PPVC modular wall grouting quality control method based on ultrasonic detection

Abstract

The invention provides a PPVC modular wall grouting quality control method based on ultrasonic detection, belongs to the technical field of grouting quality control, breaks through subjectivity and limitation of a traditional knocking method, adopts a scientific high-frequency mechanical wave detection principle, realizes accurate detection of grouting fullness, and improves the grouting quality. And the reliability and the accuracy of the quality control result of the grouting fullness are greatly improved. The unique detection process and data analysis method provide a brand-new and efficient solution for the detection and control of the PPVC wall node grouting quality, fill the blank of the precise detection method in the field, have significant innovativeness and practical value, and are worthy of popularization and application. By means of the scientific principle, the accurate detection effect and the unique patent creativity of the ultrasonic detection method, it can be effectively guaranteed that the PPVC wall joint grouting fullness in an apartment construction project reaches 100%, quality control is more accurate, and a solid guarantee is provided for the quality of a project subject.

Description

Technical Field

[0001] This invention belongs to the field of grouting quality control technology, specifically relating to a method for controlling the grouting quality of PPVC modular walls based on ultrasonic testing. Background Technology

[0002] PPVC modular wall grouting, also known as PPVC wall grouting technology, is a core process for achieving structural connections in prefabricated buildings. As mentioned in the existing technical solution with patent publication number "CN117488997A", ​​PPVC modular wall grouting is required.

[0003] In apartment construction projects, a PPVC (precast and pre-finished modular building) + core tube cast-in-place structural system is used. The joint between every two PPVC wall joints is typically 20mm, requiring grouting after installation. The grouting quality of the PPVC wall joints directly affects the overall structural integrity. To ensure the grouting quality of the main structure, a 100% grout fullness qualification rate is required. The traditional tapping method for testing grout fullness has the following drawbacks:

[0004] Traditional tapping methods rely on inspectors tapping PPVC wall joints (PPVC modules) with tools and judging the fullness of grouting by listening to the differences in sound. However, this method is highly subjective; different inspectors have varying hearing sensitivity and experience, leading to biased judgments and potential misjudgments. Furthermore, for minor cases of incomplete grouting, the change in sound is not obvious, making accurate identification difficult and compromising the reliability of the test results. Therefore, it cannot guarantee that the grouting fullness truly reaches 100%. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a quality control method for PPVC modular wall grouting based on ultrasonic testing. This method effectively avoids the defects of existing technologies, such as the highly subjective nature of the tapping method, the varying hearing sensitivity and experience of different testing personnel, the tendency for biased judgments of sound, the ease of misjudgment, the inability to ensure the reliability of test results, and the inability to guarantee that the grouting fullness truly reaches 100%.

[0006] The present invention employs the following technical solution.

[0007] A method for quality control of PPVC modular wall grouting based on ultrasonic testing, comprising:

[0008] Step 1: Construct an ultrasonic probe and ultrasonic receiver connected to the control unit to form a detection and control instrument;

[0009] Step 2: After the PPVC wall joint is grouted, the surface of the PPVC wall joint is cleaned before ultrasonic testing to remove dust and debris.

[0010] Step 3: Place the ultrasonic probe on the preset detection point of the PPVC wall node, control the ultrasonic probe to emit ultrasonic waves through the control unit and receive the reflected wave signal by the ultrasonic receiver. The reflected wave signal is also transmitted to the control unit.

[0011] Step 4: The control unit processes and analyzes the received reflected wave signal through its data analysis software to obtain the quality control results of grout fullness.

[0012] Furthermore, the ultrasonic probe and ultrasonic receiver are combined into an integrated ultrasonic transceiver probe.

[0013] Furthermore, step 4 specifically includes:

[0014] To obtain the quality control result of whether the grouting quality of PPVC modular wall is full, the data analysis software uses a deep learning algorithm based on convolutional neural network to analyze the reflected wave signal, thereby obtaining the quality control result of whether the grouting quality of PPVC modular wall is full. Full means that the grouting fullness reaches 100%.

