelet reconstruction of one-dimensional signals according to the lowest-layer low-frequency coefficient of wavelet decomposition and each layer high-frequency coefficient subjected to threshold quantization processing to obtain echo signals after noise reduction;

the defect identification unit is used for carrying out phased focusing imaging on each point in the discrete concrete defect according to the echo signal after noise reduction and the selected array element plane to obtain images and position information of each point in the defect;

the strength prediction unit is used for inputting the images and the position information of each point in the defect into a preset strength prediction model and outputting a strength prediction result of the concrete; wherein training the intensity prediction model comprises: obtaining images and position information of points in a plurality of defects in a concrete historical defect identification result, and taking the images and the position information as a training set; preprocessing a training set to generate input data; inputting the input data into a convolution layer of a CNN network to extract a feature vector; inputting the feature vectors to an Attention mechanism layer of a CNN (computer numerical network), calculating probability weights of different feature vectors, and outputting weighted results of the feature vectors and the probability weights; inputting the weighted results of the feature vectors and the probability weights to a full-connection layer of a CNN network, mapping the weighted feature vectors, and outputting predicted intensity; verifying the precision of the predicted intensity by using a robustness cross verification method until the precision meets the preset requirement, and generating an intensity prediction model;

And the strategy matching unit is used for matching the use strategy of the concrete according to the strength prediction result.

In a third aspect, the present invention also provides an electronic device, including: a processor and a memory for storing computer program code comprising computer instructions which, when executed by the processor, cause the electronic device to perform a method as described in the first aspect and any one of its possible implementation manners.

In a fourth aspect, the present invention also provides a computer readable storage medium having stored therein a computer program comprising program instructions which, when executed by a processor of an electronic device, cause the processor to perform a method as in the first aspect and any one of the possible implementations thereof.

Compared with the prior art, the invention has the beneficial effects that:

1) By adopting the ultrasonic phased array technology, the signal emission time can be controlled to form an integral wave front, so that the beam scanning, deflection and focusing of ultrasonic waves are realized, and the accuracy of detecting the internal defects of the concrete is improved. When the ultrasonic phased array is controlled, the transmission of the excitation signals can be accelerated by adopting the excitation signals coded by the Huffman codes, so that the test efficiency is improved; when receiving the echo signal, the wavelet decomposition is carried out, the threshold value is acted on the wavelet decomposition coefficient, the wavelet signal reconstruction is carried out after the high-frequency noise component is removed, and the stronger noise reduction effect is achieved, so that the quality of the echo signal is enhanced; finally, phase control focusing imaging is carried out on each point in the discrete concrete defect through the echo signal and the selected array element plane, so that accurate positioning of the concrete internal defect can be realized, and the image and the position information of each point in the defect can be obtained.

2) The concrete strength is predicted according to the images and the position information of each point in the defect, and different using strategies are matched according to different strengths, so that the using mode of the concrete can be scientifically guided, and the utilization rate of resources is improved.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

Drawings

In order to more clearly describe the embodiments of the present invention or the technical solutions in the background art, the following description will describe the drawings that are required to be used in the embodiments of the present invention or the background art.

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the disclosure and together with the description, serve to explain the technical aspects of the disclosure.

FIG. 1 is a schematic flow chart of a method for ultrasonic detection of concrete defects according to an embodiment of the present invention;

fig. 2 is a schematic structural diagram of an ultrasonic phased array according to an embodiment of the present invention;

FIG. 3 is a schematic flow chart of a training intensity prediction model according to an embodiment of the present invention;

FIG. 4 is a schematic diagram of an Attention mechanism layer according to an embodiment of the present invention;

Fig. 5 is a schematic diagram of a data transmission manner between hidden layers in a convolutional neural network structure according to an embodiment of the present invention;

FIG. 6 is a schematic structural diagram of an ultrasonic detection system for concrete defects according to an embodiment of the present invention;

fig. 7 is a schematic hardware structure of an electronic device according to an embodiment of the present invention.

Detailed Description

In order that those skilled in the art will better understand the present invention, a technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

The terms first, second and the like in the description and in the claims and in the above-described figures are used for distinguishing between different objects and not necessarily for describing a sequential or chronological order. Furthermore, the terms "comprise" and "have," as well as any variations thereof, are intended to cover a non-exclusive inclusion. For example, a process, method, system, article, or apparatus that comprises a list of steps or elements is not limited to only those listed steps or elements but may include other steps or elements not listed or inherent to such process, method, article, or apparatus.

The term "and/or" is herein merely an association relationship describing an associated object, meaning that there may be three relationships, e.g., a and/or B, may represent: a exists alone, A and B exist together, and B exists alone. In addition, the term "at least one" herein means any one of a plurality or any combination of at least two of a plurality, for example, including at least one of A, B, C, and may mean including any one or more elements selected from the group consisting of A, B and C.

Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the invention. The appearances of such phrases in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those of skill in the art will explicitly and implicitly appreciate that the embodiments described herein may be combined with other embodiments.

Furthermore, in the following detailed description, numerous specific details are set forth in order to provide a better illustration of the invention. It will be understood by those skilled in the art that the present invention may be practiced without some of these specific details. In some instances, well known methods, procedures, components, and circuits have not been described in detail so as not to obscure the present invention.

At present, the nondestructive testing technology for concrete defect detection generally has the problems of low testing precision and low efficiency. Therefore, the ultrasonic phased array and the Huffman code encoded excitation signal are combined to quickly transmit ultrasonic waves and receive echoes, and the echo is subjected to noise reduction treatment and phase control focusing imaging, so that an internal defect image can be accurately obtained and the position of the defect can be accurately positioned, and the concrete defect can be quickly and accurately detected.

Referring to fig. 1, fig. 1 is a schematic flow chart of a method for ultrasonic detection of a concrete defect according to an embodiment of the invention.

S10, setting an ultrasonic phased array on the surface of the concrete, and dispersing the concrete into points based on the ultrasonic phased array; the ultrasonic phased array comprises a plurality of transmitting-end excitation transducer array elements and receiving-end excitation transducer array elements.

An ultrasonic phased array is a device formed by arranging a plurality of piezoelectric wafers according to a certain rule, and forms an integral wave front by controlling the emitting time of each wafer in the array, so that the beam scanning, deflection and focusing of ultrasonic waves are realized. Typically, an ultrasound phased array has a square configuration, such as a rectangular or square configuration. Preferably, the present embodiment adopts an ultrasonic phased array of a square structure, and by disposing the ultrasonic phased array on the surface of the concrete, the concrete can be dispersed into dots to emit a wave signal.

Referring to fig. 2, fig. 2 provides an ultrasound phased array structure. Wherein the ultrasound phased array comprises a plurality of excitation transducer array elements. The excitation transducer is an energy conversion device, whose main function is to convert the input electric power into mechanical power (i.e. ultrasonic wave) and transmit it out, while consuming a small part of the power (less than 10%), and as shown in fig. 2, the excitation transducer includes 4 circles from the inner circle to the outer circle, which are respectively the array element structures of 2×2, 4*4, 6*6 and 8×8. When the excitation signal is triggered, some of the array elements in fig. 2 are used as transmitting ends, and others are used as receiving ends, namely the excitation transducer array elements comprise transmitting end excitation transducer array elements and receiving end excitation transducer array elements.

S20, acquiring echo signals through the ultrasonic phased array, wherein the ultrasonic phased array comprises the steps of transmitting out excitation signals coded by Huffman codes through the transmitting-end excitation transducer, receiving corresponding impulse echoes through the receiving-end excitation transducer, and carrying out pulse compression on the impulse echoes according to matched filtering to generate echo signals.

When the echo signals are obtained through the ultrasonic phased array, the signal transmitting end is mainly used for transmitting excitation signals, and the receiving end is used for receiving corresponding impact echoes so as to complete ultrasonic detection once.

It should be noted that, if the transmitting end is directly used to excite the transducer to transmit signals without any processing, a larger receiving delay is usually caused, which affects the testing efficiency. For this purpose, the present embodiment employs excitation signals based on huffman code encoding. After Huffman code encoding, the high frequency part of the wave signal is allocated with a shorter encoding mode, so as to realize rapid encoding of the wave. The transmitting end excitation transducer can transmit the coded excitation signal, so that the transmission speed of the signal is improved, and the receiving end excitation transducer can receive the signal more quickly and acquire the impact echo signal through decoding.

In one embodiment, to further improve the transmission quality of the signal, the chirped pulse compression may be performed before the huffman coding, which significantly improves the average transmit power of the ultrasound by increasing the time-bandwidth product, and then restores the due longitudinal resolution at the receiving end by pulse compression, and can significantly enhance the signal-to-noise ratio.

Further, after the receiving end excites the transducer to receive the corresponding impulse echo, impulse compression is required to be carried out on the impulse echo according to matched filtering, an echo signal is generated, and finally a narrow pulse signal with high resolution and large main and side lobes is obtained.

S30, performing wavelet decomposition on the echo signal, applying a threshold value to a wavelet decomposition coefficient, removing a high-frequency noise component, and then performing wavelet signal reconstruction to obtain a noise-reduced echo signal, wherein the method comprises the following steps:

selecting a proper wavelet base, determining the wavelet decomposition level, and performing wavelet decomposition on the signal to obtain a corresponding wavelet decomposition coefficient;

selecting proper threshold values and threshold value functions to perform threshold value quantization processing on the high-frequency coefficients under each decomposition scale to obtain new high-frequency coefficients;

and carrying out wavelet reconstruction of the one-dimensional signal according to the lowest-layer low-frequency coefficient of wavelet decomposition and each layer high-frequency coefficient subjected to threshold quantization processing to obtain the echo signal after noise reduction.

