Concrete fluidity increasing method based on vibrating rod, medium and system
By establishing a system of mechanical equations and deep learning models, combining signal processing technology to optimize the position and operating parameters of the vibrator, the problem of vibrator vibration optimization in the existing technology depends on experience and neglecting key factors, and concrete flow is achieved precision optimization and improvement of construction efficiency.
Patent Information
- Application Number
- CN202411049541.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-01
- Publication Date
- 2024-11-26
AI Technical Summary
The existing vibration optimization methods of vibrating rods rely on empirical adjustments, making it difficult to achieve accurate optimization, and ignore the impact of vibrating rod position, concrete surface deformation and internal vibration response on fluidity, resulting in insufficient comprehensive fluidity evaluation.
By obtaining the position and operating parameters of the vibrator, the concrete surface image and vibration signal, the mechanical equation system is established and analyzed and solved. Combined with Fourier transform, wavelet analysis and principal component analysis, the concrete vibration field matrix is constructed and the depth is designed. Learn the model to predict fluidity and optimize the position and operating parameters of the vibrator.
Accurate optimization of concrete fluidity has been achieved, construction quality and efficiency have been improved, and through scientific modeling methods and intelligent optimization methods, the problems of inaccurate adjustment of vibration parameters and incomplete liquidity evaluation in the existing technology have been solved. .
Smart Images

Figure CN119026499A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of concrete construction, and in particular relates to a method, medium and system for increasing the fluidity of concrete based on a vibrating rod. Background Art
[0002] Concrete is a vital building material widely used in current construction projects. Its fluidity is a key factor influencing the quality and efficiency of concrete construction. Good concrete fluidity not only ensures that concrete fills the mold during pouring, minimizing air bubbles and honeycomb defects, but also improves concrete density and durability. Therefore, improving concrete fluidity has long been a key concern in the industry.
[0003] Currently, there are two main methods for improving concrete fluidity: one is to adjust the concrete mix, such as reducing the water-cement ratio or using admixtures, to improve the viscosity and plasticity of concrete; the other is to use mechanical vibration, using a vibrator or vibration table to apply force to the concrete to increase its fluidity. Mechanical vibration is widely used in engineering practice because it can directly act on poured concrete.
[0004] However, existing vibrator vibration optimization methods have some limitations. First, the operating parameters of the vibrator, such as vibration frequency, amplitude, and power, are usually adjusted empirically based on engineering experience, lack a scientific basis, and are difficult to achieve precise optimization. Second, existing optimization methods mainly focus on optimizing a single vibrator parameter, while ignoring the influence of other important factors on fluidity, such as vibrator position, concrete surface deformation, and internal vibration response. Third, existing fluidity evaluation methods are too simplistic and cannot fully reflect the microstructural properties and macroscopic flow behavior of concrete. Therefore, there is an urgent need to develop a new vibrator-based concrete fluidity enhancement method to overcome the limitations of existing technologies and improve the quality and efficiency of concrete construction. Summary of the Invention
[0005] In view of this, the present invention provides a method, medium and system for increasing the fluidity of concrete based on a vibrating rod, which can solve the technical problem that the adjustment of the operating parameters of the vibrating rod in the prior art often relies on manual experience and is difficult to achieve precise optimization.
[0006] The present invention is achieved in that:
[0007] A first aspect of the present invention provides a method for increasing the fluidity of concrete using a vibrating rod, comprising the following steps:
[0008] S10, acquiring the position and operating parameters of the vibrator, as well as a surface image of the concrete and vibration signals at multiple points, wherein the operating parameters include vibration frequency, amplitude, power, and action time;
[0009] S20. Establish a set of mechanical equations, including the equations of motion, continuity equations, constitutive equations, boundary condition equations, and initial condition equations;
[0010] S30, analytically solving the mechanical equations to obtain multiple analytical solutions, including a velocity field analytical solution for describing concrete flow, a pressure field analytical solution for describing pressure distribution, a stress field analytical solution for describing stress distribution, and a displacement field analytical solution for describing deformation;
[0011] S40, performing numerical calculations on the multiple analytical solutions according to preset boundary conditions to obtain multiple numerical ranges, namely, a velocity field numerical range, a pressure field numerical range, a stress field numerical range, and a displacement field numerical range;
[0012] S50, based on the multiple analytical solutions, multiple numerical ranges, and vibration signals at multiple points, establish a concrete vibration field matrix by using a method combining Fourier transform, wavelet analysis, and principal component analysis;
[0013] S60. Input the concrete vibration field matrix and the surface image of the concrete into a pre-trained concrete fluidity prediction model to obtain a concrete vibration optimization matrix, where each element of the matrix is a vector representing the position and operating parameters of the vibrator; the concrete vibration optimization matrix is output to an operator user to adjust the position and operating parameters of the vibrator.
[0014] The concrete fluidity prediction model includes a micro-feature extraction network, a macro-feature extraction network and a feature fusion network; the micro-feature extraction network is used to capture the distribution of fine particles in concrete, the microscopic pore structure and the local flow characteristics; the macro-feature extraction network is used to analyze large-scale characteristics, including the overall fluidity of concrete, temperature distribution, and pressure field; the feature fusion network is used to integrate micro- and macro-features to generate the final concrete fluidity prediction.
[0015] The specific structure of the microscopic feature extraction network is:
[0016] 1. Multi-scale convolution layer: Use 3x3, 5x5, and 7x7 convolution kernels to process input data in parallel and capture micro features of different scales.
[0017] 2. Spatial Pyramid Pooling (SPP) layer: Divides the feature map into 1x1, 2x2, and 4x4 regions, performs maximum pooling, and extracts multi-scale local features.
[0018] 3. Attention mechanism: including channel attention and spatial attention, adaptively focusing on important micro features.
[0019] 4. Residual connections: Use residual blocks to deepen the network while maintaining gradient flow.
[0020] 5. Batch Normalization: Used after each convolutional layer to speed up training and improve model stability.
[0021] The specific structure of the macro feature extraction network is:
[0022] 1. Deep convolutional network: ResNet-50 is used as the backbone network to extract powerful semantic features.
[0023] 2. Dilated convolution: Use dilated convolution with different dilation rates at the back end of the network to expand the receptive field.
[0024] 3. Feature Pyramid Network (FPN): fuses features from different depth layers to obtain multi-scale representations.
[0025] 4. Global average pooling: captures global context information.
[0026] 5. Self-attention mechanism: Establish long-range dependencies in feature graphs to enhance the expressiveness of features.
[0027] The specific structure of the feature fusion network is:
[0028] 1. Feature splicing layer: splicing micro and macro features in the channel dimension.
[0029] 2.1x1 convolution layer: reduces the number of feature channels and reduces computational complexity.
[0030] 3. Adaptive feature weighting: learning the importance weights of micro and macro features.
[0031] 4. Multi-layer Perceptron (MLP): performs nonlinear transformation and enhances feature expression.
[0032] 5. Skip connection: retains original feature information and prevents information loss.
[0033] 6.Softmax classification layer: outputs the final liquidity prediction results.
[0034] The process of constructing a training dataset includes the following steps: First, raw data such as vibrating rod parameters, concrete surface images, and vibration signals are collected through actual concrete vibration experiments, and a large amount of simulation data is generated using computational fluid dynamics software. Then, the experimental data and the simulation data are matched and annotated to establish a correlation between vibration parameters, field quantity data, and fluidity measurement results. Next, the data is preprocessed, including image normalization, numerical data standardization, and vibration signal denoising. To increase data diversity, a variety of data enhancement techniques are used, such as image rotation, cropping, flipping, adding noise and interpolation to numerical data, and time offset and amplitude scaling of vibration signals. Finally, the processed dataset is divided into training set, validation set, and test set in a ratio of 8:1:1 to prepare for model training.
[0035] The training process of the concrete fluidity prediction model is as follows: First, the macro-feature extraction network is initialized using pretrained ResNet-50 weights, and the remaining network components are randomly initialized. A hybrid loss function and the Adam optimizer are used, with a learning rate decay strategy. Training is divided into three stages: first, the macro-network is fixed while the remaining components are trained, and then the entire model is gradually unfrozen and fine-tuned. Training is performed using mini-batch gradient descent, while implementing an early stopping strategy to prevent overfitting. Model performance is evaluated through cross-validation, and hyperparameters are automatically tuned using Bayesian optimization. To improve prediction stability, multiple models are trained and ensembled. Finally, model performance is evaluated on the test set, and feature importance is analyzed using SHAP values. Visualizations are generated to explain the model's decision-making process. This training process aims to build a concrete fluidity prediction model that is both accurate and has good generalization capabilities.