[0015] Furthermore, step 4 specifically includes:

[0016] Step 4-1: Preprocess the reflected wave signal;

[0017] Step 4-2: Design the CNN network structure and extract features;

[0018] Step 4-3: Perform classification and judgment of fully connected layers;

[0019] Step 4-4: Perform network training and optimization.

[0020] Furthermore, step 4-1 specifically includes:

[0021] Step 4-1-1, Signal format conversion: Convert the reflected wave signal collected by the ultrasonic receiver into a digital signal to obtain one-dimensional time series data with one dimension;

[0022] Step 4-1-2, Noise Filtering: A wavelet thresholding denoising algorithm is used to filter noise from the digital signal. This algorithm selects the db4 wavelet basis for a 3-level decomposition and sets thresholds for high-frequency coefficients. ,in The standard deviation of noise;

[0023] Step 4-1-3, Signal Standardization: Normalize the noise-filtered digital signal. The normalization formula is as follows: ,in The digital signal after normalization. The mean of the digital signal after noise filtering. This represents the standard deviation of the digital signal after noise filtering.

[0024] Furthermore, step 4-2 specifically includes:

[0025] Step 4-2-1: Convolutional layer parameter settings;

[0026] Step 4-2-2: Pooling layer parameter settings;

[0027] Step 4-2-3: Perform the feature extraction process.

[0028] Furthermore, step 4-2-1 specifically includes:

[0029] The first convolutional layer uses 16 small convolutional kernels of size 1×7 with a stride of 1 and "same" padding. This layer is used to capture local abrupt changes in the normalized digital signal. The ReLU activation function is used, and the formula for the ReLU function is... ;

[0030] The second convolutional layer uses 32 medium convolutional kernels of size 1×15 with a stride of 1 and padding mode of "same". This layer is used to extract the mid-range fluctuation features in the normalized digital signal.

[0031] The second convolutional layer uses 64 large convolutional kernels of size 1×31 with a stride of 1 and padding mode of "same". This layer is used to capture the overall trend features of the normalized digital signal.

[0032] Furthermore, step 4-2-2 specifically includes:

[0033] • Each convolutional layer is followed by a max pooling layer. The max pooling layer has a pooling window size of 1×2 and a stride of 2. Downsampling is performed by selecting the maximum value within the pooling window.

[0034] Furthermore, step 4-2-3 specifically includes:

[0035] The convolutional kernel extracts features by performing a cross-correlation operation with the normalized digital input signal. Its calculation formula is as follows:

[0036] ;

[0037] in, For the first The convolutional kernel at the _th ... Output at each position, For convolution kernel weights, The first normalized digital signal is the input signal. One sampling point, For bias terms, is the kernel size.

[0038] Furthermore, step 4-3 specifically includes:

[0039] Step 4-3-1, Feature flattening: Convert the 1×128 feature map after the third pooling layer into a 128-dimensional one-dimensional vector, which is used as the input of the fully connected layer;

[0040] Step 4-3-2, Fully Connected Layer Structure: Two fully connected layers are set up. The first fully connected layer contains 64 neurons, and the second fully connected layer contains 2 neurons. The first and second fully connected layers correspond to the "full" and "sparse" outputs, respectively. The first fully connected layer uses the ReLU activation function, and the second fully connected layer uses the Softmax function. The formula for the Softmax function is:

[0041] ;

[0042] in It is a natural constant. For the second layer of neurons One input, For the sample to belong to the first The probability of a class is used to determine the class, and the class with the higher probability value is the final classification result.

[0043] Furthermore, step 4-4 specifically includes:

[0044] Step 4-4-1, Training dataset construction: Collect 10,000 sets of labeled data, including 5,000 sets of normalized digital signals of full grouting from history and 5,000 sets of normalized digital signals of incomplete grouting with defects of different sizes. Divide the 10,000 sets of labeled data into training set and validation set in a 7:3 ratio.

