Concrete internal defect detection method and system based on ultrasonic method

Abstract

The application relates to the field of concrete defect detection, in particular to a concrete internal defect detection method and system based on an ultrasonic method, which comprises the following steps: acquiring a voltage sequence corresponding to a concrete sample and a defect label; acquiring a plurality of voltage features of the voltage sequence, and calculating the importance of any voltage feature to the defect label; constructing an eigenvalue vector, training a defect detection model according to the eigenvalue vector and the defect label; collecting a plurality of to-be-detected voltage sequences of a to-be-detected concrete sample, inputting the eigenvalue vector of any to-be-detected voltage sequence into the trained defect detection model, outputting a to-be-detected defect probability, and obtaining the to-be-detected defect probability of each to-be-detected voltage sequence through iteration; and taking the average of all to-be-detected defect probabilities as the final defect probability of the to-be-detected concrete sample. Through the technical scheme, the precision of the concrete internal defect detection result can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of concrete defect detection. In particular, it relates to a method and system for detecting internal defects of concrete based on ultrasonic method. BACKGROUND

[0002] As one of the most commonly used building materials in modern engineering construction, the structural safety and durability of concrete are directly related to the service life of the project and personnel safety. However, during the construction, maintenance and service of concrete, internal defects such as cracks and cavities may occur due to material properties, construction technology, environmental factors and other factors. If these defects are not discovered in time, they may lead to a decrease in structural bearing capacity, leakage and even collapse and other serious accidents. As a non-destructive testing technology, ultrasonic method detects internal defects by transmitting ultrasonic waves into the concrete and using changes in speed, amplitude, frequency, waveform and other parameters of the sound waves in the medium to determine internal defects.

[0003] However, traditional ultrasonic detection technology still faces some challenges, especially in terms of precision, efficiency and consistency in detecting complex structures and multi-level concrete volumes. The existing technology does not consider the correlation between different features and defects in the ultrasonic detection results, resulting in inaccurate defect detection results. SUMMARY

[0004] To solve the above technical problems, the present application provides solutions in the following aspects.

[0005] In a first aspect, a method for detecting internal defects of concrete based on ultrasonic method includes: obtaining ultrasonic detection results of concrete samples of any one defect type in history, the ultrasonic detection results being voltage sequences, one concrete sample corresponding to one voltage sequence and one defect label; for any voltage sequence, obtaining a plurality of voltage features of the voltage sequence, and calculating the importance of any voltage feature to the defect label; retaining the feature values of the voltage features with importance greater than a preset threshold, and setting the feature values of the voltage features with importance not greater than the preset threshold to zero to construct a feature value vector, training a defect detection model according to the feature value vector and the defect label; collecting a plurality of to-be-detected voltage sequences of to-be-detected concrete samples, inputting the feature value vector of any to-be-detected voltage sequence into the trained defect detection model, outputting a to-be-detected defect probability, traversing to obtain the to-be-detected defect probability of each to-be-detected voltage sequence, taking the average of all to-be-detected defect probabilities as the final defect probability of the to-be-detected concrete sample, and completing defect detection.

[0006] Preferably, the calculating the importance of any voltage feature to the defect label comprises: dividing the concrete samples according to the defect label, grouping the concrete samples with the same defect label, dimension reduction on the voltage features of any group of concrete samples, and obtaining the load of any voltage feature to any principal component after dimension reduction; taking the square sum of the load of any voltage feature to all principal components after dimension reduction as a first accumulated value, taking the accumulated value of the square sum of the load of all voltage features to all principal components after dimension reduction as a second accumulated value, and taking the ratio of the first accumulated value to the second accumulated value as the importance of any voltage feature to the defect label.