Applying discrete wavelet analysis can decompose the signal into a set of orthogonal bases corresponding to different time and frequency scales; in the first level of decomposition, the original signal is decomposed into an approximation coefficient and a detail coefficient, the approximation coefficient is further decomposed to obtain an approximation and detail coefficient of a second level, and the process is repeated to obtain approximation and detail coefficients of different decomposition levels; the approximate part is the high-scale, low-frequency component of the signal; the detail part is a low-scale, high-frequency component, namely a noise-containing component:

Wherein, c _j,k Is approximately coefficient, d _j,k H and g are a pair of orthogonal low-pass and high-pass filter banks for detail coefficients; j is the number of decomposition layers; k=0, 1..l-1, L is the number of discrete samples.

S40, performing phase control focusing imaging on each point in the discrete concrete defect according to the echo signals after noise reduction and the selected array element plane, and obtaining images and position information of each point in the defect.

Specifically, in this embodiment, the single focus scanning sound field is used to obtain the focusing and deflection of the transverse and lateral sound beams, and the calculation method is as follows:

F＝(x ² _f +y ² _f +z ² _f ) ^1/2

x _f ＝z _f tanθ _f ,y _f ＝z _f tanφ _f

z _f ＝[F ² /(1+tan ² θ _f +tan ² φ _f )]

in the above, θ _f Is the transverse deflection angle of the focus, phi _f Is the lateral deflection angle, F is the focal length, x _f 、y _f And z _f Corresponding to the distances of the origin of coordinates in the x, y and z directions of the focal point F of the sound beam.

The sound waves emitted by each array element of the transducer reach the focus at the same time, namely, the collective focusing is realized, and finally, the focusing imaging is carried out according to the phased focusing imaging method:

in the formula, h _tx,rx For the signal amplitude of the echo corresponding to the transmitting-receiving array element, x _tx To transmit the array element coordinates, x _rx For receiving the array element coordinates, I is the pixel value in the imaging result diagram, and x and z are the abscissa of the focal point.

By adopting the ultrasonic phased array technology, the concrete can be discretized into points for defect detection, and the ultrasonic phased array can control the signal emission time to form an integral wave front, so that the beam scanning, deflection and focusing of ultrasonic waves are realized, and the accuracy of detecting the defects in the concrete is improved. When the ultrasonic phased array is controlled, the transmission of the excitation signals can be accelerated by adopting the excitation signals coded by the Huffman codes, so that the test efficiency is improved; when receiving the echo signal, the wavelet decomposition is carried out, the threshold value is acted on the wavelet decomposition coefficient, the wavelet signal reconstruction is carried out after the high-frequency noise component is removed, and the stronger noise reduction effect is achieved, so that the quality of the echo signal is enhanced; finally, phase control focusing imaging is carried out on each point in the discrete concrete defect through the echo signal and the selected array element plane, so that accurate positioning of the concrete internal defect can be realized, and the image and the position information of each point in the defect can be obtained.

S50, inputting the images and the position information of each point in the defect into a preset strength prediction model, and outputting a strength prediction result of the concrete.

In this embodiment, an intensity prediction model may be trained first, where the intensity prediction model can identify the image and position information of each point in the input defect, so as to predict the intensity level of the concrete.

In order to ensure the prediction precision, the embodiment preferably adopts a convolutional neural network, namely a CNN network, and an Attention mechanism is added in the convolutional neural network, so that network parameters of a model are further optimized, and the generalization capability is improved.

Referring to fig. 3, in one embodiment, training the intensity prediction model includes the sub-steps of:

s501, obtaining images and position information of each point in a plurality of defects in a concrete historical defect identification result, and taking the images and the position information as a training set;

s502, preprocessing a training set to generate input data, and inputting the input data into a convolution layer of a CNN network to extract feature vectors;

s503, inputting the feature vectors to an Attention mechanism layer of a CNN network, calculating probability weights of different feature vectors, and outputting weighted results of the feature vectors and the probability weights;

S504, inputting the weighted results of the feature vectors and the probability weights to a full connection layer of a CNN network, mapping the weighted feature vectors, and outputting predicted intensity;

s505, verifying the precision of the predicted intensity by using a robustness cross verification method, and generating an intensity prediction model until the precision meets the preset requirement.

The Attention mechanism is used for giving different phase relation number weights to different information processing after simulating human brain to accept information and having different Attention degrees to different information. The structure of the Attention mechanism is shown in fig. 4, in fig. 4: x is x ₁ ,x ₂ ...,x _m ...,x _n The method comprises the steps of (1) inputting the load size and influence factors thereof as raw inputs; h is a ₁ ,h ₂ ...,h _m ...,h _n A state value output for the LSTM hidden layer; alpha ₁ ,α ₂ ...,α _m ...,α _n Assigning a probability distribution value of an output state value of each hidden layer for the Attention mechanism; y is the output of the final network.