[0045] Step 4-4-2, Constructing the Loss Function and Optimizer: The loss function used is the cross-entropy loss function, the formula for which is... ,in For the first A real label, For the sample to belong to the first The probability of the class; the optimizer is Adam, the learning rate of Adam is set to 0.001, and the decay rate of Adam is... =0.9、 =0.999;

[0046] Step 4-4-3, Model Validation: The trained model is validated against the normalized digital signal to obtain the quality control results of whether the grouting quality of the PPVC modular wall is full.

[0047] The beneficial effects of the present invention are as follows, compared with the prior art:

[0048] This invention overcomes the subjectivity and limitations of traditional tapping methods, employing a scientific high-frequency mechanical wave detection principle to achieve precise detection of grout fullness, significantly improving the reliability and accuracy of grout fullness quality control results. Its unique detection process and data analysis methods provide a novel and efficient solution for the detection and control of PPVC wall joint grouting quality, filling a gap in precise detection methods in this field. It possesses significant innovation and practical value. Furthermore, the ultrasonic detection method, with its scientific principles, precise detection effects, and unique patented inventiveness, can effectively ensure 100% grout fullness of PPVC wall joints in apartment construction projects, providing a more precise quality control guarantee for the overall project quality. Attached Figure Description

[0049] Figure 1 This is a flowchart of the PPVC modular wall grouting quality control method based on ultrasonic testing in this invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, other embodiments obtained by those skilled in the art without creative effort are all within the protection scope of this invention.

[0051] like Figure 1 As shown, a method for quality control of PPVC modular wall grouting based on ultrasonic testing includes:

[0052] This invention employs ultrasonic testing to control the grouting quality of PPVC modular walls. Ultrasonic testing utilizes high-frequency mechanical waves, offering high scientific rigor and accuracy. The principle is as follows: when high-frequency mechanical waves (ultrasound) propagate in different media, they undergo reflection, refraction, and attenuation. When ultrasound propagates through a fully grouted PPVC wall joint, the reflected wave characteristics are relatively consistent due to the uniform medium and stable propagation path. However, when the grouting is incomplete and gaps exist, the ultrasound will experience significant reflection and scattering at these gaps, causing changes in the amplitude and propagation time of the reflected wave. By receiving and analyzing these reflected wave signals using specialized testing instruments, the fullness of the grouting in the PPVC modular wall can be accurately determined.

[0053] Step 1: Construct an ultrasonic probe and a high-sensitivity ultrasonic receiver connected to the control unit to form a detection and control instrument. The detection instrument is used to receive and analyze the reflected wave signal, so as to accurately determine whether the wall grouting formed by the PPVC module is full.

[0054] Step 2: After the PPVC wall joint is grouted, the surface of the PPVC wall joint is cleaned before ultrasonic testing to remove dust, debris, etc., to ensure that the ultrasonic probe can be well coupled with the surface of the PPVC wall joint.

[0055] Step 3: Place the ultrasonic probe on the preset detection point of the PPVC wall node, control the ultrasonic probe to emit ultrasonic waves through the control unit and receive the reflected wave signal by the ultrasonic receiver. The reflected wave signal is also transmitted to the control unit.

[0056] After the ultrasonic probe is placed at the preset detection point, the emitted ultrasonic signal propagates according to the wave equation. Assume the speed of ultrasonic wave propagation in a homogeneous medium (such as an ideal saturated grout) is... Its propagation process can be described by the following wave equation: ,in Let be the displacement function of the ultrasonic wave. For time, It's the Laplace operator. When ultrasound encounters an unfilled gap, its propagation speed changes due to the altered properties of the medium. Changes will occur, and reflection and scattering will occur. According to Huygens' principle, each point on the wavefront can be regarded as a new wave source, emitting wavelets in all directions, thus generating reflected and scattered waves.

[0057] Step 4: The control unit processes and analyzes the received reflected wave signal using its data analysis software to obtain the quality control result of grout fullness. The PPVC wall node is the PPVC module. The control unit can be a microcontroller, PLC, or FPGA chip.

