Concrete vibration quality cross-modal detection system and data processing method

Abstract

The invention provides a concrete vibration quality cross-modal detection system. The concrete vibration quality cross-modal detection system comprises a signal acquisition module, a data processing module and a quality analysis module. The data processing method comprises the following steps that a signal collecting module composed of a plurality of acceleration sensors is used for collecting the axial movement acceleration of the vibrating rod and acceleration response data of different positions in the effective vibrating radius; the method comprises the following steps: processing a multi-source sensing signal by adopting a symmetric point mode analysis method, converting a time sequence into a polar coordinate system scatter diagram, and fusing all sensor signals into the same SDP image by adjusting initial parameters of symmetric point mode analysis so as to increase characterization differences of different fault states; performing vibration construction quality analysis on the SDP image by using a convolutional neural network; through manual classification and machine learning training, the CNN model can accurately judge that the vibration construction quality is under vibration, compact or over vibration. According to the scheme, the vibration construction quality can be accurately judged, and the construction quality and efficiency are improved.

Description

Technical Field

[0001] The invention relates to the field of concrete construction quality detection, and in particular to a concrete vibration quality cross-modal detection system and a data processing method. Background Art

[0002] During the concrete construction process, vibration quality is one of the key factors to ensure the uniformity and structural strength of concrete. Traditional vibration quality detection mainly relies on manual observation and empirical judgment. This method is not only highly subjective, but also inefficient, and it is difficult to ensure the accuracy and consistency of the detection results. With the development of sensor technology and signal processing technology, more and more automated detection methods are being applied to the monitoring of concrete vibration quality. However, the existing automated detection methods often rely only on a single sensor, and the data information collected is limited, which makes it difficult to fully reflect the complex dynamic characteristics of the vibration process. In practical applications, there are often various unpredictable noises and other interferences that affect the results of signal analysis, resulting in insufficient accuracy and reliability of the detection results. Referring to the current experience of the development and application of multimodal technology, the use of cross-modal feature interaction technology can reduce random interference and amplify the representation differences between the features to be extracted. Therefore, a highly robust cross-modal detection system and data processing method for concrete vibration quality are proposed, which can achieve more accurate vibration quality detection in complex scenes of concrete vibration construction. Summary of the invention

[0003] The purpose of the present invention is to solve the technical problems in the above background, and to propose a concrete vibration quality cross-modal detection system, which is characterized by: comprising a signal acquisition module, a data processing module and a quality analysis module;

[0004] The signal acquisition module is used to collect the axial motion acceleration of the vibrating rod and the acceleration response data of different positions within the effective vibrating radius of the vibrating rod;

[0005] The data processing module comprehensively processes the signals from multiple sensors through symmetrical point pattern analysis;

[0006] The quality analysis module uses convolutional neural networks to analyze the quality of vibration construction.

[0007] In a preferred solution, the signal acquisition module includes several acceleration sensors, all of which are in 16-channel dynamic data acquisition mode, with an acquisition frequency of 1000 Hz and an effective vibration period of 30 s.

[0008] In a preferred solution, the acceleration sensor includes one acceleration sensor and a plurality of three-axis acceleration sensors, wherein the uniaxial acceleration sensor is mounted on the shaft wall of the vibrating rod to collect the axial motion acceleration of the vibrating rod;

[0009] Several three-axis acceleration sensors are installed at different positions from the vibrating rod and are on the same horizontal line as the installation position of the uniaxial acceleration sensor. They are used to collect acceleration response data at different positions within the effective vibration radius of the vibrating rod.

[0010] The present invention also provides a data processing method for a concrete vibration quality cross-modal detection system, comprising the following steps:

[0011] S1. Using a signal acquisition module composed of several acceleration sensors, respectively collect the axial motion acceleration of the vibrating rod and the acceleration response data of different positions within the effective vibration radius;

[0012] S2, using symmetrical point pattern analysis to process multi-source sensor signals, converting the time series into a polar coordinate scatter plot, and by adjusting the initial parameters of the symmetrical point pattern analysis, fusing all sensor signals into the same SDP image to increase the representation differences of different fault states;

[0013] S3. Use convolutional neural network to analyze the vibration construction quality of SDP images. Through manual classification and machine learning training, the CNN model can accurately identify the category of SDP images, that is, determine whether the vibration construction quality is under-vibration, dense or over-vibration.