[0007] Preferably, the calculating the importance of any voltage feature to the defect label comprises: constructing a feature matrix, wherein the rows of the feature matrix represent the feature values of each voltage feature of the same concrete sample, and the columns of the feature matrix represent the feature values of the same voltage feature of different concrete samples; clustering the feature values of any column, obtaining the best clustering result according to the contour coefficient method, and taking the center point of any clustering cluster as a distinguishing value, traversing to obtain the distinguishing value of each clustering cluster, sorting the clustering clusters in ascending order according to the distinguishing values to obtain the serial numbers of each clustering cluster, replacing the feature values in the clustering cluster with the serial numbers, and sequentially obtaining the serial numbers of each position in the feature matrix to obtain a serial number matrix; for any voltage feature, calculating the probability of each serial number appearing in any defect type, constructing a probability vector of any voltage feature and any defect type, and calculating the importance of any voltage feature to the defect label based on the probability vector.

[0008] Preferably, the calculating the importance of any voltage feature to the defect label based on the probability vector comprises: taking the inverse of the entropy of the probability vector of any voltage feature to the defect label as the importance of any voltage feature to the defect label.

[0009] Preferably, the calculating the importance of any voltage feature to the defect label based on the probability vector comprises: calculating the square of each element value in the probability vector, normalizing the sum of the squares of all element values, and taking the sum as the importance of any voltage feature to the defect label.

[0010] Preferably, the defect detection model is a BP network, the input of the BP network is a feature value vector, and the output of the BP network is a defect probability.

[0011] Preferably, the training process of the defect detection model is: taking the feature value vector of any one of the defect types of the historical concrete sample as input, taking the defect label of any one of the defect types of the historical concrete sample as network label, obtaining a set of training data; inputting the training data into the defect detection model to obtain an output result; calculating the loss value of the output result and the network label through the cross-entropy loss function, propagating the error signal back according to the loss value, updating the network parameters of the defect detection model, and making the loss value smaller; iteratively updating the network parameters of the defect detection model, and stopping updating when the defect detection model reaches a set maximum training number or the loss value is less than a set loss value, thereby obtaining the trained defect detection model.

[0012] In a second aspect, a concrete internal defect detection system based on an ultrasonic method includes a processor and a memory, and the memory stores computer program instructions that, when executed by the processor, implement any of the concrete internal defect detection methods based on the ultrasonic method.

[0013] The present application has the following effects:

[0014] The present application can accurately identify and evaluate various internal defects in concrete by effectively extracting and optimizing the voltage features in the concrete ultrasonic detection data. The present application uses the correlation between voltage features and defect labels to build a training model that can efficiently detect different defect types. By screening and analyzing the importance of voltage features, the features that have a significant impact on defect judgment are retained, thereby improving the accuracy and robustness of the detection. In addition, by collecting data of the sample to be detected multiple times and performing feature extraction and model prediction, the final defect probability of the concrete sample can be given. This process not only reduces the error caused by human intervention, but also greatly improves the detection efficiency, meeting the demand for efficient and accurate concrete detection in actual engineering. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 is a flowchart of the concrete internal defect detection method based on the ultrasonic method according to an embodiment of the present application. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments of the present application.

[0017] The specific embodiments of the present application will be described in detail below with reference to the drawings.

[0018] Reference Figure 1 The concrete internal defect detection method based on the ultrasonic method includes steps S1-S3, and specifically as follows:

[0019] S1: Obtain the ultrasonic detection results of any one of the defect types of the concrete sample in the history, and the ultrasonic detection results are voltage sequences, one concrete sample corresponds to one voltage sequence and one defect label.

[0020] It should be noted that ultrasonic detection is a common non-destructive testing technique. By transmitting ultrasonic signals into the interior of the concrete structure, the reflected signals are detected to determine the internal defect conditions of the material. In practical applications, the concrete is detected by an ultrasonic detection device, and a voltage sequence is obtained. These voltage values represent the intensity and time characteristics of the reflected signals received by the ultrasonic probe, reflecting the acoustic wave propagation conditions inside the structure. Each voltage sequence corresponds to a specific time point and acoustic intensity variation, which can reveal the defect types inside the material, such as health, cracks, cavities, delamination, etc.