[0036] The motion equation is specifically expressed as:
[0037]
[0038] Where: ρ is the density of concrete, obtained through experimental measurement; v is the velocity vector field; t is time; p is the pressure field; μ is the dynamic viscosity, measured through rheological experiments; g is the gravitational acceleration vector; F is the external force field, including the force generated by the vibrator; The gradient operator represents the rate of change of a scalar field in space and is defined as
[0039] Wherein, the continuity equation is specifically expressed as:
[0040]
[0041] Wherein, the constitutive equation is specifically expressed as:
[0042] When τ>τ0;
[0043] When τ≤τ0;
[0044] Where: τ is the shear stress; τ0 is the yield stress, determined by experiment; μ p is the plastic viscosity, determined experimentally; is the shear rate.
[0045] Wherein, the boundary condition equation is specifically expressed as:
[0046] v = v0 on the surface of the vibrator;
[0047] p=p0 at the free surface;
[0048] T = T0 at the outer boundary;
[0049] Where: v0 is the surface velocity of the vibrating rod, which is determined by the vibrating rod parameters; p0 is the atmospheric pressure; T0 is the ambient temperature.
[0050] The initial condition equation is specifically expressed as:
[0051] v(x,0)=v i (x);
[0052] p(x,0)=p i (x);
[0053] T(x,0)=T i (x);
[0054] Where: v i (x) is the initial velocity field distribution; p i (x) is the initial pressure field distribution; T i (x) is the initial temperature field distribution.
[0055] The concrete vibration optimization matrix is specifically expressed as:
[0056]
[0057] Where: M is the concrete vibration optimization matrix, each element m ij is a vector representing the position and operating parameters of the vibrator; m is the number of vibrators; n is the number of vibrator parameters that need to be optimized, usually including position coordinates (x, y, z), vibration frequency f, amplitude A, power P and action time t; m ij =[x i ,y i ,z i ,f j ,A j ,P j ,tj ] is the jth optimized parameter vector of the i-th vibrator; (x i ,y i ,z i ) is the position coordinate of the i-th vibrator, in m; f j is the jth optimized vibration frequency, in Hz; A j is the jth optimized vibration amplitude, in m; P j is the jth optimized vibration power, in W; t j is the jth optimized vibration action time, in seconds.
[0058] Specifically, the step S10 includes:
[0059] Sensors installed on the vibrator provide real-time data on its operating parameters, including its position, posture, vibration frequency, amplitude, and power. High-speed cameras or structured light scanning technologies are also used to capture real-time image data of the concrete surface. Accelerometers are placed at key locations within the concrete to record the evolution of vibration signals over time. This data provides crucial foundational information for subsequent mechanical analysis and fluidity prediction.
[0060] Wherein, the step S20 includes:
[0061] First, the equations of motion describing concrete flow were established based on the Navier-Stokes equations, taking into account factors such as density, velocity field, pressure field, viscosity, gravity, and external forces. Second, the continuity equation for concrete density changes was established based on the law of conservation of mass. Third, the Bingham viscoplastic model was used to describe the stress-strain relationship in concrete. Velocity boundary conditions on the vibrator surface, pressure boundary conditions on the free surface, and temperature boundary conditions on the external boundaries were also established. Finally, the initial velocity, pressure, and temperature field distributions were given. This set of mechanical equations laid the foundation for subsequent numerical analysis.
[0062] Wherein, the step S30 includes:
[0063] For the equations of motion, an analytical solution for the velocity field is derived, expressed as a series consisting of parameters such as the time exponential function, the sine function, and the wave number vector. For the continuity equation, an analytical solution for the pressure field is derived, expressed as a series consisting of parameters such as the time exponential function, the cosine function, and the wave number vector. For the Bingham constitutive model, an analytical solution for the stress field is derived, expressed as a series consisting of parameters such as the time exponential function, the cosine function, and the wave number vector. For the initial conditions, an analytical solution for the displacement field is derived, expressed as a series consisting of parameters such as the time integral, the sine function, and the wave number vector. These analytical solutions provide an important theoretical basis for subsequent numerical calculations and fluidity predictions.
[0064] Wherein, the step S40 includes:
[0065] First, appropriate spatial and temporal step sizes are selected to generate grid points within the computational domain. Then, the analytical solutions for the velocity, pressure, stress, and displacement fields are substituted into the aforementioned formulas to calculate the corresponding field quantities at each grid point. Next, the mean and standard deviation of each field quantity are calculated, and their numerical ranges are determined according to the 3σ principle. This numerical calculation method yields the numerical ranges for the velocity, pressure, stress, and displacement fields, providing important basic data for subsequent vibration field analysis and fluidity prediction.
[0066] Wherein, the step S50 includes:
[0067] First, the Fast Fourier Transform (FFT) algorithm is used to perform frequency domain analysis on the vibration signals collected at multiple measuring points, obtaining the amplitude and phase information of each frequency component, reflecting the periodic characteristics of particle movement within the concrete. Second, a continuous wavelet transform or discrete wavelet transform is used to perform multi-scale time-frequency analysis on the vibration signals, extracting wavelet coefficient features that can describe the microscopic pore structure and local flow characteristics of the concrete. Third, the feature quantities extracted from the Fourier spectrum and wavelet coefficients are input into the principal component analysis (PCA) algorithm for feature dimensionality reduction, obtaining principal component scores that comprehensively reflect the vibration field characteristics of each measuring point. Finally, these vibration field feature quantities are integrated with the aforementioned numerical range data to construct a matrix model describing the vibration state of concrete, providing important input data for subsequent flowability prediction.
[0068] Wherein, the step S60 includes:
[0069] First, the micro-feature extraction network utilizes parallel multi-scale convolutional layers, spatial pyramid pooling, channel-level and spatial attention mechanisms, as well as residual connections and batch normalization techniques to effectively capture micro-scale features such as the distribution of fine particles, microscopic pore structure, and local flow characteristics in concrete. Second, the macro-feature extraction network, based on the ResNet-50 backbone, employs dilated convolutions, a feature pyramid network, global average pooling, and a self-attention mechanism to analyze macro-scale features such as the overall concrete flowability, temperature distribution, and pressure field. Finally, the feature fusion network concatenates and weights the micro- and macro-features, and after a nonlinear transformation, outputs the final concrete flowability prediction result. During the training process, strategies such as staged optimization, cross-validation, and Bayesian hyperparameter optimization were employed to improve the model's prediction accuracy and generalization capabilities.
[0070] Optionally, step S10 further includes:
[0071] Measuring instruments provide real-time information on the vibrator's position and posture in three-dimensional space, including parameters such as xyz coordinates, pitch, yaw, and roll angles. High-speed cameras also capture real-time images of the concrete surface deformation during vibration. Furthermore, accelerometers are placed at key locations within the concrete to record vibration acceleration signals along the three coordinate axes. This rich data provides comprehensive foundational support for subsequent fluid dynamics analysis and machine learning predictions.
[0072] Optionally, step S20 further includes:
[0073] When establishing the equations of motion, the Navier-Stokes equations were appropriately simplified and linearized, ignoring some minor terms to obtain an analytical solution. When addressing the constitutive equations, the Bingham viscoplastic model was used to describe the stress-strain relationship of concrete. This model accurately reflects the yield characteristics and non-Newtonian rheology of concrete. When setting boundary conditions, a velocity boundary was set on the vibrator surface, a pressure boundary was set on the free surface, and a temperature boundary was set on the external boundary to more realistically describe the actual site conditions. This mathematical modeling work laid a good foundation for subsequent analytical solutions and numerical calculations.
[0074] Optionally, step S30 further includes:
[0075] When deriving the analytical solution for the velocity field, the separation of variables method was used to express the velocity field as a series consisting of parameters such as the time exponential function, the sine function, and the wave number vector. This effectively describes the motion characteristics of particles within concrete. When deriving the analytical solution for the pressure field, the analytical solution of the velocity field was substituted into the continuity equation, and the analytical expression for the pressure field was obtained by solving the Poisson equation, reflecting the temporal and spatial variations of the pressure distribution. When deriving the analytical solution for the stress field, the Bingham constitutive model was combined to express the stress field as a series consisting of parameters such as the time exponential function, the cosine function, and the wave number vector. This accurately characterizes the stress distribution characteristics within concrete. These analytical solutions provide an important theoretical foundation for subsequent numerical analysis and parameter optimization.
