Separation detection device in deep foundation pit concrete pouring process

Abstract

The invention provides a segregation detection device in a deep foundation pit concrete pouring process, and belongs to the technical field of concrete construction, the segregation detection device comprises a pipe body, monitoring units and a single-chip microcomputer, a plurality of circular through holes are formed in the side wall of the pipe body, the plurality of circular through holes are uniformly arranged in two rows or three rows in the side wall, one monitoring unit is embedded in each circular through hole, and the single-chip microcomputer is connected with the monitoring units. The monitoring unit comprises a tempered glass sheet, a concave lens, an isolation glass sheet, a light condensation sheet and an optical fiber bundle, a round hole is formed in the bottom of the light condensation sheet, one end of the optical fiber bundle is embedded in the round hole to serve as an acquisition end, and a controller and an image acquisition device are arranged on the single chip microcomputer; the other end of the optical fiber bundle serves as an output end to be connected to an image collector, the image collector is used for receiving optical signals transmitted by the optical fiber bundle and converting the optical signals into electric signals, and area images of a monitoring area corresponding to the monitoring unit are obtained after analog-to-digital conversion and output to a controller. And a segregation result is calculated by using a concrete pouring process segregation analysis module in the controller.

Description

Technical Field

[0001] The invention belongs to the technical field of concrete construction, and in particular relates to a segregation detection device during a deep foundation pit concrete pouring process. Background Art

[0002] As one of the most widely used materials in modern construction projects, the quality of concrete is directly related to the safety and durability of buildings. In the process of deep foundation pit construction, the quality of concrete pouring is particularly important, because deep foundation pit structures often bear key load-bearing and waterproofing functions. However, during the pouring process of deep foundation pit concrete, due to its large height and long pouring time, segregation is prone to occur, that is, the components in the concrete (cement, fine aggregate, coarse aggregate) separate under the action of gravity, resulting in uneven concrete strength and reduced durability.

[0003] At present, the detection of concrete segregation in deep foundation pit pouring mainly relies on the following methods:

[0004] 1. Visual inspection method: Engineers judge the degree of segregation by observing the aggregate distribution on the concrete surface and cut surface. This method is simple and intuitive, but it is highly dependent on the inspector's experience, easily affected by subjective factors, and cannot achieve real-time monitoring.

[0005] 2. Coring method: After the concrete hardens, core samples of concrete are drilled from different locations, and the degree of segregation is evaluated by analyzing the distribution of aggregates in the core samples. Although this method is highly accurate, it is destructive and can only be tested after the fact, and it is impossible to detect and solve problems during the pouring process in a timely manner.

[0006] In other words, the current segregation detection in the deep foundation pit concrete pouring process has technical problems such as reliance on the experience of engineering personnel and difficulty in real-time evaluation. Summary of the invention

[0007] In view of this, the present invention provides a segregation detection device for deep foundation pit concrete pouring process, which can solve the technical problems that the current segregation detection in the deep foundation pit concrete pouring process relies on the experience of engineers and is difficult to evaluate in real time.

[0008] The present invention is achieved in that:

[0009] The present invention provides a segregation detection device for a deep foundation pit concrete pouring process, which is used for detecting the segregation of concrete during the concrete pouring process. The device comprises a tube body, a monitoring unit and a single chip computer. A plurality of circular through holes are provided on the side wall of the tube body. The plurality of circular through holes are evenly arranged in 2 or 3 rows on the side wall. A monitoring unit is embedded in each circular through hole. The monitoring unit comprises a tempered glass sheet, a concave lens, an isolation glass sheet, a condenser sheet and an optical fiber bundle. A circular hole is provided at the bottom of the condenser sheet. One end of the optical fiber bundle is embedded in the circular hole as a collection end. The single chip computer is provided with a controller and image collectors at least the number of the monitoring units. The other end of the optical fiber bundle is embedded in the circular hole as a collection end. The end is connected to the image collector as an output end, and the image collector is used to receive the optical signal transmitted by the optical fiber bundle and convert it into an electrical signal. After analog-to-digital conversion, the regional image of the monitoring area corresponding to the monitoring unit is obtained and output to the controller; the cross-section of the condenser is parabolic, and the focus of the concave lens located on the inner side of the tube body and the focus of the condenser are both located at the geometric center of the surface of the collection end; a concrete pouring process separation and analysis module is provided in the controller, which is used to calculate the separation result of the concrete pouring process based on the regional images collected by multiple image collectors and the pre-trained separation analysis model, and transmit it to the host computer through the wireless transmission module on the single-chip microcomputer.

[0010] Wherein, the outer diameter of the tube body is 3~5cm.

[0011] Furthermore, the spacing of each column of circular through holes, the radius of the tempered glass sheet, the shape parameters of the concave lens, the shape parameters of the focusing sheet and the parameters of the optical fiber bundle are all optimized and limited by an optimization equation group, and the optimization equation group includes an optical imaging equation, a spatial sampling equation, a signal-to-noise ratio equation, a resolution equation and a depth detection equation.

[0012] The optical imaging equation takes into account the imaging quality of the optical system and is used to optimize the parameters of the concave lens and the focusing plate; the spatial sampling equation takes into account the spatial sampling frequency of the image and is used to optimize the spacing of the circular through holes and the radius of the tempered glass sheet; the signal-to-noise ratio equation takes into account the signal-to-noise ratio of the image and is used to optimize the parameters of the optical fiber bundle; the resolution equation takes into account the spatial resolution of the image and is used to comprehensively optimize the parameters of each optical element; the depth detection equation takes into account the detection depth inside the concrete and is used to optimize the outer diameter of the tube body and the size of the through hole.

[0013] The following is a detailed description of the optimization equations:

[0014] 1. Optical imaging equation:

[0015] ;

[0016] In the formula, is the focal length of the concave lens; is the object distance; is the image distance; is the refractive index of the lens material; is the center thickness of the lens; , is the radius of curvature of the two surfaces of the lens;

[0017] Optimization of shape parameters of concentrator:

[0018] ;

[0019] In the formula, is the focus parameter; , are the coordinates of the point on the parabola;

[0020] 2. Spatial sampling equation:

[0021] ;

[0022] In the formula, is the minimum resolvable distance; is the wavelength of light; is the focal length of the system; is the numerical aperture;

[0023] 3. Signal-to-noise ratio equation:

[0024] ;

[0025] In the formula, is the signal power; is the noise power

[0026] Fiber bundle transmission loss:

[0027] ;

[0028] In the formula, is the output power; is the input power; is the attenuation coefficient; is the fiber length;

[0029] 4. Resolution equation:

[0030] ;

[0031] In the formula, is the system resolution; is the wavelength of light; is the focal length of the system; is the effective aperture diameter;

[0032] 5. Depth detection equation:

[0033] ;

[0034] In the formula, is the maximum detection depth; is the speed of light; is the pulse width; is the relative dielectric constant of concrete;

[0035] Relationship between the outer diameter of the tube and the size of the through hole:

[0036] ;

[0037] In the formula, is the through hole area; is the filling factor (0 < k < 1); is the outer radius of the tube.

[0038] Furthermore, the specific steps of performing optimization and limitation by optimizing the set of equations include:

[0039] Establish the objective function and sum the weighted equations;

[0040] Set the initial values ​​and constraints of each parameter;

[0041] Particle swarm algorithm is used for iterative optimization;

[0042] When the number of iterations reaches the preset value or the objective function converges, the optimization result is output.