Illustratively, the main steps of the robust cross-validation method cross-validation are as follows:

the dataset was divided into K parts, with K-1 parts as training set and the remainder as validation set. Each epoch cycle selects a different portion as the validation set. Since class-to-class distribution in the dataset is not necessarily uniform. Thus, it is desirable to have each fold have the same or at least similar class distribution during folding.

The method comprises the steps of inputting feature vectors to an Attention mechanism layer of a CNN (computer numerical network), calculating probability weights of different feature vectors, outputting weighting results of the feature vectors and the probability weights, mapping the weighted feature vectors through a full-connection layer of the CNN, and finally verifying accuracy by combining a robustness cross-validation method, so that an intensity prediction model with high recognition efficiency and high accuracy can be obtained.

S60, matching the use strategy of the concrete according to the strength prediction result.

After the intensity level is obtained, different usage strategies may be matched according to the intensity level. For example, the material can be used as a core bearing stress component when the strength level is in the range of A level, can be used as an edge component (non-core bearing stress component) when the strength level is in the range of B level, and can be used as a disqualified component for further processing when the strength level is lower than B level.

In summary, according to the embodiment, through the strength prediction model trained based on the convolutional neural network, the strength grade of the concrete component can be rapidly and accurately predicted according to the image and the position information of each point in the defects detected by the phased focusing imaging, and the corresponding use suggestions are matched, so that the use scheme of the concrete can be scientifically guided, and the utilization rate of materials is improved.

In a preferred embodiment, the disposing an ultrasonic phased array on a concrete surface comprises:

It will be appreciated that the problem of determining the optimal array element layout can be seen as an optimisation problem. In this optimization problem, the parameters to be optimized may include at least the position parameters of each array element in the ultrasound phased array, for example, the X-axis coordinates and the y-axis coordinates of each array element. The objective of the optimization problem is to determine an optimal non-uniform array element layout, and before determining the optimal array element layout, an evaluation mode of the array element layout, i.e. defining what array element layout is the desired array element layout, needs to be determined first, and the evaluation mode can be described by a loss function.

The merits of the array element layout are related to the suppression effect of the array element layout on the side lobes. For an array element layout, if the beam directions formed by the array element layout are different, the generated side lobes are different, for example, the array element layout may be low side lobe in the beam direction A and has better side lobe suppression effect, but may be high side lobe in the beam direction B and has poor side lobe suppression effect.

In one embodiment, if only the ultrasonic phased array has higher detection accuracy in the specific direction, the side lobe level corresponding to the specific direction can be used to evaluate the advantages and disadvantages of the array element layout, that is, the lower the side lobe level corresponding to the specific direction is, the better the suppression effect on the side lobe is, and the better the array element layout is considered. However, for the ultrasonic phased array, since the perception of the environment is multidirectional, the ultrasonic phased array is required to have higher detection accuracy in multiple directions, that is, to have better sidelobe suppression effects in multiple beam directions. Therefore, in one embodiment, the sidelobe levels corresponding to the multiple beam directions can be fused, and the fused results of the sidelobe levels corresponding to the multiple beam directions are used for evaluating the advantages and disadvantages of the array element layout, when the method is specifically implemented, a loss function which can describe the fused results can be established, the advantages and disadvantages of the array element layout can be quantized into the value of the loss function, and the lower the value of the loss function corresponding to one array element layout is, the better the comprehensive effect of sidelobe suppression of the array element layout in the multiple directions is, and the better the array element layout is.

After the loss function is determined, constraint conditions corresponding to the loss function can be established. For example, an aperture range corresponding to the ultrasonic phased array may be determined according to the detection environment, and a constraint condition corresponding to the aperture range may be established. In one embodiment, the number range of array elements may also be determined according to the cost control requirement, and a constraint condition corresponding to the number range may be established. In one embodiment, the size of the array elements may be determined according to the process level, and the minimum spacing between the array elements may be determined according to the size, so as to establish a constraint condition corresponding to the minimum spacing.

Further, when determining the target parameter to be optimized which minimizes the value of the loss function, the target parameter to be optimized can be obtained through optimizing through a designated optimizing algorithm. The optimization algorithm specified herein may include, but is not limited to, genetic algorithms, particle swarm algorithms, ant colony algorithms, simulated annealing algorithms, and the like.

To aid understanding, the following is determined by taking a genetic algorithm as an example, and includes the following steps:

1) Initializing parameters to be optimized to obtain a first generation population formed by a plurality of groups of parameters to be optimized.

2) The value of its corresponding loss function is calculated for each individual in the population (i.e. each set of parameters to be optimized).

3) And selecting, crossing and mutating the population.

4) Determining whether a termination condition is satisfied, and if the termination condition is satisfied (the termination condition may be the number of times of selection reaching a set number of times, or the value of the loss function is smaller than a preset value, or the value of the loss function converges), entering step 5); if the termination condition is not satisfied, the process may return to step 2).