[0076] Optionally, step S40 further includes:
[0077] When discretizing the computational domain, appropriate spatial and temporal steps are selected based on the actual scale of concrete vibration and the requirements for numerical convergence. When calculating the numerical values of each field quantity, the corresponding analytical solution formula is substituted for each grid point, and the calculation is automatically performed through programming. When statistically analyzing the numerical range, the mean and standard deviation of each field quantity are calculated, and their upper and lower limits are determined according to the 3σ principle to cover the vast majority of numerical variations. This comprehensive numerical analysis lays a solid foundation for subsequent vibration field construction and flow prediction.
[0078] Optionally, step S50 further includes:
[0079] When using Fourier transforms, the fast Fourier transform (FFT) algorithm is selected to efficiently convert time-domain vibration signals into the frequency domain and extract the amplitude and phase characteristics of each frequency component. When using wavelet analysis, either the continuous wavelet transform (CWT) can be used to obtain detailed time-frequency characteristics, or the discrete wavelet transform (DWT) can be used to reduce computational complexity. When performing principal component analysis (PCA), the features extracted from the Fourier spectrum and wavelet coefficients are used as input, and dimensionality reduction is used to obtain principal component scores that comprehensively reflect the vibration state of each measuring point. This fusion of multiple signal processing techniques enables the vibration field matrix to comprehensively describe the microstructure and macroscopic flow characteristics within the concrete.
[0080] Optionally, the microscopic feature extraction network described in step S60 further includes:
[0081] Parallel multi-scale convolutional layers, including convolution kernels with different receptive fields (3x3, 5x5, and 7x7), are used to extract micro-features at different scales in parallel. Spatial Pyramid Pooling (SPP) layers are used to divide feature maps into multiple regions (1x1, 2x2, and 4x4), performing max pooling to obtain multi-scale local features. Channel and spatial attention mechanisms are introduced to adaptively focus on micro-features that are more important for fluidity prediction. Residual connections and batch normalization techniques are used to deepen the network structure and accelerate training convergence. These innovative network designs effectively capture the complex characteristics of concrete microstructure.
[0082] Optionally, the macro feature extraction network described in step S60 further includes:
[0083] A ResNet-50 network is used as the backbone to extract powerful semantic features. Atrous convolution is used at the back end to expand the receptive field and capture a wider range of contextual information. A Feature Pyramid Network (FPN) is employed to fuse features from different depth layers to obtain multi-scale representations. Global average pooling is used to capture global features, and a self-attention mechanism is introduced to establish long-range dependencies within the feature map. This network structure is capable of effectively analyzing macroscopic characteristics such as the overall fluidity, temperature distribution, and pressure field of concrete.
[0084] Optionally, the feature fusion network described in step S60 further includes:
[0085] First, micro- and macro-features are concatenated in the channel dimension to increase feature richness. A 1x1 convolutional layer is then used to reduce the number of feature channels and reduce computational complexity. An adaptive feature weighting mechanism is introduced to learn the importance weights of micro- and macro-features. A multi-layer perceptron (MLP) is used for nonlinear transformation to enhance feature representation. Skip connections are used to preserve original feature information and prevent information loss. Finally, a softmax classification layer is used to output the final concrete flowability prediction results. This feature fusion network fully exploits the complementarity of micro- and macro-features, improving prediction accuracy.
[0086] Optionally, the model training process in step S60 further includes:
[0087] First, the macro feature extraction network is initialized using a pretrained ResNet-50 to accelerate convergence. The entire model is then trained in stages, first freezing the macro network and training the remaining components, and then gradually unfreezing and fine-tuning the entire model. A hybrid loss function and the Adam optimizer are used, along with a learning rate decay strategy, to improve training stability. Model performance is evaluated through cross-validation, and Bayesian optimization methods are used to automatically adjust hyperparameters and enhance model generalization. To improve the stability of prediction results, multiple models are trained and integrated. Finally, the model's prediction accuracy is evaluated on the test set, and feature importance is analyzed using SHAP values. Visualizations are generated to explain the model's decision-making process. This integrated training strategy ensures model accuracy and good generalization.
[0088] Optionally, when constructing the training dataset, the following steps are also included:
[0089] First, through actual concrete vibration experiments, raw data such as vibrator parameters, concrete surface images, and vibration signals were collected. Simultaneously, computational fluid dynamics software was used to generate a large amount of simulation data. The experimental and simulation data were then matched and annotated to establish correlations between vibration parameters, field data, and flowability measurements. Next, data preprocessing was performed, including image normalization, numerical data standardization, and vibration signal noise reduction. To increase data diversity, data enhancement techniques such as image rotation, cropping, and flipping, numerical data noise addition and interpolation, and vibration signal time shifting and amplitude scaling were employed.
[0090] Optionally, further, the step S10 further includes:
[0091] When acquiring vibrator parameters, in addition to position and posture information, operating parameters such as vibration frequency, amplitude, and power are measured in real time. These parameters reflect the dynamic load characteristics applied by the vibrator to the concrete, providing an important basis for subsequent mechanical analysis. Furthermore, when capturing images of the concrete surface, high-resolution cameras are used to capture even more detailed surface deformation information. Furthermore, when deploying vibration sensors, not only acceleration signals are measured, but also velocity and displacement signals, to comprehensively describe the vibration state within the concrete. This rich data facilitates the construction of a more accurate vibration field matrix model.
[0092] Optionally, further, the step S20 also includes:
[0093] When establishing the equations of motion, in addition to considering inertial forces, pressure gradient forces, viscous forces, and gravity, an external excitation force term generated by the vibrator is also introduced. This precise description of external forces can more realistically reflect the dynamic loads acting on concrete during the vibration process. When addressing the constitutive equations, in addition to using the Bingham viscoplastic model, other non-Newtonian fluid models, such as the Herschel-Bulkley model, can also be considered to better capture the complex rheological properties of concrete. When setting boundary conditions, in addition to setting corresponding conditions on the vibrator surface, free surface, and external boundaries, contact boundary conditions between the concrete and the mold can also be considered to more comprehensively describe the on-site construction environment. These modeling details help improve the accuracy and reliability of mechanical analysis.
[0094] Optionally, further, in the analytical solution derivation process of step S30, the following is also included:
[0095] For the equation of motion, the Fourier series expansion method, based on the separation of variables method, is further employed to represent the velocity field as an infinite series consisting of a time exponential function, a sine function, and a series of wave number vectors. This representation provides a more detailed description of the motion characteristics of particles within concrete. For the continuity equation, when solving the Poisson equation, the separation of variables method or the Green function method can be considered to obtain an analytical form of the pressure field. These analytical solutions have clear physical meaning and contribute to a deeper understanding of the mechanism of concrete flow. For the constitutive equation, when deriving analytical solutions to the stress field, in addition to the Bingham model, other constitutive relations, such as the Herschel-Bulkley model, can also be considered to more comprehensively characterize the stress-strain behavior of concrete. These analytical solutions provide a theoretical basis for subsequent numerical calculations and parameter optimization.
[0096] Optionally, further, the step S40 also includes:
[0097] When discretizing the computational domain, in addition to selecting appropriate spatial and time steps, different mesh types, such as structured or unstructured meshes, can be used to improve computational efficiency, depending on the geometry and boundary conditions of the actual problem. When calculating the numerical values of various field quantities, in addition to directly substituting analytical solutions into the equations, numerical discretization methods such as the finite volume method and the finite element method can also be used to improve computational stability and accuracy. When statistically analyzing numerical ranges, in addition to using the 3σ criterion, other statistical quantities, such as extreme values and confidence intervals, can also be considered to more comprehensively characterize the numerical characteristics of each field quantity. These numerical calculation strategies help improve the applicability and reliability of the model.