[0014] In the preferred solution, the symmetrical point pattern analysis method is specifically as follows: using the signal amplitudes xi and xi+1 at the i-th moment and the i+1-th moment in the time series x, the coordinates of the i-th pair of scattered points in the polar coordinate system (r(i), θ(i), ), the coordinate calculation formula is:

[0015]

[0016] Where: x max and x min are the maximum and minimum values ​​in the time series x; θ is the initial angle of the drawing; ξ is the angular domain gain factor; l is the time lag coefficient.

[0017] In the preferred scheme, the symmetrical point pattern analysis continuously converts the time series elements into scattered point pairs in the polar coordinate system, and finally forms two symmetrical lobes within the specified angle range [θ-ξ,θ+ξ], wherein θ is the symmetry axis angle of the two lobes, θ determines the symmetry property of the scatter plot, the angular domain gain factor ξ limits the angular domain width of the lobe in the polar coordinate system, and the time delay coefficient l affects the morphology of the lobe.

[0018] In a preferred solution, when there are several sensors, the symmetrical point pattern analysis fuses the signals of all sensors into the same SDP image by adjusting the initial drawing angle θ and the angular domain gain factor ξ.

[0019] In the preferred solution, in the initial parameter setting of the symmetric point pattern analysis, the pixel matrix size of the SDP image is set to 64×64 to ensure sufficient image resolution;

[0020] In order to make full use of the information of each sensor, the initial drawing angle of each sensor is set to 0°, 72°, 144°, 216° and 288° respectively to ensure the uniform distribution of the scatter plot in the polar coordinate system; in addition, the angular domain gain factor ξ and the time delay coefficient l are uniformly set to 18° and 10 to ensure the analysis effect and simplify the parameter adjustment process.

[0021] In the preferred scheme, CNN is mainly composed of an input layer, a convolution layer, a pooling layer and a fully connected layer. The input layer inputs a 64×64 SDP image. The convolution layer extracts features from the input SDP image to form a feature map. Then the pooling layer reduces the dimension of the features extracted by the convolution layer to reduce the feature dimension and the number of features to prevent feature overfitting. Finally, the features are stretched to one dimension through the fully connected layer, and the classification probability is calculated to obtain the classification calculation result of the model.

[0022] In the preferred solution, after obtaining a large number of SDP images, the SDP images are manually classified and divided into three categories according to the actual corresponding vibration construction quality: under-vibration, dense and over-vibration;

[0023] After the classification is completed, the data is randomly divided into machine learning training set and test set in a ratio of 7:3. Through repeated optimization and adjustment, hyperparameters such as CNN training learning rate and number of iterations are set until the CNN training loss converges well.

[0024] The trained CNN model can identify the quality of vibration construction: under-vibration, denseness and over-vibration.

[0025] The beneficial effects of the present invention are as follows:

[0026] (1) Improving detection accuracy: Through multi-source sensor signal acquisition and fusion, the present invention can fully reflect the complex dynamic characteristics of the vibration process, effectively improving the detection accuracy.

[0027] (2) Enhanced detection reliability: By adopting advanced SDP data processing methods and CNN machine learning algorithms, the present invention can accurately judge the quality of vibration construction and enhance the reliability of detection.

[0028] (3) Improve detection efficiency: Compared with traditional manual detection methods, the present invention realizes automated detection, greatly improves detection efficiency and reduces labor costs.

[0029] (4) Easy to promote and apply: The detection method of the present invention is not limited to specific scenarios, has strong versatility and applicability, and is easy to promote and apply in various concrete construction projects. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a schematic diagram of the installation of the acceleration sensor in the present invention.

[0031] Figure 2 It is a work flow chart of a cross-modal detection system for concrete vibration quality in the present invention. DETAILED DESCRIPTION

[0032] Example 1

[0033] A concrete vibration quality cross-modal detection system, comprising:

[0034] The signal acquisition module is used to collect the axial motion acceleration of the vibrating rod and the acceleration response data of different positions within the effective vibrating radius of the vibrating rod;

[0035] The data processing module comprehensively processes the signals from multiple sensors through the symmetrical point pattern analysis method (SDP);

[0036] The quality analysis module uses convolutional neural networks to analyze the quality of vibration construction.

[0037] Preferably, the signal acquisition module includes several acceleration sensors, all of which are in 16-channel dynamic data acquisition mode, with an acquisition frequency of 1000 Hz and an effective vibration period of 30 s.

[0038] Preferably, the acceleration sensor includes one acceleration sensor and a plurality of three-axis acceleration sensors, wherein the uniaxial acceleration sensor is mounted on the shaft wall of the vibrating rod to collect the axial motion acceleration of the vibrating rod;

[0039] Several three-axis acceleration sensors are installed at different positions from the vibrating rod and are on the same horizontal line as the installation position of the uniaxial acceleration sensor. They are used to collect acceleration response data at different positions within the effective vibration radius of the vibrating rod.