[0021] In one embodiment, in order to perform ultrasonic detection analysis of concrete defect types, first, concrete samples of different defect types in history are obtained and tested by ultrasonic detection technology. Each concrete sample will generate a voltage sequence, which records the changes in the internal structure of the concrete through the reflection characteristics of ultrasonic propagation. Each concrete sample not only includes voltage sequence data, but also has a defect label indicating the defect type or defect-free state of the concrete sample.

[0022] S2: For any voltage sequence, obtain several voltage features of the voltage sequence, calculate the importance of any voltage feature to the defect label; retain the feature values of the voltage features with importance greater than the preset threshold, and set the feature values of the voltage features with importance not greater than the preset threshold to zero, to construct a feature value vector, and train a defect detection model according to the feature value vector and the defect label.

[0023] In one embodiment, since the internal defects of concrete (such as cracks, cavities, delamination, etc.) have different degrees of influence on the propagation of ultrasonic waves, these defects will change the waveform characteristics of the voltage sequence. In order to effectively extract these features and analyze them, the time domain features and frequency domain features of the voltage sequence are calculated. Time domain features usually reflect the basic statistical characteristics of the waveform, such as peak amplitude, peak difference, skewness (reflecting the asymmetry of the waveform) and kurtosis (the sharpness of the waveform, defects often cause the kurtosis to increase). Frequency domain features reveal the frequency distribution of the signal, such as average frequency, spectral entropy (used to measure the complexity or uncertainty of the signal), high frequency energy ratio (reflecting the energy proportion of high frequency components), sample entropy (used to measure the complexity of the signal), etc. In addition, wave velocity ratio and attenuation coefficient are also commonly used to characterize the ultrasonic wave propagation characteristics of concrete. The wave velocity ratio can reflect the influence of different defect types on wave velocity, and the attenuation coefficient indicates the degree of energy attenuation in the signal propagation process.

[0024] The importance of any voltage feature to the defect label is calculated, including:

[0025] According to the defect label, the concrete samples are divided into groups, and the voltage features of any group of concrete samples are reduced in dimension to obtain the load of any voltage feature to any principal component after dimension reduction. The sum of the squares of the load of any voltage feature to all principal components after dimension reduction is taken as the first cumulative value, the sum of the squares of the load of all voltage features to all principal components after dimension reduction is taken as the second cumulative value, and the ratio of the first cumulative value to the second cumulative value is taken as the importance of any voltage feature to the defect label. The importance satisfies the relationship:

[0026] , represents the voltage feature to the defect label, represents the voltage feature to the load of the principal component after dimension reduction, represents the number of principal components selected after dimension reduction, represents the number of voltage features, reflects the contribution of the voltage feature to the principal component after dimension reduction. The greater the total contribution of the voltage feature to the principal component after dimension reduction before the voltage sequence is reduced, the more important the voltage sequence.

[0027] The above method obtains the importance of all voltage features to the same defect label. According to this method, the importance of all voltage features to each defect label can also be obtained.

[0028] The feature values of the voltage features with importance greater than the preset threshold value are retained, and the feature values of the voltage features with importance not greater than the preset threshold value are set to zero to construct a feature value vector. For example, the voltage features are peak amplitude, peak difference, spectral entropy, high-frequency energy ratio, and sample entropy, and their importance to the defect label of crack is 0.6, 0.4, 0.8, 0.7, and 0.2, respectively. The preset threshold value is 0.5. Therefore, the feature values greater than the preset threshold value are retained, the feature values not greater than the preset threshold value are set to 0, and each element in the feature value vector is 0.6, 0, 0.8, 0.7, and 0, respectively.