[0058] In a preferred but non-limiting embodiment of the present invention, the ultrasonic probe and the ultrasonic receiver are an integrated ultrasonic transceiver probe.

[0059] In a preferred but non-limiting embodiment of the present invention, step 4 specifically includes:

[0060] To obtain the quality control result of whether the grouting quality of PPVC modular wall is full, the data analysis software uses a deep learning algorithm based on convolutional neural network (CNN) to analyze the reflected wave signal, thereby obtaining the quality control result of whether the grouting quality of PPVC modular wall is full. Full means that the grouting fullness reaches 100%.

[0061] This deep learning algorithm, based on Convolutional Neural Networks (CNNs), constructs multiple convolutional and pooling layers. The convolutional kernels in the convolutional layers slide across the signal data, extracting features at different scales. For example, smaller kernels capture local details of the signal, while larger kernels capture more macroscopic features. Pooling layers, through downsampling, reduce the amount of data while preserving key features, preventing overfitting. The processed features are then input into a fully connected layer for classification, outputting a result indicating whether the data is fully or partially filled.

[0062] In a preferred but non-limiting embodiment of the present invention, step 4 specifically includes:

[0063] In the data analysis software processing stage, the analysis of reflected signals using deep learning algorithms based on convolutional neural networks (CNNs) is achieved by constructing a complete workflow of "signal preprocessing - feature extraction - classification judgment," as detailed below:

[0064] Step 4-1: Preprocess the reflected wave signal;

[0065] In a preferred but non-limiting embodiment of the present invention, step 4-1 specifically includes:

[0066] Step 4-1-1, Signal Format Conversion: Convert the reflected wave signal (voltage changing over time waveform, the reflected wave signal is an analog signal) acquired by the high-precision ultrasonic receiver into a digital signal. Set the sampling frequency to 10MHz to ensure that the high-frequency reflection characteristics are fully preserved, and obtain one-dimensional time series data with one dimension (N is the number of sampling points, usually 1024).

[0067] Step 4-1-2, Noise Filtering: A wavelet thresholding denoising algorithm is used to filter noise from the digital signal. This algorithm selects the db4 wavelet basis for a 3-level decomposition and sets thresholds for high-frequency coefficients. ,in The noise standard deviation is estimated by the first 100 sampling points of the digital signal. This standard deviation filters out noise such as mechanical vibration and electromagnetic interference in the construction environment, improving the signal-to-noise ratio to over 30dB.

[0068] Step 4-1-3, Signal Standardization: Normalize the noise-filtered digital signal. The normalization formula is as follows: ,in The digital signal after normalization. The mean of the digital signal after noise filtering. Let be the standard deviation of the noise-filtered digital signal. The noise-filtered digital signal is compressed to the [-1,1] interval to eliminate the influence of signal amplitude differences on network training.

[0069] Step 4-2: Design the CNN network structure and extract features;

[0070] In a preferred but non-limiting embodiment of the present invention, step 4-2 specifically includes:

[0071] Step 4-2-1: Convolutional layer parameter settings;

[0072] In a preferred but non-limiting embodiment of the present invention, step 4-2-1 specifically includes:

[0073] The first convolutional layer uses 16 small convolutional kernels of size 1×7 with a stride of 1 and "same" padding (keeping the output dimension consistent with the input). This layer is primarily used to capture local abrupt changes in the normalized digital signal, such as the starting point of the reflected wave and the peak amplitude (corresponding to the boundary reflection of the incomplete grouting area). The ReLU activation function is used, and the formula for the ReLU function is... This enhances the network's nonlinear fitting ability;

[0074] The second convolutional layer uses 32 1×15 kernels with a stride of 1 and a padding mode of "same". This layer is used to extract the mid-term fluctuation features in the normalized digital signal, such as the duration of the reflected wave and the amplitude attenuation rate (reflecting the size and distribution range of the grouting defect).