[0040] Preferably, the method comprises the following steps:

[0041] S1. Using a signal acquisition module composed of several acceleration sensors, respectively collect the axial motion acceleration of the vibrating rod and the acceleration response data of different positions within the effective vibration radius;

[0042] S2. Use the symmetrical point pattern analysis method to process multi-source sensor signals, convert the time series into a polar coordinate scatter plot, and fuse all sensor signals into the same SDP image by adjusting the initial parameters of the symmetrical point pattern analysis to increase the representation differences of different fault states;

[0043] S3. Use convolutional neural network to analyze the vibration construction quality of SDP images. Through manual classification and machine learning training, the CNN model can accurately identify the category of SDP images, that is, determine whether the vibration construction quality is under-vibration, dense or over-vibration.

[0044] Preferably, the symmetrical point pattern analysis method is specifically: using the signal amplitude x at the i-th moment and the i+1-th moment in the time series x i 、x i+l , calculate the coordinates of the i-th pair of scattered points in the polar coordinate system (r(i), θ(i), ), the coordinate calculation formula is as follows:

[0045]

[0046] Where: x max and x min are the maximum and minimum values ​​in the time series x; θ is the initial angle of the drawing; ξ is the angular domain gain factor; l is the time lag coefficient.

[0047] Preferably, the symmetrical point pattern analysis forms two symmetrical lobes in the specified angle range [θ-ξ,θ+ξ] by continuously converting the time series elements into scattered point pairs in the polar coordinate system. Among them, θ is the symmetry axis angle of the two lobes, which determines the symmetry property of the scatter plot. The angular domain gain factor ξ limits the angular domain width of the lobe in the polar coordinate system, while the time delay coefficient l has a significant impact on the morphology of the lobe.

[0048] Preferably, when there are multiple sensors, the symmetric point pattern analysis can fuse the signals of all sensors into the same SDP image by cleverly adjusting the initial angle θ of the drawing and the angular domain gain factor ξ. This fusion strategy not only enhances the characterization difference between different fault states, but also effectively makes up for the deficiency of a single sensor in extracting feature information.

[0049] Preferably, in the initial parameter setting of the symmetrical point pattern analysis, the pixel matrix size of the SDP image is set to 64×64 to ensure sufficient image resolution. At the same time, in order to make full use of the information of the five sensors, their initial drawing angles are set to 0°, 72°, 144°, 216° and 288° respectively to ensure the uniform distribution of the scatter plot in the polar coordinate system. In addition, the angular domain gain factor ξ and the time delay coefficient l are uniformly set to 18° and 10 to simplify the parameter adjustment process while ensuring the analysis effect.

[0050] Preferably, CNN is mainly composed of an input layer, a convolution layer, a pooling layer and a fully connected layer. The input layer inputs a 64×64 SDP image. The convolution layer extracts features from the input SDP image to form a feature map. Then, the pooling layer reduces the dimension of the features extracted by the convolution layer to reduce the feature dimension and the number of features to prevent feature overfitting. Finally, the features are stretched to one dimension through the fully connected layer, and the classification probability is calculated to obtain the classification calculation result of the model.

[0051] Preferably, after obtaining a large number of SDP images, the SDP images are manually classified and divided into three categories according to the actual corresponding vibration construction quality: under-vibration, dense, and over-vibration. After the classification is completed, it is randomly divided into a machine learning training set and a test set in a ratio of 7:3. Through repeated optimization and adjustment of parameters, hyperparameters such as CNN training learning rate and number of iterations are set until the CNN training loss converges well. The trained CNN model can accurately identify the category of SDP images, that is, the vibration construction quality: under-vibration, dense, and over-vibration.

[0052] Example 2

[0053] In order to better understand the present invention, the content of the present invention is further explained below in conjunction with an example. The concrete vibration quality cross-modal detection system and data processing method are not limited to this example.