[0043] Furthermore, the concrete pouring process separation and analysis module specifically performs the following steps:

[0044] Acquire the regional image collected by each image collector;

[0045] Performing preprocessing including denoising, enhancement and correction on the regional image to obtain a preprocessed image;

[0046] All pre-processed images are spliced ​​using image registration and fusion methods to obtain a spliced ​​image;

[0047] Inputting the spliced ​​image into a pre-trained segregation analysis model to obtain a segregation vector of the concrete pouring process;

[0048] The similarity is calculated based on the segregation vector and the preset segregation standard vector. If the similarity is greater than the preset similarity threshold, it means that the segregation of the current concrete is large; if the similarity is less than the preset similarity threshold, it means that the segregation of the current concrete is small; and the similarity is used as the segregation result.

[0049] The step of obtaining the preset standard separation vector specifically includes:

[0050] Step 1: Prepare multiple groups of concrete specimens with different degrees of segregation:

[0051] a) Design concrete mix proportions with different water-cement ratios and aggregate gradations.

[0052] b) For each mix ratio, different vibration times and vibration frequencies were used to prepare the specimens to obtain different degrees of segregation.

[0053] c) Prepare at least 3 parallel specimens for each condition to ensure the reliability of the data.

[0054] Step 2: Use this device to collect images of the specimen:

[0055] a) Cut the specimen to expose the fresh section.

[0056] b) Use the device to collect images at different positions and depths of the specimen.

[0057] c) Collect at least 9 images at different positions for each specimen to ensure the representativeness of the data.

[0058] Step 3: Mark the segregation degree of each specimen by experimental method:

[0059] a) Cement distribution uniformity: X-ray fluorescence analysis (XRF) was used to measure the cement content at different locations and the coefficient of variation was calculated.

[0060] b) Uniformity of coarse aggregate distribution: Use image analysis software to calculate the volume fraction of coarse aggregate in different areas and calculate the coefficient of variation.

[0061] c) Uniformity of fine aggregate distribution: Use a laser particle size analyzer to measure the fine aggregate particle size distribution at different locations and calculate the similarity of the distribution curves.

[0062] d) Porosity: The porosity at different locations of the specimen was measured using the mercury intrusion method.

[0063] e) Interface transition zone thickness: The interface transition zone thickness at multiple locations was observed and measured using a scanning electron microscope (SEM).

[0064] Step 4: Data processing and standardization:

[0065] a) Take the average value of each index of each specimen.

[0066] b) Each indicator is normalized to the interval [0, 1], where 0 indicates no segregation at all and 1 indicates severe segregation.

[0067] Step 5: Construct a standard separation vector dataset:

[0068] a) The five normalized indices of each specimen are combined into a vector.

[0069] b) Summarize the vectors of all specimens to form a standard separation vector data set.

[0070] Step 6: Data analysis and verification:

[0071] a) Use principal component analysis (PCA) to analyze the data set and identify the main influencing factors.

[0072] b) Perform cross-validation to evaluate the reliability and representativeness of the dataset.

[0073] Step 7: Establish a classification standard for the degree of segregation:

[0074] a) Based on the data distribution, the degree of segregation is divided into multiple levels (such as mild, moderate, severe, etc.).

[0075] b) Determine a representative standard segregation vector for each level.

[0076] Step 8: Expert review and revision:

[0077] a) Invite experts in the field of concrete materials to review the grading standards.

[0078] b) Make necessary revisions and improvements to the standards based on expert opinions.

[0079] Step 9: Finally determine the standard separation vector:

[0080] a) Comprehensively consider the data analysis results and expert opinions, and finally determine the standard segregation vector for each segregation level.

[0081] b) Use these standard separation vectors as reference for subsequent separation analysis.

[0082] Furthermore, the segregation vector is specifically [cement distribution uniformity, coarse aggregate distribution uniformity, fine aggregate distribution uniformity, porosity, interface transition zone thickness].

[0083] Furthermore, the segregation analysis model adopts a multi-branch deep learning structure, specifically including a cement branch network, a coarse aggregate branch network, a fine aggregate branch network, a pore branch network, an interface branch network and a fusion network.

[0084] Furthermore, the cement branch network is used to extract cement-related features, the input is a preprocessed grayscale image, the output is a cement distribution feature map, and the structure is a U-Net convolutional neural network;

[0085] The coarse aggregate branch network is used to extract coarse aggregate related features, the input is a preprocessed color image, the output is a coarse aggregate distribution feature map, and the structure is a ResNet convolutional neural network;

[0086] The fine aggregate branch network is used to extract fine aggregate related features, the input is a preprocessed color image, the output is a fine aggregate distribution feature map, and the structure is a DenseNet convolutional neural network;

[0087] The pore branch network is used to extract pore-related features, the input is a preprocessed binary image, the output is a pore distribution feature map, and the structure is a FCN fully convolutional network;

[0088] The interface branch network is used to extract the interface features between aggregate and cement paste. The input is the preprocessed edge enhanced image, the output is the interface distribution feature map, and the structure is the SegNet segmentation network.

[0089] The fusion network is used to integrate the output features of each branch network, the input is the output feature map of each branch network, the output is the final separation feature vector, and the structure is a multi-layer perceptron.

[0090] The training step of the separation analysis model includes constructing a training data set and a training model; wherein constructing the training data set specifically includes:

[0091] a) Sample collection:

[0092] Prepare at least 100 sets of concrete specimens with different mix proportions and different degrees of segregation;

[0093] This device is used to collect images of each specimen at multiple angles and depths;

[0094] At least 20 images were collected for each specimen to ensure data diversity;

[0095] b) Data annotation:

[0096] Manually annotate the collected images, including: cement distribution area; coarse aggregate position and size; fine aggregate distribution area; pore position and size; aggregate and cement paste interface; and use professional image annotation software (such as LabelMe) for annotation;

[0097] c) Data Augmentation:

[0098] Rotate, scale, flip, and other operations on the original image to expand the data set;

[0099] Adjust image brightness and contrast to simulate different lighting conditions;

[0100] Add random noise to improve model robustness;

[0101] d) Data preprocessing:

[0102] Image normalization: scaling pixel values ​​to the [0, 1] interval;

[0103] Image segmentation: split large-size images into patches of fixed size (e.g. 256x256 pixels);

[0104] Data format conversion: converting images and annotation data into the format required by the model (such as TFRecord or HDF5);

[0105] e) Dataset division: The dataset is randomly divided into training set, validation set and test set in a ratio of 7:2:1; ensuring that each set contains samples with different degrees of segregation;

[0106] f) Normalized separation vector:

[0107] For each sample, a standardized separation vector is calculated based on the annotation information;

[0108] Segregation vectors include: cement distribution uniformity, coarse aggregate distribution uniformity, fine aggregate distribution uniformity, porosity, and interface transition zone thickness;

[0109] g) Data quality control: cross-validation is performed to eliminate abnormal samples; concrete experts review some samples to ensure the quality of annotation;

[0110] Among them, the training model is specifically:

[0111] a) Model architecture design: Implement each branch network based on the multi-branch deep learning structure described previously; use deep learning frameworks such as PyTorch or TensorFlow to build the model;