[0075] The second convolutional layer uses 64 large convolutional kernels of size 1×31 with a stride of 1 and a padding mode of "same". This layer is used to capture the overall trend characteristics of the normalized digital signal, such as the periodicity of the waveform and the energy distribution (corresponding to the overall uniformity of the grouting body).

[0076] Step 4-2-2: Pooling layer parameter settings;

[0077] In a preferred but non-limiting embodiment of the present invention, step 4-2-2 specifically includes:

[0078] • Each convolutional layer is followed by a max-pooling layer. The max-pooling layer has a pooling window size of 1×2 and a stride of 2. Downsampling is performed by selecting the maximum value within the pooling window. For example, the 1×1024 feature map output from the first convolutional layer becomes a 1×512 feature map after pooling, and the feature map dimension is compressed to 1×128 after the third pooling layer. This reduces the amount of data (compression ratio of up to 8 times) while retaining key features and effectively preventing network overfitting.

[0079] Step 4-2-3: Perform the feature extraction process.

[0080] In a preferred but non-limiting embodiment of the present invention, step 4-2-3 specifically includes:

[0081] The convolutional kernel extracts features by performing a cross-correlation operation with the normalized digital input signal. Its calculation formula is as follows:

[0082] ;

[0083] in, For the first The convolutional kernel at the _th ... Output at each position, For convolution kernel weights, The first normalized digital signal is the input signal. One sampling point, For bias terms, The kernel size is denoted by . By stacking multiple convolutional layers, the network automatically learns multi-scale features from local details to global trends, without the need for manual feature design.

[0084] Step 4-3: Perform classification and judgment of fully connected layers;

[0085] In a preferred but non-limiting embodiment of the present invention, step 4-3 specifically includes:

[0086] Step 4-3-1, Feature flattening: Convert the 1×128 feature map after the third pooling layer into a 128-dimensional one-dimensional vector, which is used as the input of the fully connected layer;

[0087] Step 4-3-2, Fully Connected Layer Structure: Two fully connected layers are set up. The first fully connected layer contains 64 neurons, and the second fully connected layer contains 2 neurons. The first and second fully connected layers correspond to the "full" and "sparse" outputs, respectively. The first fully connected layer uses the ReLU activation function, and the second fully connected layer uses the Softmax function. The formula for the Softmax function is:

[0088] ;

[0089] in It is a natural constant. For the second layer of neurons One input, For the sample to belong to the first The probability of a class is used to determine the class, and the class with the higher probability value is the final classification result.

[0090] Step 4-4: Perform network training and optimization.

[0091] In a preferred but non-limiting embodiment of the present invention, step 4-4 specifically includes:

[0092] Step 4-4-1, Training dataset construction: Collect 10,000 sets of labeled data, including 5,000 sets of normalized digital signals of full grouting from history and 5,000 sets of normalized digital signals of incomplete grouting with defects of different sizes. Divide the 10,000 sets of labeled data into training set and validation set in a 7:3 ratio.

[0093] Step 4-4-2, Constructing the Loss Function and Optimizer: The loss function used is the cross-entropy loss function, the formula for which is... ,in For the first A real label, For the sample to belong to the first The probability of the class; the optimizer is Adam, the learning rate of Adam is set to 0.001, and the decay rate of Adam is... =0.9、 =0.999, and the network parameters are updated through backpropagation, so that the loss function converges to below 0.01;

[0094] Step 4-4-3, Model Validation: The trained model is validated against the normalized digital signal obtained in Step 4-1 to obtain the quality control results of whether the grouting quality of the PPVC modular wall is full. The trained model achieves an accuracy of 99.2% on the normalized digital signal obtained in Step 4-1, a precision (correct recognition rate of full samples) of 99.5%, and a recall rate (correct recognition rate of incomplete samples) of 98.8%, meeting the high precision requirements of engineering testing.