[0098] Optionally, further, the step S50 further includes:
[0099] When employing Fourier transforms, in addition to the FFT algorithm, other variants, such as the short-time Fourier transform, can be used to obtain more refined time-frequency features. When using wavelet analysis, in addition to the CWT and DWT, wavelet packet analysis can also be considered to better capture the non-stationary characteristics of concrete vibration signals. When performing PCA, in addition to extracting principal component scores, other dimensionality reduction techniques, such as linear discriminant analysis and t-SNE, can be combined to further compress feature dimensions and improve computational efficiency. The combined application of these signal processing techniques enables the constructed vibration field matrix to more comprehensively characterize the microstructure and macroscopic flow characteristics of concrete.
[0100] Optionally, further, the deep learning model described in step S60 also includes:
[0101] In micro-feature extraction networks, in addition to using multi-scale convolutional layers, SPP layers, attention mechanisms, residual connections, and batch normalization techniques, other innovative modules such as sparse convolutional layers, separable convolutional layers, and squeeze-and-excitation modules can be introduced to further improve the efficiency and accuracy of feature extraction. In macro-feature extraction networks, in addition to using ResNet-50 as the backbone, other mainstream CNN models such as VGGNet, Inception, and MobileNet can also be explored to achieve more powerful semantic feature representations. In feature fusion networks, in addition to feature concatenation, adaptive weighting, MLP, and Softmax layers, modules such as attention mechanisms, gating mechanisms, and transformers can be introduced to enhance the modeling capabilities of feature fusion. These innovative network designs can help further improve the performance of prediction models.
[0102] Optionally, further, the model training process in step S60 further includes:
[0103] When initializing network weights, in addition to using pre-trained ResNet-50 parameters, other pre-trained models, such as ImageNet pre-trained weights or self-supervised pre-trained weights, can also be considered to obtain better initial feature representations. In a phased training strategy, in addition to first fixing the macro network and then fine-tuning the entire model, other strategies can be explored, such as training the micro network first and then the macro network, or adopting a joint training approach. When selecting an optimization algorithm, in addition to Adam, other optimizers such as SGD, RMSProp, and AdamW can also be used, combined with techniques such as dynamic learning rate adjustment and gradient clipping to further improve training stability. When ensembling models, in addition to simple averaging, more complex ensemble methods such as weighted averaging, stacking, and blending can also be explored to obtain more robust prediction results. These innovations in training strategies can help further improve the model's generalization and predictive performance.
[0104] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for increasing concrete fluidity based on a vibrating rod.
[0105] A third aspect of the present invention provides a vibrator-based concrete fluidity increasing system, which includes the above-mentioned computer-readable storage medium.
[0106] Compared with the prior art, the method, medium and system for increasing concrete fluidity based on a vibrating rod provided by the present invention have the following beneficial effects:
[0107] By combining fluid mechanics analysis and machine learning technology, in-depth research and intelligent optimization of the concrete vibration process are achieved.
[0108] First, the method fully utilizes multi-source data, including vibrator operating parameters, concrete surface images, and internal vibration signals, to construct a mechanical model describing concrete flow and derive analytical solutions for the velocity, pressure, stress, and displacement fields. These analytical solutions not only reflect the microstructural properties of concrete but also describe its macroscopic flow behavior, laying the theoretical foundation for subsequent numerical analysis and parameter optimization.
[0109] Secondly, the method uses signal processing techniques such as Fourier transform, wavelet analysis, and principal component analysis to construct a matrix model that comprehensively reflects the vibration state of concrete. This vibration field matrix integrates both microscopic and macroscopic features, providing rich input data for the machine learning model and significantly improving the accuracy and reliability of flowability prediction.
[0110] Thirdly, this method employs an innovative deep learning model, comprising a micro-feature extraction network, a macro-feature extraction network, and a feature fusion network. This multi-scale, multi-modal feature fusion architecture fully exploits the correlation between the concrete's internal microstructure and overall flow behavior, providing more accurate flow predictions. Furthermore, the model employs strategies such as staged training, cross-validation, and Bayesian optimization to further enhance its predictive performance and generalization capabilities.
[0111] In general, the vibrator-based concrete fluidity enhancement method proposed in this invention fully integrates fluid mechanics analysis and machine learning technology, and has the following three advantages over existing technologies:
[0112] 1. A more comprehensive concrete flow dynamics model was established, which not only took into account the parameters of the vibrating rod, but also integrated key factors such as the concrete surface deformation and internal vibration response, providing theoretical support for the precise optimization of the vibration process.
[0113] 2. A prediction model based on vibration field matrix was constructed, which can better capture the microstructure and macroscopic flow characteristics of concrete, and significantly improve the accuracy and reliability of fluidity prediction.
[0114] 3. Deep learning technology is used to realize intelligent prediction and optimization of concrete fluidity, providing strong support for improving construction quality and efficiency in engineering practice.
[0115] In summary, the solution of the present invention solves the technical problem in the prior art that the adjustment of the operating parameters of the vibrating rod often relies on manual experience and is difficult to achieve precise optimization. BRIEF DESCRIPTION OF THE DRAWINGS
[0116] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION
[0117] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
[0118] like Figure 1 FIG. 1 is a flow chart of a method for increasing concrete fluidity based on a vibrating rod provided by the present invention. The method comprises the following steps:
[0119] S10, obtaining the position and operating parameters of the vibrator, as well as the surface image of the concrete and vibration signals at multiple points, the operating parameters including vibration frequency, amplitude, power and action time;
[0120] S20. Establish a set of mechanical equations, including the equations of motion, continuity equations, constitutive equations, boundary condition equations, and initial condition equations;
[0121] S30, analytically solving the mechanical equations to obtain multiple analytical solutions, including a velocity field analytical solution for describing concrete flow, a pressure field analytical solution for describing pressure distribution, a stress field analytical solution for describing stress distribution, and a displacement field analytical solution for describing deformation;
[0122] S40, performing numerical calculations on the multiple analytical solutions according to the preset boundary conditions to obtain multiple numerical ranges, namely, a velocity field numerical range, a pressure field numerical range, a stress field numerical range, and a displacement field numerical range;
[0123] S50, based on multiple analytical solutions, multiple numerical ranges, and vibration signals at multiple points, using a method combining Fourier transform, wavelet analysis, and principal component analysis, establish a concrete vibration field matrix;
[0124] S60. Input the concrete vibration field matrix and the surface image of the concrete into a pre-trained concrete fluidity prediction model to obtain a concrete vibration optimization matrix, where each element of the matrix is a vector representing the position and operating parameters of the vibrator; the concrete vibration optimization matrix is output to the operator user to adjust the position and operating parameters of the vibrator.
[0125] The specific implementation of the above steps is described in detail below:
[0126] Step S10: Obtain the position and operating parameters of the vibrator, as well as the surface image of the concrete and the vibration signals of multiple points
[0127] The specific implementation of this step is as follows:
[0128] First, the sensors installed on the vibrator are used to collect the vibrator's operating parameters such as position, posture, vibration frequency, amplitude and power in real time.
[0129] At the same time, high-speed cameras or structured light scanning technologies are used to obtain real-time image data of the concrete surface.
[0130] Acceleration sensors are placed at several key locations inside the concrete to record the changes in vibration signals over time.
[0131] These data include the kinematic parameters of the vibrator, the deformation changes of the concrete surface, and the vibration response of the internal particles, providing important basic data for subsequent mechanical analysis and fluidity prediction.
[0132] Step S20: Establishing a set of mechanical equations
[0133] The specific implementation of this step is as follows:
[0134] 1. Equation of motion: Based on the Navier-Stokes equations, the equation of motion describing the flow of concrete is established. This equation takes into account factors such as the density ρ, velocity vector field v, pressure field p, viscosity μ, gravitational acceleration g, and the external force F generated by the vibrator:
[0135]
[0136] 2. Continuity equation: According to the law of conservation of mass, establish the continuity equation that describes the change of concrete density:
[0137]
[0138] 3. Constitutive equation: Use Bingham viscoplastic model to describe the stress-strain relationship of concrete:
[0139] When τ>τ0
[0140] When τ≤τ0
[0141] Where τ is the shear stress, τ0 is the yield stress, μ p is the plastic viscosity, is the shear rate.
[0142] 4. Boundary condition equation:
[0143] On the surface of the vibrator, v = v0, where v0 is the velocity of the vibrator surface;
[0144] At the free surface of concrete, p = p0, where p0 is the atmospheric pressure;
[0145] At the external boundary, T = T0, where T0 is the ambient temperature.