[0112] b) Loss function design:

[0113] Design an appropriate loss function for each branch network:

[0114] Cement branch: use cross entropy loss;

[0115] Coarse aggregate branch: Use Focal Loss;

[0116] Fine aggregate branch: Use Dice Loss;

[0117] Porous branch: using weighted binary cross entropy;

[0118] Interface branch: using edge-aware loss;

[0119] Fusion network: use mean square error loss;

[0120] The total loss function is the weighted sum of the losses of each branch;

[0121] c) Optimizer selection: Use the Adam optimizer with an initial learning rate of 0.001; implement a learning rate decay strategy, such as cosine annealing;

[0122] d) Model initialization: Use He initialization for the convolutional layer; use a normal distribution with a mean of 1 and a variance of 0.02 for the batch normalization layer;

[0123] e) Training process: Set the batch size to 32 and the number of training rounds to 100; implement the early stopping strategy and stop training when the validation set loss does not decrease for 5 consecutive rounds; use mixed precision training to increase the training speed; implement gradient clipping to prevent gradient explosion;

[0124] f) Model fine-tuning: Freeze the backbone network and only train the last few layers; gradually unfreeze and fine-tune;

[0125] g) Regularization technique: L2 regularization is used, and the weight decay coefficient is set to 0.0001; Dropout is used in the fully connected layer with a ratio of 0.5;

[0126] h) Model evaluation: Calculate the mean absolute error (MAE) and root mean square error (RMSE) on the validation set; draw a scatter plot of the predicted value and the true value, and calculate the correlation coefficient;

[0127] i) Model interpretation: Use techniques such as Grad-CAM to visualize the areas that the model focuses on; analyze the contribution of different features to the final prediction;

[0128] j) Model integration: train multiple models, use different initialization and data augmentation strategies; use soft voting or averaging to integrate the prediction results of multiple models;

[0129] k) Select the optimal model as the training result.

[0130] Wherein, the tube body is made of stainless steel; and a conical structure is also arranged at the bottom end of the tube body.

[0131] It is convenient to insert or remove the pipe body into the poured concrete.

[0132] Furthermore, the preset similarity threshold is 0.85.

[0133] The choice of this threshold is based on the following considerations:

[0134] 1. Engineering practice: In concrete engineering, a certain degree of error and variation is usually allowed. A similarity of 85% means that the prediction result is highly consistent with the standard separation vector, while also leaving a certain tolerance space.

[0135] 2. Model performance: Considering the noise and interference that may exist in the actual environment, it is unrealistic to require the model to achieve 100% similarity. The 85% threshold can ensure high accuracy without being too harsh.

[0136] 3. Balance between sensitivity and specificity: A threshold of 85% can achieve a good balance between reducing false positives (misjudged as severe segregation) and false negatives (missed mild segregation).

[0137] 4. Adjustability: Setting the threshold to 0.85 also allows fine-tuning in actual applications according to specific circumstances. For example, for particularly important projects, the threshold can be slightly increased to 0.90; while for general projects, it can be slightly reduced to 0.80.

[0138] 5. Grading judgment: Based on this threshold, you can establish a multi-level judgment standard, for example:

[0139] - Similarity ≥ 0.85: Concrete has a small degree of segregation and is of good quality

[0140] - 0.70 ≤ Similarity < 0.85: Concrete may have slight segregation, which requires attention

[0141] - Similarity < 0.70: The concrete may be severely segregated and measures need to be taken

[0142] Therefore, we can modify the judgment condition as follows:

[0143] If the similarity is ≥ 0.85 (the preset similarity threshold), it means that the current concrete has a small degree of segregation;

[0144] If the similarity is < 0.85 (the preset similarity threshold), it means that the current concrete may have segregation problems and requires further inspection and treatment.

[0145] It should be noted that this threshold should be fine-tuned and optimized in actual applications through a large amount of experimental data and field verification. During the initial deployment of the model, feedback can be collected and the threshold can be continuously adjusted to achieve the best judgment effect.

[0146] Compared with the prior art, the beneficial effects of the segregation detection device for deep foundation pit concrete pouring process provided by the present invention are:

[0147] 1. Real-time monitoring capability: This device uses optical imaging and fiber optic transmission technology to continuously collect image data during the concrete pouring process. Combined with high-speed image processing and deep learning algorithms, it realizes real-time analysis and evaluation of segregation. This feature enables construction personnel to discover and deal with segregation problems in a timely manner, greatly improving the efficiency and effectiveness of quality control.

[0148] 2. High-precision detection: Through optimized optical system design and multi-parameter comprehensive analysis, this device can capture subtle changes in the microstructure of concrete. Compared with traditional visual inspection and simple instrument detection, the segregation evaluation results provided by this device are more accurate and objective, significantly reducing human errors.

[0149] 3. Comprehensive monitoring range: This device adopts a design with multiple monitoring units evenly distributed, which can monitor different depths and positions of the concrete structure at the same time. This all-round monitoring capability overcomes the problem of limited monitoring range in the existing technology and provides the possibility for comprehensive evaluation of the segregation of the entire deep foundation pit concrete structure.

[0150] 4. Intelligent data analysis: This device integrates a segregation analysis model based on deep learning, which can automatically identify and quantify multiple key parameters such as cement distribution uniformity, aggregate distribution uniformity, porosity and interface transition zone thickness. This intelligent data analysis method not only provides more comprehensive and accurate segregation evaluation results, but also greatly reduces the workload of manual analysis.

[0151] In summary, the device of the present invention solves the technical problems that the current segregation detection in the deep foundation pit concrete pouring process relies on the experience of engineers and is difficult to evaluate in real time. BRIEF DESCRIPTION OF THE DRAWINGS

[0152] Figure 1 It is a structural schematic diagram of a segregation detection device during the pouring process of deep foundation pit concrete;

[0153] Figure 2 It is a structural diagram of the monitoring unit;

[0154] In the accompanying drawings, the components represented by the reference numerals are listed as follows:

[0155] 10. Tube body; 11. Conical structure; 20. Monitoring unit; 21. Tempered glass sheet; 22. Concave lens; 23. Isolation glass sheet; 24. Focusing sheet; 25. Optical fiber bundle; 30. Single chip microcomputer. DETAILED DESCRIPTION

[0156] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0157] like Figure 1~2 FIG. 1 is a schematic diagram of the structure of a segregation detection device for a deep foundation pit concrete pouring process provided by the present invention. The device is used to detect the segregation of concrete during the concrete pouring process. The specific implementation of the device is described in detail below:

[0158] 1. Device structure

[0159] The device comprises a tube body 10, a monitoring unit 20 and a single chip microcomputer. A plurality of circular through holes are provided on the side wall of the tube body. The plurality of circular through holes are evenly arranged in 2 or 3 rows on the side wall. A monitoring unit is embedded in each circular through hole. The monitoring unit comprises a tempered glass sheet 21, a concave lens 22, an isolation glass sheet 23, a focusing sheet 24 and an optical fiber bundle 25. A circular hole is provided at the bottom of the focusing sheet. One end of the optical fiber bundle is embedded in the circular hole as a collection end. The single chip microcomputer is provided with a controller and image collectors of at least the number of monitoring units; the other end of the optical fiber bundle is connected to the image collector as an output end. The image collector is used to receive the optical signal transmitted by the optical fiber bundle and convert it into an electrical signal. After analog-to-digital conversion, the regional image of the monitoring area corresponding to the monitoring unit is obtained and output to the controller; the cross section of the condenser is parabolic, and the focus of the concave lens located inside the tube body and the focus of the condenser are both located at the geometric center of the surface of the collection end; the controller is provided with a concrete pouring process separation and analysis module, which is used to calculate the separation results of the concrete pouring process based on the regional images collected by multiple image collectors and the pre-trained separation analysis model, and transmit it to the host computer through the wireless transmission module on the single-chip microcomputer. The tube body is made of stainless steel; the bottom end of the tube body is also provided with a conical structure 11.