[0029] The defect detection model is trained according to the eigenvector and the defect label, the defect detection model is a BP network, the input of the BP network is the eigenvector, and the output is a defect probability. The training process is as follows: the eigenvector of any one defect type of the concrete sample in the history is taken as the input, the defect label of any one defect type of the concrete sample in the history is taken as the network label, and a set of training data is obtained; the training data is input into the defect detection model to obtain an output result; the loss value of the output result and the network label is calculated through a cross-entropy loss function, an error signal is propagated in a reverse direction according to the loss value, the network parameters of the defect detection model are updated, and the loss value is reduced; the network parameters of the defect detection model are iteratively updated, and when the defect detection model reaches a set maximum training number or the loss value is less than a set loss value, the updating is stopped, and a trained defect detection model is obtained.

[0030] In one embodiment, the importance of any one voltage feature to the defect label includes: constructing a feature matrix, the rows of the feature matrix representing the eigenvectors of each voltage feature of the same concrete sample, and the columns of the feature matrix representing the eigenvectors of the same voltage feature of different concrete samples.

[0031] The eigenvectors of any column are clustered, the best clustering result is obtained according to the contour coefficient method, the center point of any clustering cluster is taken as a distinguishing value, the distinguishing value of each clustering cluster is obtained by traversal, the clustering clusters are sorted from small to large according to the distinguishing values to obtain the serial numbers of the clustering clusters, the serial numbers are used to replace the eigenvectors in the clustering clusters, the serial numbers of each position in the feature matrix are sequentially obtained, a serial number matrix is obtained, for any one voltage feature, the probability of each serial number appearing in any one defect type is calculated, a probability vector of any one voltage feature and any one defect type is constructed, and the importance of any one voltage feature to the defect label is obtained by mapping the result of the reciprocal of the probability vector entropy of any one voltage feature to the defect label through an exponential function. The importance satisfies the following relationship:

[0032] , denotes the voltage feature the importance of the defect label , denotes the element value at the th position in the probability vector of the voltage feature and the defect label , is the reciprocal of the probability vector entropy, and the greater the probability vector entropy, the more unstable the relationship between the voltage feature and the defect label, and the lower the importance.

[0033] In one embodiment, calculating the importance of any voltage feature to the defect label based on the probability vector comprises: calculating the square of each element value in the probability vector, normalizing the sum of the squares of all element values as the importance of any voltage feature to the defect label. The importance satisfies the relationship:

[0034] , denotes the importance of the voltage feature to the defect label , denotes the element value at the th position in the probability vector of the voltage feature and the defect label , is the energy value of the probability vector, and the greater the energy value, the more certain the relationship and the greater the importance of the voltage feature to the defect label.

[0035] S3: Collect multiple voltage sequences of the concrete sample to be detected, input the feature value vector of any voltage sequence to be detected into the trained defect detection model, output the defect probability to be detected, traverse to obtain the defect probability to be detected of each voltage sequence to be detected, take the mean value of all defect probabilities to be detected as the final defect probability of the concrete sample to be detected, and complete the defect detection.

[0036] It should be noted that in actual detection, due to environmental factors, electromagnetic interference, vibration and other external disturbances, a single collected voltage sequence may be affected, resulting in inaccurate measurement data. Therefore, in order to improve the reliability and accuracy of the detection result, the same concrete sample to be detected will usually be collected multiple times. Obtain multiple voltage sequences, which will then be input into the previously trained defect detection model. Through the model, each voltage sequence is analyzed and the corresponding defect probability is output. Since the data collected each time may fluctuate to some extent, the detection result of a single sequence may be biased. Therefore, by traversing all collected voltage sequences and calculating the defect probability corresponding to each sequence, the influence of accidental factors can be reduced. Finally, the defect probabilities of all voltage sequences are averaged to obtain the final defect probability of the concrete sample to be detected. This can effectively integrate the results of multiple detections, improve the accuracy of defect judgment, and reduce the influence of environmental noise or other interference factors on the detection result.

[0037] The system comprises a processor and a memory, and the memory stores computer program instructions which, when executed by the processor, implement the ultrasonic method-based concrete internal defect detection method according to the first aspect of the present application.