[0146] 5. Initial condition equation:
[0147] v(x,0)=v i (x),p(x,0)=p i (x),T(x,0)=T i (x)
[0148] Among them, v i (x),p i (x) and T i (x) are the initial velocity field, pressure field and temperature field distribution respectively.
[0149] These mechanical equations describe the various physical processes of concrete flow and lay the foundation for subsequent numerical analysis.
[0150] Step S30: Analytically solving the mechanical equations
[0151] The specific implementation of this step is as follows:
[0152] 1. Derivation of the analytical solution of the velocity field:
[0153] First, by linearizing the Navier-Stokes equations and separating the variables, the analytical solution of the velocity field can be obtained as follows:
[0154]
[0155] Where v is the kinematic viscosity, k n is the wave number vector, A n and λ n Determined by the initial conditions.
[0156] 2. Derivation of analytical solution of pressure field:
[0157] Using the analytical solution of the velocity field, the analytical solution of the pressure field can be derived as:
[0158]
[0159] Among them, B n and μ n Determined by the velocity field and boundary conditions, l n is the wave number vector of the pressure field.
[0160] 3. Derivation of analytical solution of stress field:
[0161] Substituting the analytical solution of the velocity field into the strain rate calculation and combining it with the Bingham viscoplastic model, the analytical solution of the stress field can be obtained as follows:
[0162]
[0163] Among them, C n and v n Determined by the analytical solution of the velocity field, m n is the wave number vector of the stress field.
[0164] 4. Derivation of analytical solution of displacement field:
[0165] By integrating the analytical solution of the velocity field over time, the analytical solution of the displacement field can be obtained as:
[0166]
[0167] Through the above derivation, the analytical solutions of four key field quantities describing the flow characteristics of concrete were obtained, including velocity field, pressure field, stress field and displacement field, which provided a theoretical basis for subsequent numerical calculations and fluidity predictions.
[0168] Step S40: Numerical calculation of the multiple analytical solutions
[0169] The specific implementation of this step is as follows:
[0170] 1. Discretize the computational domain:
[0171] Select appropriate spatial step Δx, Δy, Δz and time step Δt, and x ]×[0,L y ]×[0,L z ]×[0,T] to generate grid points.
[0172] 2. Calculate the numerical value of the analytical solution:
[0173] For each grid point (x i ,y j ,z k ,t l ), substitute into the above analytical solution formula and calculate the corresponding velocity field value v ijkl , pressure field value p ijkl , stress field value τ ijkl and displacement field value u ijkl .
[0174] 3. Statistical analysis value range:
[0175] For each field quantity, calculate its mean value and standard deviation, and then determine its numerical range according to the 3σ principle:
[0176] The velocity field value range is arrive
[0177] The pressure field value range is arrive
[0178] The stress field value range is arrive
[0179] The displacement field value range is arrive
[0180] Through this numerical calculation method, the numerical ranges of the above four key field quantities can be obtained, providing important basic data for subsequent vibration field analysis and fluidity prediction.
[0181] The following is the derivation process of each analytical solution:
[0182] First, we start with the analytical solution of the Navier-Stokes equations:
[0183] 1. Analytical solution of velocity field:
[0184] Starting with the simplified Navier-Stokes equations:
[0185] for
[0186] where ν = μ / ρ is the kinematic viscosity.
[0187] Step a. Linearization:
[0188] Assume that the pressure gradient and gravity terms can be expressed as Then the equation becomes:
[0189]
[0190] Step b. Separate variables:
[0191] Assume that the solution is of the form v(x,t)=T(t)X(x)
[0192] Step c. Fourier series expansion:
[0193]
[0194] where k n is the wave number vector.
[0195] Step d. Solve the ordinary differential equation:
[0196] Substituting into the original equation we get:
[0197]
[0198] Separating the variables yields:
[0199]
[0200] in is the separation constant.
[0201] Solve the equation for the time portion:
[0202]
[0203] The solution is:
[0204]
[0205] Step e. Apply boundary conditions and initial conditions:
[0206] Assume that at t = 0, v(x,0) = v0(x)
[0207] Step f. Obtain the analytical solution of the velocity field:
[0208]
[0209] Among them A n Determined by the initial conditions.
[0210] 2. Analytical solution of pressure field:
[0211] Step a. Use velocity field analytical solution:
[0212]
[0213] Step b. Substitute into the continuity equation:
[0214]
[0215] Step c. Solve Poisson's equation:
[0216]
[0217] Assume that the pressure field is of the form:
[0218]
[0219] Step d. Apply boundary conditions:
[0220] On the free surface, p = p0
[0221] Step e. Obtain the analytical solution of the pressure field:
[0222]
[0223] Among them B n and μ n Determined by the velocity field and boundary conditions. 3. Analytical solution of stress field:
[0224] Step a. Calculate the strain rate using the velocity field analytical solution:
[0225] Step b. Substitute into the constitutive equation (Bingham model):
[0226] When τ>τ0
[0227] Step c. Obtain the analytical solution of the stress field:
[0228]
[0229] Among them C n and ν n Determined by the analytical solution of the velocity field.
[0230] 4. Analytical solution of displacement field:
[0231] Step a. Time-integrate the velocity field analytical solution:
[0232]
[0233] Step b. Apply initial conditions:
[0234] Assume u(x,0)=0
[0235] Step c. Obtain the analytical solution of the displacement field:
[0236]
[0237] The calculation process of the numerical range:
[0238] 1. Discretization:
[0239] Choose appropriate spatial step sizes Δx, Δy, Δz and time step Δt.
[0240] 2. Generate grid points:
[0241] In the computational domain [0,L x ]×[0,L y ]×[0,L z ]×[0,T] to generate grid points:
[0242] x i =iΔx,i=0,1,...,N x
[0243] y j =jΔy,j=0,1,...,N y
[0244] z k =kΔz, k=0,1,...,N z
[0245] t l =lΔt,l=0,1,...,N t
[0246] 3. Calculate the value of the analytical solution:
[0247] For each grid point (x i ,y j ,z k , t l), calculate the corresponding field value.
[0248] For example, for the velocity field:
[0249]
[0250] Where N is the number of truncation terms.
[0251] 4. Statistical Analysis
[0252] Calculate the mean:
[0253]
[0254] Calculate the standard deviation:
[0255]
[0256] 5. Determine the value range:
[0257] According to the 3σ principle, the numerical range of the velocity field is:
[0258] 3σ v
[0259]
[0260] Repeat steps 3-5 for other fields (pressure field, stress field, displacement field).
[0261] It should be noted that the divergence operator
[0262] definition:
[0263] Function: Represents the divergence of a vector field, that is, the total outflow of the vector field at a certain point. For example, the continuity equation in It is the divergence of the product of density and velocity fields.
[0264] Laplace operator
[0265] definition:
[0266] Function: It represents the second-order derivative of a scalar field in space, reflecting the local curvature of the scalar field at a certain point. For example, in the Navier-Stokes equations, represents the viscous force term.
[0267] Time derivative
[0268] Function: Indicates the rate of change of a field quantity over time. For example, in the equation of motion, Represents the rate of change of the velocity field over time.
[0269] Inner Product
[0270] definition:
[0271] Function: It represents the convection term of the velocity field and reflects the nonlinear effect of the velocity field itself.
[0272] Outer Product
[0273] definition:
[0274] Function: It represents the curl of the velocity field and reflects the vortex characteristics of the velocity field.
[0275] Step S50: Establishing concrete vibration field matrix
[0276] The specific implementation of this step is as follows:
[0277] 1. Fourier Transform:
[0278] The Fast Fourier Transform (FFT) algorithm is used to decompose the amplitude and phase of each vibration frequency component from the vibration signals obtained at multiple points. These frequency domain characteristics reflect the characteristics of the movement of particles inside the concrete.
[0279] 2. Wavelet Analysis:
[0280] Wavelet transform technology is used to perform multi-scale time-frequency analysis on vibration signals. Wavelet coefficients at different scales contain information about the microscopic pore structure and local flow characteristics of concrete.
[0281] 3. Principal Component Analysis:
[0282] The principal component analysis (PCA) method is used to extract the main characteristic variables of the concrete vibration field from the Fourier spectrum and wavelet coefficients. The principal component scores can be used to represent the comprehensive characteristics of the vibration field at each measuring point.