[0160] 2. Parameter Optimization

[0161] In order to obtain the best detection effect, the present invention adopts a set of parameter optimization methods. These parameters include the spacing of circular through holes, the radius of tempered glass, the shape parameters of concave lenses, the shape parameters of condensers, and the parameters of optical fiber bundles. The optimization process uses a set of optimization equations, including optical imaging equations, spatial sampling equations, signal-to-noise ratio equations, resolution equations, and depth detection equations.

[0162] 2.1 Optimization equations

[0163] a) Optical imaging equation:

[0164] ;

[0165] This equation is used to optimize the parameters of a concave lens, where f is the focal length of the concave lens, u is the object distance, v is the image distance, n is the refractive index of the lens material, d is the center thickness of the lens, and R1 and R2 are the radii of curvature of the two surfaces of the lens.

[0166] The shape parameters of the concentrator are optimized using the parabolic equation: , where p is the focal parameter.

[0167] b) Spatial sampling equation:

[0168] ;

[0169] This equation is used to optimize the spacing of the circular through holes and the radius of the tempered glass sheet, where Δs is the minimum resolvable distance, λ is the wavelength of light, f is the focal length of the system, and NA is the numerical aperture.

[0170] c) Signal-to-noise ratio equation:

[0171] ;

[0172] This equation is used to optimize the parameters of the fiber bundle, where Psignal is the signal power and Pnoise is the noise power.

[0173] The transmission loss of the optical fiber bundle is given by the formula Calculate, where Pout is the output power, Pin is the input power, α is the attenuation coefficient, and L is the fiber length.

[0174] d) Resolution equation:

[0175] ;

[0176] This equation is used to comprehensively optimize the parameters of each optical component, where R is the system resolution and D is the effective aperture diameter.

[0177] e) Depth detection equation:

[0178] ;

[0179] This equation is used to optimize the outer diameter of the tube and the size of the through hole, where dmax is the maximum detection depth, c is the speed of light, τ is the pulse width, and εr is the relative dielectric constant of concrete.

[0180] The relationship between the outer diameter of the tube and the size of the through hole is expressed by the formula Determine, where Ahole is the through-hole area, k is the filling factor (0 < k < 1), and r is the outer radius of the tube.

[0181] The acquisition and calculation steps of these parameters are as follows:

[0182] 1. Optical system parameters (such as , , , , ):

[0183] These parameters can be optimized by optical design software (such as Zemax) or obtained by experimental measurement. The experimental steps include:

[0184] a) Use a lens meter to measure the focal length of a concave lens

[0185] b) Use a refractometer to measure the refractive index of the lens material

[0186] c) Use a micrometer to measure the center thickness of the lens

[0187] d) Use a spherometer to measure the radius of curvature of the lens surface

[0188] 2. Wavelength :

[0189] The central wavelength of a light source can be measured using a spectrometer. Measurement steps:

[0190] a) Calibrate the spectrometer

[0191] b) Collecting light source spectrum

[0192] c) Analyze spectral data and determine the central wavelength

[0193] 3. Numerical aperture :

[0194] It can be calculated by measuring the beam divergence angle. Calculation formula:

[0195]

[0196] in is the refractive index of the medium, is the half divergence angle of the beam

[0197] 4. Signal power and noise power :

[0198] Use an optical power meter to measure. Measurement steps:

[0199] a) Measure noise power when there is no signal input

[0200] b) Input a signal of known strength and measure the total power

[0201] c) Subtract the noise power from the total power to get the signal power

[0202] 5. Fiber attenuation coefficient :

[0203] It can be measured by the cut-back method. Measurement steps:

[0204] a) The measured length is Fiber output power

[0205] b) Cut the optical fiber to length , measure output power

[0206] c) Calculate the attenuation coefficient:

[0207] 6. Effective aperture diameter :

[0208] It can be measured directly through an aperture or light barrier, or calculated from the diffraction pattern.

[0209] 7. Pulse width :

[0210] Use a high-speed oscilloscope to measure. Measurement steps:

[0211] a) Convert optical signals into electrical signals

[0212] b) Use an oscilloscope to measure the half-maximum full width of the electrical signal

[0213] 8. Relative dielectric constant of concrete :

[0214] It can be measured by time domain reflectometry (TDR). Measurement steps:

[0215] a) Insert the TDR probe into the concrete sample

[0216] b) Send out electromagnetic pulses and measure the reflected signal

[0217] c) Calculate the dielectric constant based on the reflection time

[0218] 9. Fill factor :

[0219] This is a design parameter that needs to be determined based on actual needs and engineering experience, and is usually between 0.3 and 0.7.

[0220] 2.2 Optimization process

[0221] The optimization process uses the particle swarm algorithm (PSO), and the specific steps are as follows:

[0222] a) Establish the objective function:

[0223] The objective function F is obtained by weighted summing of the equations:

[0224]

[0225] in arrive is the weight coefficient of each equation, arrive is the error function of each equation.

[0226] b) Set the initial values ​​of parameters and constraints:

[0227] The initial values ​​of each parameter are set based on existing equipment specifications, relevant literature, and preliminary experimental measurements. At the same time, the constraints of each parameter are set based on actual equipment limitations, material properties, and application requirements.

[0228] c) Particle swarm algorithm iterative optimization:

[0229] - Initialize the particle swarm and create N particles for each parameter.

[0230] - Evaluate the fitness of each particle, that is, calculate the objective function F(Xi).

[0231] - Update individual optimal positions and global optimal positions.

[0232] - Update particle speed and position.

[0233] - Check the termination condition and return to the evaluation step if it is not met.

[0234] d) Output optimization results:

[0235] When the number of iterations reaches the preset value or the objective function converges, the optimal parameter combination is output.