5) And selecting an individual with the minimum loss function value from the final generation population, and determining the parameter to be optimized corresponding to the individual as a target parameter.

6) And performing array element layout on the ultrasonic phased array according to the target parameters.

In summary, the function value of the loss function established in the embodiment can represent the fusion result of the side lobe levels corresponding to the ultrasonic phased array in the multiple beam directions, so that the obtained array element layout is optimized based on the loss function, and the ultrasonic phased array has better detection accuracy in the multiple beam directions, thereby improving the test accuracy.

In a preferred embodiment, the acquiring echo signals by the ultrasound phased array further comprises:

dividing the concrete area to generate a plurality of subareas;

In this embodiment, the concrete area may be divided into a plurality of areas by the ultrasonic phased array to generate a plurality of sub-areas, and then the ultrasonic phased array is set in one sub-area, and transmitted in different directions by a transmitting end, so as to collect echo signals.

Taking the ultrasound phased array provided in fig. 2 as an example, since the ultrasound phased array includes 4 circles, respectively, the structures of the array elements of 2 x 2, 4*4, 6*6 and 8 x 8, when transmitting signals, the array element circles surrounded by 4*4 and 6*6 can be used as transmitting ends, and the array element circles of 2 x 2 and 8 x 8 are used as receiving ends. Assuming that the ultrasonic phased array comprises 5 circles and the outermost circle is of a 10 x 10 array element structure, the intermediate 6*6 array element structure can be directly adopted as a transmitting end at the moment, and then signals are transmitted to receiving ends of 2 x 2 and 10 x 10 array element structures at the same time. Finally, after the wave of one sub-area is collected, the echo signals of other areas can be collected in the same way.

Therefore, in this embodiment, the concrete area is divided into a plurality of units to obtain different sub-areas, so that echo signals are collected in the sub-areas respectively, signal attenuation or interference caused by too long one-time propagation distance can be reduced, and then the quality of the impact echo is improved, so as to improve the test precision. When the signal excitation is carried out, the transducer array elements in the central area are used as transmitting ends, and the transducer array elements in the outer ring and the innermost ring are used as receiving ends for signal transmission, so that the wave transmission efficiency is further improved compared with the mode of transmitting from one end to the other end, and the defect detection speed is further accelerated.

In a preferred embodiment, before acquiring echo signals of other sub-regions by translating the ultrasound phased array, determining an optimal excitation frequency of the excitation signal comprises:

It will be appreciated that a smaller diffusion coefficient means that ultrasonic energy from the excitation source to the receiver is continually attenuated, while a larger dissipation coefficient means that ultrasonic energy from the excitation source to the receiver is continually attenuated.

In this embodiment, a diffusion coefficient curve and a dissipation coefficient curve are respectively constructed according to the diffusion coefficient and the dissipation coefficient. The optimal excitation frequency can be obtained by selecting a first frequency corresponding to the maximum slope of the diffusion coefficient curve and a second frequency corresponding to the maximum slope of the dissipation coefficient curve and weighting and averaging the first frequency and the second frequency.

After the optimal excitation frequency is determined, the signal frequency of the excitation transducer at the transmitting end can be timely adjusted when the ultrasonic phased array is translated to acquire the echo signal of the next subarea. Therefore, after the echo signals of the current subarea are acquired, the optimal excitation signals of the next subarea can be calculated according to the method, so that the signal quality can be gradually improved, and finally, the defect detection result is more accurate.

In a preferred embodiment, the adaptive scale threshold is determined using an AdaRound algorithm using a Daubechies wavelet function of order 6 as the basis function for wavelet decomposition.

The key point of wavelet noise reduction is to select proper wavelet basis and threshold value, and different wavelet basis functions and threshold value estimation methods will produce different signal processing results, which directly relate to the quality of signal noise reduction; the Daubechies wavelet function is a type of orthogonal wavelet function widely used in signal processing and image compression, and has important properties such as tight support, smoothness, orthogonality and the like. In view of the fact that the Daubechies wavelet function system has strong localization capability and reconstruction capability of time domains and frequency domains, the embodiment selects the 6-order Daubechies wavelet function as a basis function of wavelet analysis, and meanwhile adopts threshold quantization of an adaptive scale, and the selection of the threshold is determined according to an AdaRound algorithm.

It will be appreciated that the form, location, size and strength of the internal imperfections of the concrete are closely related, thereby affecting the strategy of use of the concrete. For example, when the defects are small and have little effect on the bearing stress and strength, the concrete component can still be used as a component with acceptable quality, and when the defects are serious and have great effect on the bearing stress and strength, the concrete component can be further processed or the use position of the concrete component can be changed. In the above-described embodiments, the results of concrete defect measurement are mainly given, with the aim of improving the accuracy of defect imaging and the accuracy of defect positioning. In one embodiment, the following method is also provided, considering the use of concrete:

in one embodiment, the preprocessing the training set to generate input data includes:

The following advantages can be realized by performing feature dimension reduction on the training set through an SAE algorithm:

removing redundant information: the SAE algorithm can reconstruct input data through the learning process of an automatic encoder, and features which have little influence on the reconstruction effect can be automatically removed in the reconstruction process, so that redundant information in the data is removed.