[0095] Step 4 utilizes a customized CNN structure to automatically extract multi-scale features from the reflected wave signal, avoiding the limitations of traditional manually designed features (such as peak amplitude and rise time). Small convolutional kernels are sensitive to the reflection details of minute defects (such as 1mm³ voids), while large convolutional kernels accurately capture the overall waveform changes in large areas of incomplete grouting. Combined with downsampling of pooling layers and probabilistic judgment of fully connected layers, the model maintains stable detection performance even in complex construction environments. Compared to existing signal analysis methods based on empirical thresholds, its innovation lies in: leveraging the end-to-end learning capabilities of deep learning to directly mine latent features related to grout fullness from the original signal, improving detection accuracy by more than three orders of magnitude. Furthermore, it exhibits strong generalization ability, adapting to the detection needs of different grouting material ratios and environmental conditions.

[0096] Compared to traditional methods, the method of this invention overcomes the subjectivity and low accuracy of traditional detection methods by utilizing scientific wave equations and precise algorithms. Through quantitative analysis of key parameters of the reflected signal and processing with advanced deep learning algorithms, it can accurately identify tiny areas of incomplete grouting, greatly improving the resolution and reliability of the detection. For example, in actual engineering inspections, traditional tapping methods may not be able to detect grouting defects smaller than 10 mm³, while this method can accurately detect defects smaller than 1 mm³, achieving a qualitative leap in detection accuracy. This provides a new and efficient solution for the quality inspection of grouting at PPVC wall joints, filling the gap in high-precision detection methods in this field and achieving significantly better technical results than existing technologies.

[0097] The beneficial effects of the present invention are as follows, compared with the prior art:

[0098] This invention overcomes the subjectivity and limitations of traditional tapping methods, employing a scientific high-frequency mechanical wave detection principle to achieve precise detection of grout fullness, significantly improving the reliability and accuracy of grout fullness quality control results. Its unique detection process and data analysis methods provide a novel and efficient solution for the detection and control of PPVC wall joint grouting quality, filling a gap in precise detection methods in this field. It possesses significant innovation and practical value. Furthermore, the ultrasonic detection method, with its scientific principles, precise detection effects, and unique patented inventiveness, can effectively ensure 100% grout fullness of PPVC wall joints in apartment construction projects, providing a more precise quality control guarantee for the overall project quality.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention without departing from the spirit and scope of the present invention. Any modifications or equivalent substitutions should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for quality control of PPVC modular wall grouting based on ultrasonic testing, characterized in that, include: Step 1: Construct an ultrasonic probe and ultrasonic receiver connected to the control unit to form a detection and control instrument; Step 2: After the PPVC wall joint is grouted, the surface of the PPVC wall joint is cleaned before ultrasonic testing to remove dust and debris. Step 3: Place the ultrasonic probe on the preset detection point of the PPVC wall node, control the ultrasonic probe to emit ultrasonic waves through the control unit and receive the reflected wave signal by the ultrasonic receiver. The reflected wave signal is also transmitted to the control unit. Step 4: The control unit processes and analyzes the received reflected wave signal through its data analysis software to obtain the quality control results of grout fullness; Step 4 specifically includes: Step 4-1: Preprocess the reflected wave signal; Step 4-2: Design the CNN network structure and extract features; Step 4-3: Perform classification and judgment of fully connected layers; Step 4-4: Perform network training and optimization; Step 4-1 specifically includes: Step 4-1-1, Signal format conversion: Convert the reflected wave signal collected by the ultrasonic receiver into a digital signal to obtain one-dimensional time series data with one dimension; Step 4-1-2, Noise Filtering: A wavelet thresholding denoising algorithm is used to filter noise from the digital signal. This algorithm selects the db4 wavelet basis for a 3-level decomposition and sets thresholds for high-frequency coefficients. ,in The standard deviation of noise; Step 4-1-3, Signal Standardization: Normalize the noise-filtered digital signal. The normalization formula is as follows: ,in The digital signal after normalization. The mean of the digital signal after noise filtering. This represents the standard deviation of the digital signal after noise filtering.

2. The method for quality control of PPVC modular wall grouting based on ultrasonic testing according to claim 1, characterized in that, The ultrasonic probe and ultrasonic receiver are combined into an integrated ultrasonic transceiver probe.