[0236] Specifically, the specific steps of optimizing and limiting by optimizing the set of equations include:

[0237] S11. Establish the objective function and perform weighted summation of each equation:

[0238] Objective Function Defined as:

[0239]

[0240] in:

[0241] is the error function of the optical imaging equation;

[0242] is the error function of the spatial sampling equation;

[0243] is the error function of the signal-to-noise ratio equation;

[0244] is the error function of the resolution equation;

[0245] is the error function of the depth detection equation;

[0246] is the weight coefficient of each equation, satisfying

[0247] The error functions are defined as follows:

[0248] 1. Error function of optical imaging equation:

[0249] ;

[0250] 2. Spatial sampling equation error function:

[0251] ;

[0252] 3. Signal-to-noise ratio equation error function:

[0253] ;

[0254] in is the target signal-to-noise ratio;

[0255] 4. Resolution equation error function:

[0256] ;

[0257] in is the target resolution;

[0258] 5. Depth detection equation error function:

[0259] ;

[0260] in is the maximum detection depth of the target;

[0261] S12. Set the initial values ​​and constraints of each parameter:

[0262] Initial value setting: Concave lens focal length , object distance , Image distance , refractive index of lens material , lens center thickness , radius of curvature of lens surface , , wavelength of light , numerical aperture , signal power , noise power , effective aperture diameter , pulse width , relative dielectric constant of concrete ;

[0263] These initial values ​​can be obtained by: selecting according to existing equipment specifications; referring to typical values ​​in relevant literature; performing preliminary experimental measurements;

[0264] Constraint setting:

[0265] ;

[0266] ;

[0267] ;

[0268] ;

[0269] ;

[0270] ;

[0271] ;

[0272] ;

[0273] ;

[0274] ;

[0275] ;

[0276] ;

[0277] ;

[0278]

[0279] These constraints are determined based on practical device limitations, material properties, and application requirements.

[0280] S13, using particle swarm algorithm for iterative optimization:

[0281] Particle swarm optimization (PSO) is a group intelligence optimization algorithm, and its main steps are as follows:

[0282] 1. Initialize the particle swarm:

[0283] Create for each parameter particles, each particle represents a set of possible parameter values.

[0284] particle The position vector is:

[0285] ;

[0286] Initial velocity vector:

[0287] ;

[0288] 2. Evaluate the fitness of each particle:

[0289] Calculate the objective function

[0290] 3. Update individual optimal position and global optimal position:

[0291] For each particle ,if , then update the individual optimal position ;

[0292] if , then update the global optimal position ;

[0293] 4. Update particle speed and position:

[0294] ;

[0295] ;

[0296] in: is the inertia weight; is the acceleration constant; is a random number between [0, 1];

[0297] 5. Check termination conditions:

[0298] If the maximum number of iterations is reached or the objective function value converges, the process terminates; otherwise, it returns to step 2.

[0299] S14. When the number of iterations reaches the preset value or the objective function converges, the optimization result is output:

[0300] Convergence judgment conditions:

[0301]

[0302] in:

[0303] For the The optimal objective function value of the iteration; is the preset convergence threshold;

[0304] Output the optimal parameter combination:

[0305]

[0306] ;

[0307] These optimized parameters will be used to determine the final design of the device, including the spacing of the circular through holes, the radius of the tempered glass sheet, the shape parameters of the concave lens and the focusing sheet, and the parameters of the optical fiber bundle.

[0308] Through this optimization method, the optimal value of each parameter can be obtained, thus ensuring that the device has the best detection performance.

[0309] 3. Segregation analysis model

[0310] The present invention adopts a multi-branch deep learning structure as a segregation analysis model, including a cement branch network, a coarse aggregate branch network, a fine aggregate branch network, a pore branch network, an interface branch network and a fusion network.

[0311] 3.1 Model Structure

[0312] a) Cement branch network: It adopts U-Net convolutional neural network structure, the input is the preprocessed grayscale image, and the output is the cement distribution feature map.

[0313] b) Coarse aggregate branch network: It adopts the ResNet convolutional neural network structure, the input is the preprocessed color image, and the output is the coarse aggregate distribution feature map.

[0314] c) Fine aggregate branch network: It adopts the DenseNet convolutional neural network structure, the input is the preprocessed color image, and the output is the fine aggregate distribution feature map.

[0315] d) Pore branch network: It adopts FCN fully convolutional network structure, the input is the preprocessed binary image, and the output is the pore distribution feature map.

[0316] e) Interface branch network: It adopts the SegNet segmentation network structure, the input is the preprocessed edge enhanced image, and the output is the interface distribution feature map.

[0317] f) Fusion network: It adopts a multi-layer perceptron structure, the input is the output feature map of each branch network, and the output is the final separated feature vector.

[0318] 3.2 Model Training

[0319] The model training process includes the following steps:

[0320] a) Constructing training data set:

[0321] - Prepare at least 100 sets of concrete specimens with different mix proportions and different degrees of segregation.

[0322] - Use this device to collect images of each specimen at multiple angles and depths, with at least 20 images for each specimen.

[0323] - Manually annotate the collected images, including cement distribution area, coarse aggregate location and size, fine aggregate distribution area, pore location and size, aggregate and cement paste interface, etc.

[0324] - Perform data enhancement, including rotation, scaling, flipping, brightness adjustment, adding random noise, etc.

[0325] - Data preprocessing, including image normalization, segmentation, and format conversion.

[0326] - The dataset is randomly divided into training set, validation set and test set in the ratio of 7:2:1.

[0327] b) Model architecture design:

[0328] - Build models using deep learning frameworks such as PyTorch or TensorFlow.

[0329] - Design an appropriate loss function for each branch network, such as cross entropy loss, focal loss, Dice loss, etc.

[0330] - Select the Adam optimizer, set the initial learning rate to 0.001, and implement the learning rate decay strategy.

[0331] c) Training process:

[0332] - Set the batch size to 32 and the number of training rounds to 100.

[0333] - Implement an early stopping strategy and stop training when the validation set loss does not decrease for 5 consecutive rounds.

[0334] - Use mixed precision training to increase training speed.

[0335] - Implement gradient clipping to prevent gradient explosion.

[0336] d) Model evaluation and optimization:

[0337] - Calculate the mean absolute error (MAE) and root mean square error (RMSE) on the validation set.

[0338] - Draw a scatter plot of predicted values ​​and true values ​​and calculate the correlation coefficient.

[0339] - Visualize the areas that the model is paying attention to using techniques such as Grad-CAM.

[0340] - Analyze the contribution of different features to the final prediction.

[0341] e) Model deployment preparation:

[0342] - Use knowledge distillation or pruning techniques to reduce model size.

[0343] - Convert floating point models to integer models to improve inference speed.

[0344] - Export the trained model to a format suitable for deployment (such as ONNX).

[0345] 4. Separation and detection process

[0346] 4.1 Image acquisition

[0347] When the device is inserted into freshly poured concrete, the concrete contacts the monitoring unit through the circular through-hole. After passing through the tempered glass sheet, concave lens and condenser sheet, the light is collected by the optical fiber bundle and transmitted to the image collector.

[0348] 4.2 Image Processing

[0349] The image collector converts the optical signal into an electrical signal, which is converted into a digital image after analog-to-digital conversion. These images are then sent to the controller for processing.

[0350] 4.3 Segregation analysis

[0351] The concrete pouring process segregation analysis module in the controller performs the following steps:

[0352] a) Obtain the regional image captured by each image collector.

[0353] b) Preprocessing the regional image, including denoising, enhancement and correction, to obtain a preprocessed image.

[0354] c) Using image registration and fusion methods to stitch all preprocessed images to obtain a stitched image.

[0355] d) Input the spliced ​​image into the pre-trained segregation analysis model to obtain the segregation vector of the concrete pouring process. The segregation vector includes five indicators: [cement distribution uniformity, coarse aggregate distribution uniformity, fine aggregate distribution uniformity, porosity, and interface transition zone thickness].

[0356] e) Calculate the similarity between the segregation vector and the preset segregation standard vector. If the similarity is greater than or equal to the preset similarity threshold (0.85), it is judged that the segregation degree of the current concrete is small; if the similarity is less than the preset similarity threshold, it is judged that the current concrete may have segregation problems and further inspection and processing are required.