[0038] The system also comprises other components well known to those skilled in the art, such as a communication bus and a communication interface, whose arrangement and function are known in the art and therefore will not be described here.

[0039] It should be noted that, for one of ordinary skill in the art, several variations and improvements can be made without departing from the inventive concept, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application patent should be subject to the appended claims.

Claims

1. A method for detecting internal defects in concrete based on ultrasonic testing, characterized in that, include: Obtain the ultrasonic test results of concrete samples with any type of defect in history. The ultrasonic test results are voltage sequences, and each concrete sample corresponds to a voltage sequence and a defect label. For any voltage sequence, obtain several voltage features of the voltage sequence and calculate the importance of any voltage feature to the defect label; retain the feature values ​​of voltage features whose importance is greater than a preset threshold, and set the feature values ​​of voltage features whose importance is not greater than the preset threshold to zero, so as to construct a feature value vector, and train the defect detection model based on the feature value vector and the defect label; Multiple collections of the voltage sequence to be detected from the concrete sample to be tested are performed. The feature value vector of any voltage sequence to be tested is input into the trained defect detection model, and the probability of the defect to be detected is output. The probability of the defect to be detected for each voltage sequence to be tested is obtained by iterating through the model. The mean of all the probability of the defect to be detected is taken as the final defect probability of the concrete sample to be tested, thus completing the defect detection. The calculation of the importance of any voltage characteristic for defect labeling includes: Construct a feature matrix where the rows of the feature matrix represent the feature values ​​of each voltage feature for the same concrete sample, and the columns of the feature matrix represent the feature values ​​of different concrete samples for the same voltage feature. Clustering is performed on the feature values ​​of any column. The optimal clustering result is obtained using the silhouette coefficient method. The centroid of any cluster is used as the discriminant value. The discriminant values ​​of each cluster are obtained by iterating through the clusters. The clusters are then sorted in ascending order of their discriminant values ​​to obtain the index of each cluster. The index is used to replace the feature values ​​within each cluster, and the index of each position in the feature matrix is ​​obtained sequentially to obtain the index matrix. For any voltage feature, calculate the probability of each number appearing in any defect type, construct a probability vector of any voltage feature and any defect type, and calculate the importance of any voltage feature to the defect label based on the probability vector.

2. The method for detecting internal defects in concrete based on ultrasonic technology according to claim 1, characterized in that, The importance of any voltage feature to the defect label based on probability vector calculation includes: The inverse of the probability vector entropy of any voltage feature to the defect label is mapped by an exponential function as the importance of any voltage feature to the defect label.

3. The method for detecting internal defects in concrete based on ultrasonic technology according to claim 1, characterized in that, The importance of any voltage feature to the defect label based on probability vector calculation includes: Calculate the square of each element in the probability vector, normalize the sum of the squares of all elements, and use this as the importance of any voltage feature to the defect label.

4. The method for detecting internal defects in concrete based on ultrasonic testing according to claim 1, characterized in that, The defect detection model is a BP network, with the input being a feature vector and the output being the defect probability.

5. The method for detecting internal defects in concrete based on ultrasonic technology according to claim 1, characterized in that, The training process of the defect detection model is as follows: The feature vector of any type of concrete sample in history is used as input, and the defect label of any type of concrete sample in history is used as the network label to obtain a set of training data. Input the training data into the defect detection model to obtain the output results; The loss value of the output result and network label is calculated by using the cross-entropy loss function. The error signal is backpropagated based on the loss value to update the network parameters of the defect detection model and reduce the loss value. The network parameters of the defect detection model are updated iteratively. When the defect detection model reaches the set maximum number of training iterations or the loss value is less than the set loss value, the update is stopped, and the trained defect detection model is obtained.

6. A concrete internal defect detection system based on ultrasonic method, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement the method for detecting internal defects in concrete based on ultrasonic methods according to any one of claims 1-5.

Images