[0283] 4. Integrate into vibration field matrix:
[0284] The characteristic quantities obtained from the above vibration signal analysis and the numerical ranges of the aforementioned velocity field, pressure field, stress field and displacement field are integrated into a concrete vibration field matrix.
[0285] Each element of the matrix is a eigenvector consisting of the vibrator's posture and operating parameters, which describes the fluidity of concrete under different vibration field conditions.
[0286] Through the integration of multiple signal analysis technologies, a matrix model that comprehensively reflects the internal vibration state of concrete is constructed, providing important input data for subsequent fluidity prediction.
[0287] Step S60: Predicting concrete fluidity using a pre-trained deep learning model
[0288] The specific implementation of this step is as follows:
[0289] 1. Micro-feature extraction network:
[0290] The network is used to capture fine-scale features in concrete, such as particle distribution, microscopic pore structure, and local flow characteristics. It includes:
[0291] (1) Multi-scale convolution layer: Convolution kernels with different receptive fields are processed in parallel to extract multi-scale micro features.
[0292] (2) Spatial pyramid pooling (SPP) layer: Divide the feature map into multiple regions, perform maximum pooling, and obtain multi-scale local features.
[0293] (3) Attention mechanism: including channel attention and spatial attention, adaptively focusing on important micro features.
[0294] (4) Residual connection: Residual blocks are used to deepen the network while maintaining gradient flow.
[0295] (5) Batch normalization: used after each convolutional layer to accelerate training and improve model stability.
[0296] 2. Macro feature extraction network:
[0297] The network is used to analyze large-scale characteristics of concrete, such as overall fluidity, temperature distribution, and pressure field. It includes:
[0298] (1) Deep convolutional network: ResNet-50 is used as the backbone network to extract powerful semantic features.
[0299] (2) Dilated convolution: Dilated convolution with different dilation rates is used at the back end of the network to expand the receptive field.
[0300] (3) Feature Pyramid Network (FPN): Fusing features from different depth layers to obtain multi-scale representation.
[0301] (4) Global average pooling: captures global context information.
[0302] (5) Self-attention mechanism: Establish long-range dependencies in feature graphs to enhance the expressiveness of features.
[0303] 3. Feature fusion network:
[0304] The network is used to integrate microscopic and macroscopic features to generate the final concrete flow prediction. It includes:
[0305] (1) Feature splicing layer: splicing micro and macro features in the channel dimension.
[0306] (2) 1x1 convolutional layer: reduces the number of feature channels and reduces computational complexity.
[0307] (3) Adaptive feature weighting: learning the importance weights of micro and macro features.
[0308] (4) Multi-layer Perceptron (MLP): performs nonlinear transformation and enhances feature expression.
[0309] (5) Skip connection: retain the original feature information and prevent information loss.
[0310] (6) Softmax classification layer: outputs the final liquidity prediction result.
[0311] During training, the macro feature extraction network was first initialized using a pretrained ResNet-50 model, and the entire model was then trained in stages. A hybrid loss function and the Adam optimizer were used, with learning rate decay. Performance was evaluated through cross-validation, and hyperparameters were tuned using Bayesian optimization. Finally, prediction accuracy was evaluated on the test set, and feature importance was analyzed using SHAP metrics.
[0312] This deep learning model can effectively integrate microscopic and macroscopic features, accurately predict the fluidity of concrete, and provide decision support for the optimization of vibrator parameters.
[0313] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for increasing concrete fluidity based on a vibrating rod.
[0314] A third aspect of the present invention provides a vibrator-based concrete fluidity increasing system, which includes the above-mentioned computer-readable storage medium.
[0315] Specifically, the principle of the present invention is:
[0316] First, this method fully utilizes classical fluid mechanics theory to establish a mathematical model describing the concrete flow process. Specifically, the method establishes an equation of motion based on the Navier-Stokes equations to describe the evolution of the concrete velocity field; a continuity equation based on the law of conservation of mass to describe the change in concrete density; and a constitutive equation using the Bingham viscoplastic model to describe the stress-strain relationship of concrete. Appropriate boundary and initial conditions are also provided, laying the foundation for the subsequent derivation of an analytical solution.
[0317] Secondly, this method uses mathematical analysis to derive analytical expressions for the velocity, pressure, stress, and displacement fields of concrete. These analytical solutions not only reflect the changing characteristics of concrete's microstructure, such as particle distribution and pore evolution, but also describe its macroscopic flow behavior, such as overall deformation and pressure distribution. These rich theoretical results provide an important basis for subsequent numerical calculations and parameter optimization.
[0318] Thirdly, the method employs a variety of signal processing techniques, such as Fourier transform, wavelet analysis, and principal component analysis, to construct a matrix model that comprehensively reflects the concrete vibration field state. This vibration field matrix incorporates microscopic and macroscopic features extracted from vibrator operating parameters, concrete surface deformation, and internal vibration signals, providing comprehensive input data for the subsequent deep learning model construction.
[0319] Finally, this method designed an innovative deep learning model architecture, comprising a micro-feature extraction network, a macro-feature extraction network, and a feature fusion network. This multi-scale, multi-modal feature fusion approach effectively exploits the complex relationships between concrete's internal microstructure and overall flow behavior, providing more accurate flow predictions. Furthermore, the model employs strategies such as staged training, cross-validation, and Bayesian optimization to further enhance its predictive performance and generalization capabilities.
[0320] In summary, the vibrator-based concrete fluidity increase method proposed in the present invention fully integrates fluid mechanics analysis and machine learning technology, and uses scientific modeling methods and intelligent optimization means to solve the limitations of existing technologies in concrete fluidity evaluation and parameter optimization, providing strong support for improving the quality and efficiency of concrete construction.
[0321] The following is an example of a specific application scenario of the present invention: A high-rise building project uses cast-in-place reinforced concrete. During foundation and wall construction, extensive use of vibrators is required to compact the poured concrete. To ensure the fluidity and compactness of the concrete, the construction company decided to optimize the vibration parameters using the vibrator-based concrete fluidity enhancement method proposed in this invention.
[0322] 1. Data Collection and Preprocessing
[0323] First, at the construction site, high-precision position sensors and accelerometers were installed on the vibrator rods, enabling real-time monitoring of the vibrator's three-dimensional position coordinates (x, y, z), tilt angles (pitch angle θ, yaw angle φ, roll angle ψ), and vibration accelerations (ax, ay, az). Simultaneously, a high-speed camera was placed on the concrete surface, capturing real-time deformation images of the concrete surface at a rate of 10 frames per second during the vibration process. Furthermore, three orthogonally arranged accelerometers were embedded in key locations within the concrete to record vibration acceleration signals along the three coordinate axes.
[0324] Through the collection of the above instruments and equipment, the following three types of basic data were obtained:
[0325] 1. Vibrator operating parameter data: including position coordinates (x, y, z), inclination angle (θ, φ, ψ), vibration frequency f, amplitude A and power P, etc.
[0326] 2. Concrete surface deformation image data: Concrete surface deformation image sequence at 10 frames / second taken by a high-speed camera.
[0327] 3. Concrete internal vibration signal data: vibration acceleration time series in the three coordinate axis directions (x, y, z) recorded by three orthogonal accelerometers.
[0328] For subsequent data analysis and model training, the collected raw data needs to be preprocessed. Specifically, it includes:
[0329] 1. Vibrator operating parameters:
[0330] -Perform time series smoothing filtering on the position coordinate and tilt angle data to remove high-frequency noise.
[0331] -Calculate parameters such as vibration frequency f, amplitude A and power P as characteristic quantities of the vibrator's operating state.
[0332] 2. Concrete surface deformation image data:
[0333] -Perform geometric correction on each frame of image to remove errors such as lens distortion.
[0334] -Using digital image correlation (DIC) technology, the displacement vector field of the concrete surface during the vibration process is extracted.
[0335] -Standardize the displacement vector field data to eliminate dimension effects.
[0336] 3. Concrete internal vibration signal data:
[0337] -Perform time-frequency analysis on the acceleration signals in the three coordinate axis directions and extract the amplitude and phase of the main frequency components.
[0338] -Use wavelet transform technology to further extract multi-scale time-frequency features.
[0339] -Principal component analysis (PCA) is used to reduce the dimensionality of the above frequency domain features and time-frequency features.