[0357] f) The calculated similarity is used as the separation result and sent to the host computer through the wireless transmission module.

[0358] The preset similarity threshold can be set to 0.85 or 85%. The selection of this threshold is based on the following considerations:

[0359] 1. Engineering practice: In concrete engineering, a certain degree of error and variation is usually allowed. A similarity of 85% means that the prediction result is highly consistent with the standard separation vector, while also leaving a certain tolerance space.

[0360] 2. Model performance: Considering the noise and interference that may exist in the actual environment, it is unrealistic to require the model to achieve 100% similarity. The 85% threshold can ensure high accuracy without being too harsh.

[0361] 3. Balance between sensitivity and specificity: A threshold of 85% can achieve a good balance between reducing false positives (misjudged as severe segregation) and false negatives (missed mild segregation).

[0362] 4. Adjustability: Setting the threshold to 0.85 also allows fine-tuning in actual applications according to specific circumstances. For example, for particularly important projects, the threshold can be slightly increased to 0.90; while for general projects, it can be slightly reduced to 0.80.

[0363] 5. Grading judgment: Based on this threshold, you can establish a multi-level judgment standard, for example:

[0364] - Similarity ≥ 0.85: Concrete has a small degree of segregation and is of good quality

[0365] - 0.70 ≤ Similarity < 0.85: Concrete may have slight segregation, which requires attention

[0366] - Similarity < 0.70: The concrete may be severely segregated and measures need to be taken

[0367] Therefore, we can modify the judgment condition as follows:

[0368] If the similarity is ≥ 0.85 (the preset similarity threshold), it means that the current concrete has a small degree of segregation;

[0369] If the similarity is < 0.85 (the preset similarity threshold), it means that the current concrete may have segregation problems and requires further inspection and treatment.

[0370] It should be noted that this threshold should be fine-tuned and optimized in actual applications through a large amount of experimental data and field verification. During the initial deployment of the model, feedback can be collected and the threshold can be continuously adjusted to achieve the best judgment effect.

[0371] The steps for obtaining the standard separation vector are described in detail below:

[0372] 5.1 Specimen preparation

[0373] a) Design concrete mix proportions with different water-cement ratios and aggregate gradations.

[0374] b) For each mix ratio, different vibration times and vibration frequencies were used to prepare the specimens to obtain different degrees of segregation.

[0375] c) Prepare at least 3 parallel specimens for each condition to ensure the reliability of the data.

[0376] 5.2 Image acquisition

[0377] a) Cut the specimen to expose the fresh section.

[0378] b) Use the device to collect images at different positions and depths of the specimen.

[0379] c) Collect at least 9 images at different positions for each specimen to ensure the representativeness of the data.

[0380] 5.3 Segregation degree marking

[0381] The degree of segregation of each specimen is marked by experiment:

[0382] a) Cement distribution uniformity: X-ray fluorescence analysis (XRF) was used to measure the cement content at different locations and the coefficient of variation was calculated.

[0383] b) Uniformity of coarse aggregate distribution: Use image analysis software to calculate the volume fraction of coarse aggregate in different areas and calculate the coefficient of variation.

[0384] c) Uniformity of fine aggregate distribution: Use a laser particle size analyzer to measure the fine aggregate particle size distribution at different locations and calculate the similarity of the distribution curves.

[0385] d) Porosity: The porosity at different locations of the specimen was measured using the mercury intrusion method.

[0386] e) Interface transition zone thickness: The interface transition zone thickness at multiple locations was observed and measured using a scanning electron microscope (SEM).

[0387] 5.4 Data processing and standardization

[0388] a) Take the average value of each index of each specimen.

[0389] b) Each indicator is normalized to the interval [0, 1], where 0 indicates no segregation at all and 1 indicates severe segregation.

[0390] 5.5 Constructing a Standard Separation Vector Dataset

[0391] a) The five normalized indices of each specimen are combined into a vector.

[0392] b) Summarize the vectors of all specimens to form a standard separation vector data set.

[0393] 5.6 Data Analysis and Validation

[0394] a) Use principal component analysis (PCA) to analyze the data set and identify the main influencing factors.

[0395] b) Perform cross-validation to evaluate the reliability and representativeness of the dataset.

[0396] 5.7 Establishing a classification standard for the degree of segregation

[0397] a) Based on the data distribution, the degree of segregation is divided into multiple levels (such as mild, moderate, severe, etc.).

[0398] b) Determine a representative standard segregation vector for each level.

[0399] 5.8 Expert Review and Revision

[0400] a) Invite experts in the field of concrete materials to review the grading standards.

[0401] b) Make necessary revisions and improvements to the standards based on expert opinions.

[0402] 5.9 Final determination of the standard separation vector

[0403] a) Comprehensively consider the data analysis results and expert opinions, and finally determine the standard segregation vector for each segregation level.

[0404] b) Use these standard separation vectors as reference for subsequent separation analysis.

[0405] 6. Practical application process

[0406] 6.1 Device Installation

[0407] The device is inserted into the intended location before concrete pouring. Multiple devices can be installed at different depths and locations to obtain more comprehensive monitoring data.

[0408] 6.2 Real-time monitoring

[0409] During the concrete pouring process, the device continuously collects images and performs segregation analysis. The analysis results are transmitted to the host computer in real time for reference by on-site engineers.

[0410] 6.3 Early warning mechanism

[0411] If the segregation level is detected to exceed the preset threshold, the system will immediately issue an alarm to remind engineers to take corresponding measures, such as adjusting the vibration frequency, changing the pouring speed, etc.

[0412] 6.4 Data Recording and Analysis

[0413] All monitoring data will be recorded for subsequent quality analysis and process improvement. These data can help engineers optimize concrete mix and pouring process.

[0414] Specifically, the principle of the present invention is:

[0415] 1. Optical imaging principle:

[0416] This device uses a carefully designed optical system, including a concave lens and a parabolic condenser, which can efficiently capture the optical characteristics of the internal structure of concrete. The concave lens is used to converge the light scattered from the concrete, while the parabolic condenser further improves the utilization of light energy. The focal points of these two optical components are located at the geometric center of the collection end of the fiber bundle, ensuring the maximum collection of effective information. The design of the parabolic condenser can accurately focus parallel light to one point, maximizing the efficiency of light energy collection.

[0417] 2. Spatial sampling principle:

[0418] In order to obtain sufficient spatial resolution, the design of this device takes into account the diffraction limit theory. According to the Rayleigh criterion, the spatial resolution of the system can be expressed as: ; where λ is the wavelength of light, f is the focal length of the system, and NA is the numerical aperture. By optimizing these parameters, high-resolution imaging of the concrete microstructure can be achieved.

[0419] 3. Signal-to-noise ratio optimization principle:

[0420] In order to improve the image quality, this device adopts the signal-to-noise ratio optimization design. The signal-to-noise ratio can be expressed as: ; Where Psignal is the signal power and Pnoise is the noise power. By optimizing the optical fiber transmission system, such as selecting the appropriate optical fiber type and length, signal attenuation and noise interference can be minimized. Optical fiber transmission loss follows the formula: , where Pout is the output power, Pin is the input power, α is the attenuation coefficient, and L is the fiber length.