The important characteristics are reserved: the SAE algorithm may pass input data through the coding layer to get a feature representation with a lower dimension through the learning process of the auto-encoder. During this process, SAE retains features that have an important impact on data reconstruction and model performance, thereby better presenting the essential features of the data set.

Data visualization: the data after feature dimension reduction is more suitable for visual analysis. The SAE can map the high-dimensional data to the low-dimensional space, so that the data visualization is more visual, and further data exploration and analysis are facilitated.

The calculation cost is reduced: reducing the data dimension can reduce the computational complexity of model training and evaluation and speed up the execution of the algorithm. This is very beneficial for processing large-scale data sets and real-time applications.

Relief of dimension disasters: with the increase of the feature quantity, the dimension disaster can cause the problems of increased model complexity, reduced calculation efficiency, reduced model generalization capability and the like. The SAE can alleviate dimension disasters to a certain extent by reducing the dimension of the data, and the effect and generalization capability of the model are improved.

Further, fitting the dimension reduced training set by using a random forest algorithm, an adaptive integration algorithm and/or a gradient lifting tree algorithm.

The random forest algorithm is insensitive to noise and abnormal data and is suitable for high-dimensional characteristic large samples, but has no good training effect on small data and low-dimensional data; the GBDT algorithm can flexibly process various types of data, but cannot process the data in parallel; the AdaBoost algorithm can obtain better learning performance after integrating learners with poorer generalization performance, is not easy to be subjected to fitting, and is sensitive to abnormal data. When different tree models are used for training data, besides inherent defects, the training effect is also influenced by model parameters, so that the importance of the features obtained after training is unreliable. In addition, the feature importance index calculation methods adopted by different tree models are also different. In order to adapt the feature selection algorithm to different types of features, overcome the defects of different tree models, and comprehensively evaluate the importance of the features from multiple angles, the embodiment provides a multi-model comprehensive feature quantity selection method, which comprehensively screens and retains the features with larger correlation coefficients according to training results of a random forest algorithm, a GBDT algorithm and an AdaBoost algorithm.

The random forest algorithm is provided according to a Bagging integrated learning theory and a random subspace theory, the algorithm comprises a plurality of decision trees, and a final classification result is determined by all the decision trees. The basic steps of the random forest algorithm are that firstly, a Bagging thought is adopted to draw part of training samples in a put-back mode, then, a decision tree is established in the training samples obtained through random drawing, node features of the decision tree are drawn from a fixed number of feature spaces which are randomly drawn according to feature selection criteria, and finally, training is stopped until all features and the training samples are drawn. After training, the random forest algorithm can calculate the importance of each feature vector and sort the feature vectors, and the way of calculating the importance is that the data outside the bag calculate the error value, the base coefficient (Gini) and the like. The decision tree of the random forest algorithm randomly replaces each time a fixed number of samples are extracted, the same sample can be extracted for a plurality of times during training, and partial unselected data samples exist, and the ignored samples are called 'out-of-bag data'. The out-of-bag data can be used to evaluate the performance of decision trees and the prediction error rate of models in random forests, which is also referred to as "out-of-bag error".

The GBDT algorithm is similar to the random forest algorithm and is also a decision tree-based learner. However, the GBDT algorithm adopts a Boosting idea, each iteration is improved based on the last learning, and each decision tree learns the conclusion and residual errors of all decision trees at the last stage. In the iterative process, the GBDT algorithm adopts the gradient descent concept to continuously fit the loss function, so that the optimal model is found. The GBDT algorithm selects the split nodes of the decision tree according to the Gini coefficient impure degree variation quantity of the variable, so that the importance degree of the variable in the decision tree, namely the importance of the feature, can be measured by calculating the Gini coefficient impure degree variation quantity.

Specifically, define feature X _Z Characteristic complex correlation coefficient I of (2) _SYNz The method comprises the following steps:

wherein,the characteristic quantity correlation coefficient after being fitted by the random forest model; />Characteristic quantity correlation coefficients after GBDT model fitting are obtained; />And (5) fitting the characteristic quantity correlation coefficient for the AdaBoost model.

After the comprehensive feature importance coefficients are output, the proportion of each feature importance in the total feature importance is calculated, and features with the proportion far smaller than other feature importance proportions are removed, so that feature screening is completed.

And finally, fusing the characteristic quantity with the images and the position information of each point in the defect to serve as input data of the CNN network, so that the training effect of the model is greatly improved.

In one embodiment, the inputting the feature vector into the Attention mechanism layer of the CNN network calculates probability weights of different feature vectors, including:

Specifically, the present embodiment includes the steps of:

1) For the data input layer, a random sampling mode is used, so that the diversity of data is maximized.