[0340] Through the above preprocessing, three types of data were obtained: vibrator operating parameters, concrete surface deformation and internal vibration, which laid the foundation for subsequent mechanical analysis and machine learning modeling.
[0341] 2. Mechanical Analysis and Numerical Calculation
[0342] According to steps S20 and S30 of the present invention, a mechanical model describing the concrete flow process is established, and analytical solutions for various physical field quantities are derived. Specifically, the following are the results:
[0343] 1. Equation of motion:
[0344] According to the Navier-Stokes equation, the motion equation describing the concrete flow is established:
[0345]
[0346] Among them, ρ is the density of concrete, v is the velocity vector field, t is time, p is the pressure field, μ is the dynamic viscosity, g is the gravity acceleration vector, and F is the external force generated by the vibrator.
[0347] 2. Continuity equation:
[0348] According to the law of conservation of mass, the continuity equation of concrete density change is established:
[0349]
[0350] 3. Constitutive equation:
[0351] The Bingham viscoplastic model is used to describe the stress-strain relationship of concrete:
[0352] When τ>τ0
[0353] When τ≤τ0
[0354] Where τ is the shear stress, τ0 is the yield stress, μ p is the plastic viscosity, is the shear rate.
[0355] 4. Boundary conditions:
[0356] On the surface of the vibrator, v = v0, where v0 is the velocity of the vibrator surface;
[0357] At the free surface of concrete, p = p0, where p0 is the atmospheric pressure;
[0358] At the external boundary, T = T0, where T0 is the ambient temperature.
[0359] 5. Initial conditions:
[0360] v(x,0)=v i (x),p(x,0)=p i (x),T(x,0)=T i (x)
[0361] Among them, v i (x),p i (x) and T i (x) are the initial velocity field, pressure field and temperature field distribution respectively.
[0362] On this basis, the above mechanical equations were solved analytically, and the analytical expressions of velocity field, pressure field, stress field and displacement field were obtained.
[0363] To calculate the numerical range of each field quantity, a spatial grid step size of Δx = Δy = Δz = 0.1 m and a time step size of Δt = 0.02 s were selected within the computational domain of [0,3 m] × [0,3 m] × [0,2 m] × [0,10 s]. A total of 3001 grid points were calculated. For each grid point, the aforementioned analytical solution was substituted to calculate the corresponding values of the velocity, pressure, stress, and displacement fields.
[0364] The statistical analysis results are as follows:
[0365] 1. Velocity field value range:
[0366] average value Standard deviation σ v =0.12m / s
[0367] According to the 3σ principle, the velocity field has a numerical range of v min =0.09m / s to v max =0.81m / s.
[0368] 2. Pressure field value range:
[0369] average value Standard deviation σ p =0.13MPa
[0370] The numerical range of the pressure field is p min =0.13MPa to p max =0.91MPa.
[0371] 3. Stress field value range:
[0372] average value Standard deviation σ τ =0.09MPa
[0373] The numerical range of the stress field is τ min =0.10MPa to τ max =0.64MPa.
[0374] 4. Displacement field value range:
[0375] average value Standard deviation σ u =0.42mm
[0376] The displacement field has a numerical range of u min =0.36mm to u max =2.88mm.
[0377] These numerical calculation results provide important reference for subsequent vibration field analysis and machine learning model construction.
[0378] 3. Vibration Field Analysis and Feature Extraction
[0379] According to step S50 of the present invention, a matrix model that comprehensively reflects the vibration state of concrete is constructed using techniques such as Fourier transform, wavelet analysis, and principal component analysis. The specific process is as follows:
[0380] 1. Fourier Transform:
[0381] The 9 collected acceleration sensor signals (3 coordinate axes x, y, z directions, a total of 3 sensors) were subjected to fast Fourier transform (FFT) to obtain the amplitude and phase of each frequency component:
[0382] A x =[A x,10Hz ,A x,20Hz ,…,A x,100Hz ]
[0383] φ x =[φ x,10Hz ,φ x,20Hz ,…,φ x,100Hz ]
[0384] A y ,φ y ,A z ,φ z
[0385] These frequency domain characteristics reflect the periodic characteristics of particle movement inside concrete.
[0386] 2. Wavelet Analysis:
[0387] Using Morlet wavelet, the acceleration signal is transformed into continuous wavelet transform (CWT), and the wavelet coefficients at different time-frequency scales are obtained:
[0388]
[0389] W y (a,b),W z (a,b)
[0390] Among them, a is the scale parameter, b is the time parameter, and ψ is the wavelet basis function.
[0391] These time-frequency characteristics can reflect the dynamic changes of concrete microstructure and local flow characteristics.
[0392] 3. Principal Component Analysis:
[0393] The above features extracted from the Fourier spectrum and wavelet coefficients are input into the PCA algorithm to obtain the principal component score:
[0394] s=[σ1,σ2…,σ p ]
[0395] Among them, σ i is the score of the i-th principal component, and p is the number of principal components.
[0396] These principal component scores comprehensively reflect the characteristics of the vibration field at each measuring point.
[0397] Finally, the vibrator operating parameters (f, A, P), concrete surface deformation data (u, v, w), and the principal component scores s of the vibration field are integrated into a concrete vibration field matrix, which serves as input for the subsequent machine learning model. This vibration field matrix comprehensively describes the microstructure and macroscopic flow characteristics of concrete, providing an important basis for accurately predicting concrete flow properties.
[0398] 4. Deep Learning Model Construction and Training
[0399] According to step S60 of the present invention, a multi-scale, multi-modal deep learning model is designed to predict the fluidity of concrete. The model includes three main network modules: a micro-feature extraction network, a macro-feature extraction network, and a feature fusion network.
[0400] 1. Micro-feature extraction network:
[0401] The network aims to capture microscopic features of concrete, such as the distribution of fine particles, microscopic pore structure, and local flow characteristics. It includes:
[0402] (1) Multi-scale convolution layer: 3x3, 5x5, and 7x7 convolution kernels are used to extract micro features of different receptive fields in parallel.
[0403] (2) Spatial pyramid pooling (SPP) layer: Divide the feature map into 1x1, 2x2, and 4x4 regions, perform maximum pooling, and obtain multi-scale local features.
[0404] (3) Channel attention mechanism and spatial attention mechanism: adaptively focus on micro-features that are important for liquidity prediction.
[0405] (4) Residual connection and batch normalization: enhance the network's expressiveness and training stability.
[0406] 2. Macro feature extraction network:
[0407] The purpose of this network is to analyze the macroscopic characteristics of concrete, such as fluidity, temperature distribution, and pressure field. It includes:
[0408] (1) ResNet-50 backbone network: extracts powerful semantic features.
[0409] (2) Dilated convolution: used at the back end of the network to expand the receptive field and capture a wider range of contextual information.
[0410] (3) Feature Pyramid Network (FPN): Fusing features from different depth layers to obtain multi-scale representation.
[0411] (4) Global average pooling and self-attention mechanism: establish long-range dependencies in feature maps.
[0412] 3. Feature fusion network:
[0413] The purpose of the network is to integrate microscopic and macroscopic features to generate the final concrete flow prediction results. It includes:
[0414] (1) Feature stitching: stitching micro and macro features in the channel dimension.
[0415] (2) 1x1 convolution: reduces the number of feature channels and reduces computational complexity.
[0416] (3) Adaptive feature weighting: learning the importance weights of micro and macro features.
[0417] (4) Multi-layer Perceptron (MLP): Enhances feature expression capabilities.
[0418] (5) Skip connection: retain the original feature information and prevent information loss.
[0419] (6) Softmax classification layer: outputs the final liquidity prediction result.
[0420] During model training, the macro-feature extraction network was first initialized using a pretrained ResNet-50 model, followed by phased training of the entire model. A hybrid loss function and the Adam optimizer were used, along with a learning rate decay strategy. Performance was evaluated through cross-validation, and hyperparameters were automatically tuned using Bayesian optimization. To improve the stability of prediction results, multiple models were trained and ensembled.
[0421] Finally, the model's prediction accuracy was evaluated on a test set. The results showed that the deep learning model achieved high accuracy in predicting concrete flowability, with a mean absolute error of less than 5%. SHAP analysis of feature importance was also used, revealing that the vibration frequency and amplitude of the vibrator, the normal displacement of the concrete surface, and the principal component scores of the internal vibration field were key factors influencing concrete flowability.