[0421] 4. Depth detection principle:

[0422] This device can detect different depths inside concrete, which is based on the principle of electromagnetic wave propagation in the medium. The maximum detection depth can be expressed as:

[0423] ; where c is the speed of light, τ is the pulse width, is the relative dielectric constant of concrete. By adjusting these parameters, the detection depth can be optimized while ensuring sufficient spatial resolution.

[0424] 5. Multi-parameter comprehensive optimization principle:

[0425] This device uses a particle swarm algorithm (PSO) to comprehensively optimize multiple parameters. The PSO algorithm simulates the intelligent behavior of the group, continuously updates the position and velocity of the particles through iteration, and finally converges to the global optimal solution. This method can find the best combination of design parameters in a complex multi-dimensional parameter space, ensuring that the device can perform at its best in various engineering environments.

[0426] 6. Principles of Deep Learning:

[0427] This device uses a multi-branch deep learning structure for segregation analysis, which can process multiple features at the same time and improve the model's expressiveness and generalization capabilities. Each branch network (such as U-Net, ResNet, DenseNet, etc.) focuses on extracting specific concrete features (such as cement distribution, aggregate distribution, etc.), while the fusion network combines these features to obtain the final segregation evaluation results.

[0428] The deep learning model learns the complex nonlinear relationship between the microstructural characteristics of concrete and the degree of segregation through training with a large amount of labeled data. This data-driven approach can automatically extract key features, avoiding the limitations of manually designed features, and thus achieving more accurate segregation assessment.

[0429] In order to better understand and implement the present invention, an embodiment of a specific application scenario of the present invention is provided below: This embodiment describes the application of a segregation detection device in the process of deep foundation pit concrete pouring in the foundation construction of a super high-rise building. The project is located in a coastal city, with a total building height of 150 meters. The foundation adopts a raft pile foundation with a raft thickness of 4 meters and a foundation pit depth of 15 meters. Due to the large scale of the project and complex geological conditions, the quality control of concrete pouring is particularly important.

[0430] 1. Device configuration

[0431] According to the size and depth of the foundation pit, the project department decided to use 20 sets of deep foundation pit concrete pouring process segregation detection devices. Each set of devices is configured as follows:

[0432] Tube body: stainless steel, outer diameter 4cm, length 40cm;

[0433] Monitoring unit: 10 monitoring units are evenly distributed on each pipe body;

[0434] Single chip microcomputer: adopts high-performance ARM Cortex-M7 processor, main frequency 480MHz;

[0435] Image collectors: 10, each corresponding to a monitoring unit;

[0436] Wireless transmission module: using 5G communication technology to ensure real-time data transmission;

[0437] 2. Parameter Optimization

[0438] Before installation, the device parameters were optimized according to the specific conditions of this project. The optimization process used a particle swarm algorithm and iterated 500 times, and finally obtained the following key parameters:

[0439] Concave lens focal length: 5.2mm;

[0440] Focus parameter of condenser: 0.8mm;

[0441] Circular through hole spacing: 35mm;

[0442] Tempered glass radius: 7.5mm;

[0443] Fiber bundle diameter: 200 μm;

[0444] Number of optical fibers: 19 per bundle;

[0445] 3. Installation and layout

[0446] In the raft casting area, 20 sets of detection devices are evenly arranged in a 5m×5m grid. Each set of devices is vertically inserted into the reserved hole, with the top flush with the top surface of the raft. After installation, all devices are connected to the on-site control center and connected to the remote monitoring center through the 5G network.

[0447] 4. Obtaining standard separation vector

[0448] In order to obtain the standard separation vector applicable to this project, the project department conducted preliminary tests according to the following steps:

[0449] a) Preparation of test specimens: According to the concrete mix ratio of the project design (cement dosage 450kg / m³, water-binder ratio 0.36, maximum particle size of coarse aggregate 40mm), 15 groups of test specimens with different segregation degrees were prepared. Each group has 3 parallel test specimens, a total of 45.

[0450] b) Image acquisition: The device was used to acquire images at 9 different positions of each specimen, and a total of 405 images were acquired.

[0451] c) Experimental measurements:

[0452] Cement distribution uniformity: The cement content at 15 points was measured using an X-ray fluorescence analyzer, and the coefficient of variation was calculated.

[0453] Uniformity of coarse aggregate distribution: The volume fraction of coarse aggregate in nine areas was calculated using image analysis software, and the coefficient of variation was calculated.

[0454] Uniformity of fine aggregate distribution: The particle size distribution of fine aggregate at 9 locations was measured using a laser particle size analyzer, and the similarity of the distribution curves was calculated.

[0455] Porosity: The porosity was measured at 9 locations using mercury intrusion porosimetry.

[0456] Thickness of interface transition zone: The interface transition zone was observed at 15 locations using a scanning electron microscope and the thickness was measured.

[0457] d) Data processing: Normalize all measurement results to the interval [0, 1] and construct a standard separation vector.

[0458] e) Data analysis: The main factors affecting segregation were determined by principal component analysis and cross-validation was performed.

[0459] f) Expert review: Invite three experts in the field of concrete materials to review and revise the results.

[0460] The final standard separation vector is shown in Table 1:

[0461] Table 1 Standard separation vector data table

[0462]

[0463] 5. Segregation analysis model training

[0464] Based on the acquired standard separation vectors and collected image data, the project department trained the separation analysis model:

[0465] a) Data preprocessing: 405 original images were enhanced, including rotation, scaling, flipping, etc., and the number of images was expanded to 4050.

[0466] b) Model architecture: A multi-branch deep learning structure is adopted, including:

[0467] Cement branch network: U-Net structure, 5 layers deep;

[0468] Coarse aggregate branch network: ResNet-50 structure;

[0469] Fine aggregate branch network: DenseNet-121 structure;

[0470] Pore ​​branching network: FCN-8s structure;

[0471] Interface branch network: SegNet structure;

[0472] Fusion network: 3 fully connected layers;

[0473] c) Training parameters:

[0474] Batch size: 32

[0475] Learning rate: Initially 0.001, using cosine annealing strategy;

[0476] Optimizer: Adam;

[0477] Number of training rounds: 100;

[0478] Early stopping strategy: stop if the validation set loss does not decrease for 5 consecutive rounds;

[0479] d) Training results:

[0480] Training set accuracy: 97.8%;

[0481] Verification set accuracy: 95.2%;

[0482] Test set accuracy: 94.5%;

[0483] 6. Real-time monitoring process

[0484] On the day of raft slab pouring, the project department activated the segregation detection system. The pouring process lasted 48 hours and was monitored in real time. The following are the key nodes and data of the monitoring process:

[0485] a) 2 hours after pouring starts:

[0486] Detection point A (near the pumping pipe outlet):

[0487] Separation vector: [0.18, 0.22, 0.20, 0.25, 0.28];

[0488] Similarity: 0.92 (compared with slightly separated standard vector);

[0489] System judgment: The degree of segregation is slight and pouring can continue;

[0490] b) Casting for 12 hours:

[0491] Testing point B (20m from the pumping pipe outlet):

[0492] Separation vector: [0.38, 0.42, 0.40, 0.48, 0.52];

[0493] Similarity: 0.88 (compared with the medium separation standard vector);

[0494] System warning: There may be moderate segregation in some local areas;

[0495] Treatment measures: adjust the vibration time and frequency, and increase the frequency of manual inspections;

[0496] c) Casting for 24 hours:

[0497] Detection point C (near the edge of the foundation pit):

[0498] Segregation vector: [0.58, 0.62, 0.60, 0.68, 0.72];

[0499] Similarity: 0.86 (compared with severely separated standard vector);

[0500] System alarm: There is a serious risk of segregation in the local area;

[0501] Emergency measures: Stop pouring in this area, add a string tube device to adjust the fluidity of the concrete, and adjust the pumping speed.