Wherein D is _0j Data representing an input layer, j representing the input layer having j base classifiers, X _0j Representing training samples of the j-th basis classifier of the input layer, y representing the label to which the sample corresponds,representing sample weights, the same weights are set at the input layer.

2) Within a hidden layer, multiple parallel basis classifiers are used to train within a hidden layer, generating new training weights based on training errors. Each classifier classification error can be calculated by accumulating before and after classifying the base classifier. New sample weights can be calculated by the weights of the classifier.

3) Between one hidden layer, the indistinguishable transfer samples and the corresponding transfer weights between different learners are transferred between the hidden layers. The embodiment adopts a simple and effective hidden layer forwarding scheme as shown in fig. 5: for each hidden base classifier, a certain number of samples and weights are generated, called transfer samples, transfer weights, and finally combine them with the input of the next layer, the specific rules are as follows: the specific rule is dynamically selected according to the accuracy of the base classifier and the number of training as follows:

1) If the number of the newly generated maximum weights is smaller than 150 and the accuracy of the base is larger than 90%, selecting all the maximum weights and transmitting samples;

2) If the number of newly generated maximum weights is less than 150 and the base accuracy is less than 90%, then 10% of the maximum weights and 10% of the next largest weights are selected.

And combining the selected transfer weight and the weight sample with training data of other base classifiers of the same layer respectively to be used as output of a next layer. The classification data of each base learner of the next layer consists of the training data of the base learner and the transmission data of other classifiers, the training weights correspond to training samples one by one, and finally all the classifiers are integrated to form a final classification model.

In this embodiment, multiple independent base classifiers are used to train simultaneously in one hidden layer, so as to generate a new sample weight, and the hidden layers iterate based on the weight generated in the previous layer, so that the training deficiency of the previous layer is overcome. And a transfer sample and a transfer error are also provided between the hidden layers, are mutually combined and added into training data of the next layer, so that the diversity of the training data is further expanded, the feature extraction capability of the model is further enhanced, and the prediction precision of the model is further improved.

Referring to fig. 6, the present invention also provides an ultrasonic detection system for concrete defects, comprising:

an excitation source arrangement unit 100 for setting an ultrasonic phased array on a concrete surface, and dispersing concrete into points based on the ultrasonic phased array; the ultrasonic phased array comprises a plurality of transmitting-end excitation transducer array elements and receiving-end excitation transducer array elements;

an echo signal obtaining unit 200, configured to obtain an echo signal through the ultrasonic phased array, where the echo signal obtaining unit includes sending an excitation signal encoded by a huffman code through the sending end excitation transducer, receiving a corresponding impulse echo through the receiving end excitation transducer, and performing pulse compression on the impulse echo according to matched filtering, so as to generate an echo signal;

The wavelet decomposition unit 300 is configured to perform wavelet decomposition on the echo signal and apply a threshold to a wavelet decomposition coefficient, perform wavelet signal reconstruction after removing a high-frequency noise component, obtain a noise-reduced echo signal, and obtain a corresponding wavelet decomposition coefficient by selecting an appropriate wavelet basis and determining a wavelet decomposition level to perform wavelet decomposition on the signal; selecting proper threshold values and threshold value functions to perform threshold value quantization processing on the high-frequency coefficients under each decomposition scale to obtain new high-frequency coefficients; carrying out wavelet reconstruction of one-dimensional signals according to the lowest-layer low-frequency coefficient of wavelet decomposition and each layer high-frequency coefficient subjected to threshold quantization processing to obtain echo signals after noise reduction;

the defect identification unit 400 is configured to perform phased focusing imaging on each point in the discrete concrete defect according to the echo signal after noise reduction and the selected array element plane, so as to obtain an image and position information of each point in the defect;

the strength prediction unit 500 is configured to input the image and the position information of each point in the defect into a preset strength prediction model, and output a strength prediction result of the concrete; wherein training the intensity prediction model comprises: obtaining images and position information of points in a plurality of defects in a concrete historical defect identification result, and taking the images and the position information as a training set; preprocessing a training set to generate input data; inputting the input data into a convolution layer of a CNN network to extract a feature vector; inputting the feature vectors to an Attention mechanism layer of a CNN (computer numerical network), calculating probability weights of different feature vectors, and outputting weighted results of the feature vectors and the probability weights; inputting the weighted results of the feature vectors and the probability weights to a full-connection layer of a CNN network, mapping the weighted feature vectors, and outputting predicted intensity; verifying the precision of the predicted intensity by using a robustness cross verification method until the precision meets the preset requirement, and generating an intensity prediction model;

And a policy matching unit 600 for matching the usage policy of the concrete according to the strength prediction result.

In some embodiments, the functions or modules included in the system provided by the present embodiment may be used to perform the methods described in the foregoing method embodiments, and specific implementations thereof may refer to descriptions of the foregoing method embodiments, which are not repeated herein for brevity.