[0422] 5. Optimization and implementation of vibration parameters
[0423] Based on the prediction results of the above deep learning model, the operating parameters of the vibrator were optimized and adjusted according to the actual situation of the project site. The specific process is as follows:
[0424] 1. Determine the optimal vibration parameters:
[0425] Using the previously trained deep learning model, we predicted the fluidity of different vibrator operating parameter combinations (f, A, P). After multiple rounds of iterative optimization, we ultimately determined the following optimal vibrator parameters:
[0426] Vibration frequency f = 45 Hz
[0427] Vibration amplitude A = 25 mm
[0428] Vibration power P = 1.2kW
[0429] This set of parameters can maximize the fluidity index of concrete and meet the pouring quality requirements.
[0430] 2. On-site vibration construction:
[0431] Based on the determined optimal vibration parameters, operators precisely controlled the vibrator's position and set parameters during the actual pouring process. The vibrator's trajectory, vibration frequency, amplitude, and power were compared in real time with the pre-set optimized parameters, allowing for timely fine-tuning to ensure the desired vibration results.
[0432] At the same time, monitoring sensors are deployed on the concrete surface and within the concrete to collect deformation images and vibration signals in real time. These data are fed into a trained deep learning model to predict the concrete's fluidity in real time. If a decrease in fluidity is detected, operators are immediately notified to further optimize the vibration parameters.
[0433] 3.Quality acceptance:
[0434] During the curing period after pouring, the concrete structure underwent routine quality testing, including tests on strength, density, durability, and other indicators. The results showed that the concrete components vibrated and optimized using the method of the present invention met all quality requirements and achieved the expected construction results.
[0435] By implementing the three steps above, the proposed vibrator-based method for increasing concrete fluidity has been successfully applied to actual construction projects. This method leverages the advantages of fluid dynamics analysis and machine learning techniques to achieve precise modeling and intelligent optimization of the concrete vibration process, significantly improving the quality and efficiency of concrete construction.
[0436] Specifically, the technical advantages of this method are reflected in the following aspects:
[0437] 1. A more comprehensive concrete flow dynamics model was established, which not only took into account the vibrating rod parameters, but also integrated key factors such as surface deformation and internal vibration response, providing theoretical support for the precise optimization of the vibration process.
[0438] 2. A prediction model based on vibration field matrix was constructed, which can better capture the microstructure and macroscopic flow characteristics of concrete, and significantly improve the accuracy and reliability of fluidity prediction.
[0439] 3. The use of deep learning technology realizes the intelligent prediction and optimization of concrete fluidity, greatly simplifies the adjustment process of vibration parameters, and provides strong support for improving construction quality and efficiency in engineering practice.
[0440] 4. This method has good applicability and scalability. It is not only suitable for high-rise building projects, but can also be used in the construction process of various concrete structures, contributing to improving the overall level of the industry.
[0441] In general, the method for increasing concrete fluidity based on a vibrating rod proposed in the present invention fully integrates cutting-edge engineering technology, provides an innovative solution for the optimal control of concrete construction, and has important practical value in improving project quality and efficiency.
[0442] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.
Claims
1. A method for increasing concrete fluidity based on a vibrating rod, characterized in that: The following steps are involved: S10, acquiring the position and operating parameters of the vibrating rod, the surface image of the concrete, and vibration signals at multiple points, wherein the operating parameters include vibration frequency, amplitude, power, and action time; S20, establish a set of mechanical equations, including equations of motion, continuity equations, constitutive equations, boundary condition equations and initial condition equations; S30, analytically solving the mechanical equations to obtain a plurality of analytical solutions, which are a velocity field analytical solution for describing concrete flow, a pressure field analytical solution for describing pressure distribution, a stress field analytical solution for describing stress distribution, and a displacement field analytical solution for describing deformation; S40, performing numerical calculations on the multiple analytical solutions according to preset boundary conditions to obtain multiple numerical ranges, namely, a velocity field numerical range, a pressure field numerical range, a stress field numerical range, and a displacement field numerical range; S50, based on the multiple analytical solutions, multiple numerical ranges and vibration signals at multiple points, using a method combining Fourier transform, wavelet analysis and principal component analysis to establish a concrete vibration field matrix; S60. Input the concrete vibration field matrix and the surface image of the concrete into a pre-trained concrete fluidity prediction model to obtain a concrete vibration optimization matrix, wherein each element of the matrix is a vector representing the position and operating parameters of the vibrating rod; the concrete vibration optimization matrix is output to an operator user to adjust the position and operating parameters of the vibrating rod.
2. A method for increasing concrete fluidity based on a vibrating rod according to claim 1, characterized in that: The concrete fluidity prediction model includes a microscopic feature extraction network, a macroscopic feature extraction network and a feature fusion network; wherein the microscopic feature extraction network is used to capture the distribution of fine particles in concrete, the microscopic pore structure and the local flow characteristics; the macroscopic feature extraction network is used to analyze large-scale characteristics, including the overall fluidity of concrete, temperature distribution and pressure field; the feature fusion network is used to integrate microscopic and macroscopic features to generate the final concrete fluidity prediction.
3. The method for increasing concrete fluidity based on a vibrating rod according to claim 1, characterized in that: The motion equation is specifically expressed as: Among them: ρ is the density of concrete, obtained by experimental measurement; v is the velocity vector field; t is time; p is the pressure field; μ is the dynamic viscosity, measured by rheological experiments; g is the gravity acceleration vector; F is the external force field, including the force generated by the vibrator; is the gradient operator, which represents the rate of change of a scalar field in space and is defined as 4. A method for increasing concrete fluidity based on a vibrating rod according to claim 3, characterized in that: The continuity equation is specifically expressed as:
5. The method for increasing concrete fluidity based on a vibrating rod according to claim 4, characterized in that: The constitutive equation is specifically expressed as: When τ>τ0; When τ≤τ0; Where: τ is the shear stress; τ0 is the yield stress, which is determined by experiment; μ p is the plastic viscosity, determined experimentally; is the shear rate.
6. A method for increasing concrete fluidity based on a vibrating rod according to claim 5, characterized in that: The boundary condition equation is specifically expressed as: v = v0 on the surface of the vibrating rod; p = p0 on the free surface; T = T0 at the outer boundary; Where: v0 is the surface velocity of the vibrating rod, which is determined by the vibrating rod parameters; p0 is the atmospheric pressure; T0 is the ambient temperature.
7. A method for increasing concrete fluidity based on a vibrating rod according to claim 6, characterized in that: The initial condition equation is specifically expressed as: v(x,0)=v i (x); p(x,0)=p i (x); T(x,0)=T i (x); Where: v i (x) is the initial velocity field distribution; p i (x) is the initial pressure field distribution; T i (x) is the initial temperature field distribution.
8. The method for increasing concrete fluidity based on a vibrating rod according to claim 7, characterized in that: The concrete vibration optimization matrix is specifically expressed as: Where: M is the concrete vibration optimization matrix, each element m ij is a vector representing the position and operating parameters of the vibrator; m is the number of vibrators; n is the number of vibrator parameters that need to be optimized, usually including position coordinates (x, y, z), vibration frequency f, amplitude A, power P and action time t; m ij =[x i ,y i , z i , f j , A j , P j , t j ] is the jth optimized parameter vector of the i-th vibrator; (x i ,y i , z i ) is the position coordinate of the i-th vibrator, in m; f j is the jth optimized vibration frequency, in Hz; A j is the jth optimized vibration amplitude, in m; P j is the jth optimized vibration power, in W; t j is the jth optimized vibration action time, in seconds.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the method for increasing concrete fluidity based on a vibrating rod as described in any one of claims 1 to 8.
10. A concrete fluidity increasing system based on a vibrating rod, characterized in that: A computer-readable storage medium comprising the computer-readable storage medium of claim 9.
Citation Information
Cited By
Intelligent vibrating method and system for concrete
CN120563531A
Concrete vibration quality prediction method based on AWMI-Light GBM algorithm
CN120653954A
A concrete vibrating quality prediction method based on an AWMI-LightGBM algorithm
CN120653954B
Vibration degree judgment method based on concrete surface characteristics and related product
CN121073866A
Vibration quality online monitoring method and system based on multivariate perception
CN121612631A