[0502] d) Casting for 36 hours:

[0503] Detection point C (after processing):

[0504] Segregation vector: [0.28, 0.32, 0.30, 0.35, 0.38];

[0505] Similarity: 0.90 (compared with slightly separated standard vector);

[0506] System judgment: segregation problem is effectively controlled and pouring can continue;

[0507] e) Finish of pouring (48 hours): The final data of all 20 test points are shown in Table 2:

[0508] Table 2 Monitoring point data table

[0509]

[0510] System summary: The overall segregation is slight and the pouring quality meets the requirements

[0511] 7. Data analysis and improvement

[0512] After pouring was completed, the project department conducted a detailed analysis of the monitoring data of the entire process:

[0513] a) Segregation distribution map: Based on the data from 20 detection points, a three-dimensional distribution map of the raft segregation degree was generated, and it was found that the segregation risk was higher in the area close to the pumping pipe outlet and the edge of the foundation pit.

[0514] b) Time trend analysis: A trend graph showing the degree of segregation over time was drawn and it was found that the risk of segregation was higher in the middle period of pouring (12-24 hours), which may be related to the fatigue of the construction workers.

[0515] c) Analysis of influencing factors: Through multiple regression analysis, it was found that pumping speed, ambient temperature and concrete slump are the main factors affecting the degree of segregation.

[0516] d) Suggestions for improvement:

[0517] Optimize the layout of pumping pipes to reduce the distance of concrete transportation;

[0518] Strengthen mid-term construction management and appropriately increase the frequency of personnel rotation;

[0519] Adjust concrete mix ratio in real time according to ambient temperature;

[0520] Increase the use of string tube devices in high-risk areas.

[0521] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A device for detecting segregation during deep foundation pit concrete pouring, used to detect the segregation of concrete during the concrete pouring process, characterized in that , including a tube body, a monitoring unit and a single-chip computer, a plurality of circular through holes are opened on the side wall of the tube body, the plurality of circular through holes are evenly arranged in 2 or 3 rows on the side wall, each circular through hole is embedded with a monitoring unit, the monitoring unit includes a tempered glass sheet, a concave lens, an isolation glass sheet, a focusing sheet and an optical fiber bundle, a circular hole is opened at the bottom of the focusing sheet, one end of the optical fiber bundle is embedded in the circular hole as a collection end, the single-chip computer is provided with a controller and at least the number of image collectors of the monitoring units; the other end of the optical fiber bundle is connected to the image collector as an output end, and the image collector is used The optical signal transmitted by the optical fiber bundle is received and converted into an electrical signal, and after analog-to-digital conversion, a regional image of the monitoring area corresponding to the monitoring unit is obtained and output to the controller; the cross-section of the condenser is parabolic, and the focus of the concave lens located on the inner side of the tube body and the focus of the condenser are both located at the geometric center of the surface of the collection end; a concrete pouring process separation and analysis module is provided in the controller, which is used to calculate the separation result of the concrete pouring process based on the regional images collected by multiple image collectors and the pre-trained separation analysis model, and transmit it to the host computer through the wireless transmission module on the single-chip microcomputer.

2. A segregation detection device for deep foundation pit concrete pouring process according to claim 1, characterized in that: The outer diameter of the tube is 3-5 cm.

3. A segregation detection device for deep foundation pit concrete pouring process according to claim 2, characterized in that: The spacing of each column of circular through holes, the radius of the tempered glass sheet, the shape parameters of the concave lens, the shape parameters of the focusing sheet and the parameters of the optical fiber bundle are all optimized and limited by an optimization equation group, and the optimization equation group includes an optical imaging equation, a spatial sampling equation, a signal-to-noise ratio equation, a resolution equation and a depth detection equation.

4. A segregation detection device for deep foundation pit concrete pouring process according to claim 3, characterized in that: The specific steps of performing optimization and limitation by optimizing the set of equations include: Establish the objective function and sum the weighted equations; Set the initial values ​​and constraints of each parameter; Particle swarm algorithm is used for iterative optimization; When the number of iterations reaches the preset value or the objective function converges, the optimization result is output.

5. A segregation detection device for deep foundation pit concrete pouring process according to claim 4, characterized in that: The concrete pouring process segregation analysis module specifically performs the following steps: Acquire the regional image collected by each image collector; Performing preprocessing including denoising, enhancement and correction on the regional image to obtain a preprocessed image; All pre-processed images are spliced ​​using image registration and fusion methods to obtain a spliced ​​image; Inputting the spliced ​​image into a pre-trained segregation analysis model to obtain a segregation vector of the concrete pouring process; The similarity is calculated based on the segregation vector and the preset segregation standard vector. If the similarity is greater than the preset similarity threshold, it means that the segregation of the current concrete is large; if the similarity is less than the preset similarity threshold, it means that the segregation of the current concrete is small; and the similarity is used as the segregation result.

6. A segregation detection device for deep foundation pit concrete pouring process according to claim 5, characterized in that: The segregation vectors are specifically [cement distribution uniformity, coarse aggregate distribution uniformity, fine aggregate distribution uniformity, porosity, and interface transition zone thickness].

7. A segregation detection device for deep foundation pit concrete pouring process according to claim 6, characterized in that: The segregation analysis model adopts a multi-branch deep learning structure, specifically including a cement branch network, a coarse aggregate branch network, a fine aggregate branch network, a pore branch network, an interface branch network and a fusion network.

8. A segregation detection device for deep foundation pit concrete pouring process according to claim 7, characterized in that: The cement branch network is used to extract cement-related features, the input is a preprocessed grayscale image, the output is a cement distribution feature map, and the structure is a U-Net convolutional neural network; The coarse aggregate branch network is used to extract coarse aggregate related features, the input is a preprocessed color image, the output is a coarse aggregate distribution feature map, and the structure is a ResNet convolutional neural network; The fine aggregate branch network is used to extract fine aggregate related features, the input is a preprocessed color image, the output is a fine aggregate distribution feature map, and the structure is a DenseNet convolutional neural network; The pore branch network is used to extract pore-related features, the input is a preprocessed binary image, the output is a pore distribution feature map, and the structure is a FCN fully convolutional network; The interface branch network is used to extract the interface features between aggregate and cement paste. The input is the preprocessed edge enhanced image, the output is the interface distribution feature map, and the structure is the SegNet segmentation network. The fusion network is used to integrate the output features of each branch network, the input is the output feature map of each branch network, the output is the final separated feature vector, and the structure is a multi-layer perceptron.

9. The device for detecting segregation during deep foundation pit concrete pouring process according to claim 1, characterized in that: The tube body is made of stainless steel; the bottom end of the tube body is also provided with a conical structure.

10. The device for detecting segregation during deep foundation pit concrete pouring process according to claim 5, characterized in that: The preset similarity threshold is 0.85.

Images