[0054] A concrete vibration quality cross-modal detection system, such as Figure 2 As shown, including:

[0055] (1) Signal acquisition module: The signal acquisition module is mainly composed of 5 acceleration sensors, numbered #1, #2, #3, #4 and #5. All acceleration sensors are in 16-channel dynamic data acquisition mode, with an acquisition frequency of 1000Hz and an effective vibration period of 30s. Among them, #1 is a uniaxial acceleration sensor installed on the shaft wall of the vibrating rod, which is used to collect the axial motion acceleration of the vibrating rod. #2, #3, #4 and #5 are all three-axis acceleration sensors, which are installed at 100mm, 200mm, 400mm and 600mm away from the vibrating rod respectively, and are on the same horizontal line with #1, and are used to collect acceleration response data at different positions within the effective vibration radius of the vibrating rod. Figure 1 FIG. 4 is a schematic diagram of the specific installation of five acceleration sensors in this embodiment.

[0056] (2) Data processing module: This module uses the Symmetrized Dot Pattern (SDP) method to comprehensively process the signals from multiple sensors. SDP is an innovative data visualization technology that can convert time series data into a scatter plot in a polar coordinate system. This method uses the signal amplitudes xi and xi+l at the i-th and i+l-th moments in the time series x to calculate the coordinates of the i-th pair of scattered points in the polar coordinate system (r(i), θ(i), ), the coordinate calculation formula is as follows:

[0057]

[0058] Where: x max and x min are the maximum and minimum values ​​in the time series x; θ is the initial angle of the drawing; ξ is the angular domain gain factor; l is the time lag coefficient.

[0059] Symmetrical point pattern analysis continuously converts time series elements into scattered point pairs in the polar coordinate system, and finally forms two symmetrical lobes within the specified angle range [θ-ξ,θ+ξ]. Among them, θ is the symmetry axis angle of the two lobes, which determines the symmetry of the scatter plot. The angular domain gain factor ξ limits the angular domain width of the lobe in the polar coordinate system, while the time delay coefficient 1 has a significant impact on the shape of the lobe.

[0060] When there are multiple sensors, symmetric point pattern analysis can fuse the signals of all sensors into the same SDP image by cleverly adjusting the initial angle θ and the angular domain gain factor ξ. This fusion strategy not only enhances the characterization differences between different fault states, but also effectively makes up for the shortcomings of a single sensor in feature information extraction.

[0061] In the initial parameter setting of symmetrical point pattern analysis, the pixel matrix size of the SDP image is set to 64×64 to ensure sufficient image resolution. At the same time, in order to make full use of the information of the five sensors (#1, #2, #3, #4 and #5), their initial drawing angles are set to 0°, 72°, 144°, 216° and 288° respectively to ensure the uniform distribution of the scatter plot in the polar coordinate system. In addition, the angular domain gain factor ξ and the time delay coefficient 1 are uniformly set to 18° and 10 to simplify the parameter adjustment process while ensuring the analysis effect. Selecting the appropriate angular domain gain factor ξ and the time delay coefficient 1 according to the frequency components and amplitude changes of the high and low frequencies of the signal can make the final generated image features more obvious. When the angular domain gain factor ξ and the time delay coefficient 1 are uniformly set to 18° and 10, the image features generated by the SDP method are the most obvious and the resolution is the best, which is obtained by experimental parameter adjustment.

[0062] Quality analysis module: Convolutional Neural Networks (CNN) is used to analyze the quality of vibration construction. CNN is mainly composed of input layer, convolution layer, pooling layer and fully connected layer. The input layer inputs a 64×64 SDP image. The convolution layer extracts features from the input SDP image to form a feature map. Then the pooling layer reduces the dimension of the features extracted by the convolution layer to reduce the feature dimension and number to prevent overfitting. Finally, the fully connected layer stretches the features to one dimension and performs classification probability calculation to obtain the classification calculation results of the model.

[0063] After obtaining a large number of SDP images, the SDP images are manually classified and divided into three categories according to the actual corresponding vibration construction quality: under-vibration, dense, and over-vibration. After the classification is completed, it is randomly divided into machine learning training set and test set in a ratio of 7:3. Through repeated optimization and adjustment, hyperparameters such as CNN training learning rate and number of iterations are set until the CNN training loss converges well. The trained CNN model can accurately identify the category of SDP images, that is, the vibration construction quality: under-vibration, dense, and over-vibration.

[0064] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A concrete vibration quality cross-modal detection system, characterized by: It includes signal acquisition module, data processing module and quality analysis module; The signal acquisition module is used to collect the axial motion acceleration of the vibrating rod and the acceleration response data of different positions within the effective vibrating radius of the vibrating rod; The data processing module comprehensively processes the signals from the sensors through symmetrical point pattern analysis; The quality analysis module uses convolutional neural networks to analyze the quality of vibration construction.