The invention also provides an electronic device, comprising: a processor, a transmitting means, an input means, an output means and a memory for storing computer program code comprising computer instructions which, when executed by the processor, cause the electronic device to perform a method as any one of the possible implementations described above.

The invention also provides a computer readable storage medium having stored therein a computer program comprising program instructions which, when executed by a processor of an electronic device, cause the processor to perform a method as any one of the possible implementations described above.

Referring to fig. 7, fig. 7 is a schematic diagram of a hardware structure of an electronic device according to an embodiment of the invention.

The electronic device 2 comprises a processor 21, a memory 22, input means 23, output means 24. The processor 21, memory 22, input device 23, and output device 24 are coupled by connectors including various interfaces, transmission lines or buses, etc., as are not limited by the present embodiments. It should be appreciated that in various embodiments of the invention, coupled is intended to mean interconnected by a particular means, including directly or indirectly through other devices, e.g., through various interfaces, transmission lines, buses, etc.

The processor 21 may be one or more graphics processors (graphics processing unit, GPUs), which may be single-core GPUs or multi-core GPUs in the case where the processor 21 is a GPU. Alternatively, the processor 21 may be a processor group formed by a plurality of GPUs, and the plurality of processors are coupled to each other through one or more buses. In the alternative, the processor may be another type of processor, and the embodiment of the invention is not limited.

Memory 22 may be used to store computer program instructions as well as various types of computer program code for performing aspects of the present invention. Optionally, the memory includes, but is not limited to, a random access memory (random access memory, RAM), a read-only memory (ROM), an erasable programmable read-only memory (erasable programmable read only memory, EPROM), or a portable read-only memory (compact disc read-only memory, CD-ROM) for associated instructions and data.

The input means 23 are for inputting data and/or signals and the output means 24 are for outputting data and/or signals. The output device 23 and the input device 24 may be separate devices or may be an integral device.

It will be appreciated that in embodiments of the present invention, the memory 22 may not only be used to store relevant instructions, but embodiments of the present invention are not limited to the specific data stored in the memory.

It will be appreciated that fig. 5 shows only a simplified design of an electronic device. In practical applications, the electronic device may further include other necessary elements, including but not limited to any number of input/output devices, processors, memories, etc., and all video parsing devices capable of implementing the embodiments of the present invention are within the scope of the present invention.

Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.

It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, and are not repeated herein. It will be further apparent to those skilled in the art that the descriptions of the various embodiments of the present invention are provided with emphasis, and that the same or similar parts may not be described in detail in different embodiments for convenience and brevity of description, and thus, parts not described in one embodiment or in detail may be referred to in description of other embodiments.

In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices, and methods may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.

The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit.

Those of ordinary skill in the art will appreciate that implementing all or part of the above-described method embodiments may be accomplished by a computer program to instruct related hardware, the program may be stored in a computer readable storage medium, and the program may include the above-described method embodiments when executed. And the aforementioned storage medium includes: a read-only memory (ROM) or a random access memory (random access memory, RAM), a magnetic disk or an optical disk, or the like.

Claims

1. An ultrasonic detection method for concrete defects, which is characterized by comprising the following steps:

preprocessing a training set to generate input data, and inputting the input data into a convolution layer of a CNN (computer numerical network) to extract feature vectors;

2. The method for ultrasonic detection of defects in concrete according to claim 1, wherein said disposing an ultrasonic phased array on a surface of the concrete comprises:

3. The method for ultrasonic detection of concrete defects according to claim 1, wherein echo signals are acquired by the ultrasonic phased array, further comprising:

dividing the concrete area to generate a plurality of subareas;

4. A method of ultrasonically detecting a concrete defect according to claim 3, further comprising determining an optimal excitation frequency of an excitation signal prior to acquiring echo signals of other sub-areas by translating the ultrasonic phased array, comprising:

5. The ultrasonic detection method of concrete defects according to claim 1, wherein a 6 th order Daubechies wavelet function is used as a basis function of wavelet decomposition, and an AdaRound algorithm is used for determining a threshold of an adaptive scale.

6. The method for ultrasonic detection of concrete defects according to claim 1, wherein the preprocessing of the training set to generate input data comprises:

7. The ultrasonic detection method of concrete defects according to claim 1, wherein the inputting of feature vectors into an Attention mechanism layer of a CNN network calculates probability weights of different feature vectors, comprising:

8. An ultrasonic detection system for concrete defects, the system comprising:

9. An electronic device, comprising: a processor and a memory for storing computer program code comprising computer instructions which, when executed by the processor, cause the electronic device to perform the method of any one of claims 1 to 7.

10. A computer readable storage medium, characterized in that the computer readable storage medium has stored therein a computer program comprising program instructions which, when executed by a processor of an electronic device, cause the processor to perform the method of any of claims 1 to 7.