3. The method for quality control of PPVC modular wall grouting based on ultrasonic testing according to claim 2, characterized in that, Step 4 specifically includes: To obtain the quality control result of whether the grouting quality of PPVC modular wall is full, the data analysis software uses a deep learning algorithm based on convolutional neural network to analyze the reflected wave signal, thereby obtaining the quality control result of whether the grouting quality of PPVC modular wall is full. Full means that the grouting fullness reaches 100%.

4. The method for quality control of PPVC modular wall grouting based on ultrasonic testing according to claim 3, characterized in that, Step 4-2 specifically includes: Step 4-2-1: Convolutional layer parameter settings; Step 4-2-2: Pooling layer parameter settings; Step 4-2-3: Perform the feature extraction process.

5. The method for quality control of PPVC modular wall grouting based on ultrasonic testing according to claim 4, characterized in that, Step 4-2-1 specifically includes: The first convolutional layer uses 16 small convolutional kernels of size 1×7 with a stride of 1 and "same" padding. This layer is used to capture local abrupt changes in the normalized digital signal. The ReLU activation function is used, and the formula for the ReLU function is... ; The second convolutional layer uses 32 medium convolutional kernels of size 1×15 with a stride of 1 and padding of "same". This layer is used to extract the mid-range fluctuation features in the normalized digital signal. The second convolutional layer uses 64 large convolutional kernels of size 1×31 with a stride of 1 and padding of "same". This layer is used to capture the overall trend features of the normalized digital signal.

6. The method for quality control of PPVC modular wall grouting based on ultrasonic testing according to claim 5, characterized in that, Step 4-2-2 specifically includes: Each convolutional layer is followed by a max pooling layer. The pooling window size of the max pooling layer is 1×2, and the stride is 2. Downsampling is performed by selecting the maximum value within the pooling window. Step 4-2-3 specifically includes: The convolutional kernel extracts features by performing a cross-correlation operation with the normalized digital input signal. Its calculation formula is as follows: ； in, For the first The convolutional kernel at the _th ... Output at each position, For convolution kernel weights, The first normalized digital signal is the input signal. One sampling point, For bias terms, is the kernel size.

7. The method for quality control of PPVC modular wall grouting based on ultrasonic testing according to claim 6, characterized in that, Step 4-3 specifically includes: Step 4-3-1, Feature flattening: Convert the 1×128 feature map after the third pooling layer into a 128-dimensional one-dimensional vector, which is used as the input of the fully connected layer; Step 4-3-2, Fully Connected Layer Structure: Two fully connected layers are set up. The first fully connected layer contains 64 neurons, and the second fully connected layer contains 2 neurons. The first and second fully connected layers correspond to the "full" and "sparse" outputs, respectively. The first fully connected layer uses the ReLU activation function, and the second fully connected layer uses the Softmax function. The formula for the Softmax function is: ； in It is a natural constant. For the second layer of neurons One input, For the sample to belong to the first The probability of a class is used to determine the class, and the class with the higher probability value is the final classification result.

8. The method for quality control of PPVC modular wall grouting based on ultrasonic testing according to claim 7, characterized in that, Step 4-4 specifically includes: Step 4-4-1, Training dataset construction: Collect 10,000 sets of labeled data, including 5,000 sets of normalized digital signals of full grouting from history and 5,000 sets of normalized digital signals of incomplete grouting with defects of different sizes. Divide the 10,000 sets of labeled data into training set and validation set in a 7:3 ratio. Step 4-4-2, Constructing the Loss Function and Optimizer: The loss function used is the cross-entropy loss function, the formula for which is... ,in For the first A real label, For the sample to belong to the first The probability of the class; the optimizer is Adam, the learning rate of Adam is set to 0.001, and the decay rate of Adam is... =0.9、 =0.999; Step 4-4-3, Model Validation: The trained model is validated against the normalized digital signal to obtain the quality control results of whether the grouting quality of the PPVC modular wall is full.

Images