2. According to claim 1, a concrete vibration quality cross-modal detection system is characterized by: The signal acquisition module includes several acceleration sensors, all of which are in 16-channel dynamic data acquisition mode, with an acquisition frequency of 1000 Hz and an effective vibration period of 30 s.

3. A concrete vibration quality cross-modal detection system according to claim 2, characterized in that: The acceleration sensor includes an acceleration sensor and a plurality of three-axis acceleration sensors, wherein the uniaxial acceleration sensor is installed on the shaft wall of the vibrating rod and is used to collect the axial motion acceleration of the vibrating rod; Several three-axis acceleration sensors are installed at different positions from the vibrating rod and are on the same horizontal line as the installation position of the uniaxial acceleration sensor. They are used to collect acceleration response data at different positions within the effective vibration radius of the vibrating rod.

4. A data processing method using a cross-modal detection system for concrete vibration quality according to any one of claims 1 to 3, characterized in that: The following steps are involved: S1. Using a signal acquisition module composed of several acceleration sensors, respectively collect the axial motion acceleration of the vibrating rod and the acceleration response data of different positions within the effective vibration radius; S2, using symmetrical point pattern analysis to process multi-source sensor signals, converting the time series into a polar coordinate scatter plot, and by adjusting the initial parameters of the symmetrical point pattern analysis, fusing all sensor signals into the same SDP image to increase the representation differences of different fault states; S3. Use convolutional neural network to analyze the vibration construction quality of SDP images. Through manual classification and machine learning training, the CNN model can accurately identify the category of SDP images, that is, determine whether the vibration construction quality is under-vibration, dense or over-vibration.

5. The data processing method of a cross-modal detection system for concrete vibration quality according to claim 4 is characterized by: The specific method of symmetrical point pattern analysis is: using the signal amplitudes xi and xi+l at the i-th moment and the i+l-th moment in the time series x, calculate the coordinates of the i-th pair of scattered points in the polar coordinate system The coordinate calculation formula is: Where: x max and x min is the maximum and minimum value in the time series x; θ is the initial angle of the drawing; ξ is the angular domain gain factor; l is the time delay coefficient.

6. The data processing method of a cross-modal detection system for concrete vibration quality according to claim 4 is characterized by: Symmetrical point pattern analysis continuously converts time series elements into scattered point pairs in the polar coordinate system, and finally forms two symmetrical lobes within the specified angle range [θ-ξ,θ+ξ], where θ is the symmetry axis angle of the two lobes, θ determines the symmetry property of the scatter plot, the angular domain gain factor ξ limits the angular domain width of the lobe in the polar coordinate system, and the time delay coefficient l affects the morphology of the lobe.

7. The data processing method of the cross-modal detection system for concrete vibration quality according to claim 4 is characterized by: When there are several sensors, the symmetric point pattern analysis fuses the signals of all sensors into the same SDP image by adjusting the initial angle θ of the drawing and the angular domain gain factor ξ.

8. The data processing method of a cross-modal detection system for concrete vibration quality according to claim 4 is characterized by: In the initial parameter setting of the symmetric point pattern analysis, the pixel matrix size of the SDP image was set to 64 × 64 to ensure sufficient image resolution; In order to make full use of the information of each sensor, the initial drawing angle of each sensor is set to 0°, 72°, 144°, 216° and 288° respectively to ensure the uniform distribution of the scatter plot in the polar coordinate system; in addition, the angular domain gain factor ξ and the time delay coefficient l are uniformly set to 18° and 10 to ensure the analysis effect and simplify the parameter adjustment process.

9. The data processing method of a cross-modal detection system for concrete vibration quality according to claim 4 is characterized by: CNN is mainly composed of input layer, convolution layer, pooling layer and fully connected layer. The input layer inputs a 64×64 SDP image. The convolution layer extracts features from the input SDP image to form a feature map. Then the pooling layer reduces the dimension of the features extracted by the convolution layer to reduce the feature dimension and number to prevent overfitting. Finally, the fully connected layer stretches the features to one dimension and calculates the classification probability to obtain the classification calculation result of the model.

10. The data processing method of the cross-modal detection system for concrete vibration quality according to claim 9 is characterized by: After obtaining a large number of SDP images, the SDP images were manually classified into three categories according to the actual corresponding vibration construction quality: under-vibration, dense and over-vibration; After the classification is completed, the data is randomly divided into machine learning training set and test set in a ratio of 7:

3. Through repeated optimization and adjustment, hyperparameters such as CNN training learning rate and number of iterations are set until the CNN training loss converges well. The trained CNN model can identify the quality of vibration construction: under-vibration, denseness and over-vibration.

Images