Data-driven intelligent stirring and quality prediction method

Abstract

The invention provides a data-driven intelligent stirring and quality prediction method, and is suitable for the field of stirring pile construction. According to the method, the data-driven neural network model is established, soil layer parameters, stirring parameters and construction quality are recognized, the state of a stirring machine is monitored in real time, the stirring process is dynamically adjusted, the comprehensive construction efficiency is improved, and the strength of a stirring pile can be predicted according to real-time data such as cement paste and the guniting amount by establishing the quality prediction model; and finally, intelligent construction is achieved, and the engineering quality and safety are improved.

Description

Technical Field

[0001] The present invention relates to a data-driven intelligent mixing and quality prediction method, which is applicable to the field of mixing pile construction. Background Art

[0002] Deep cement-soil mixing piles are a method of foundation treatment. They use cement as a solidifying agent. Using specialized mixing machinery, the solidifying agent is sprayed deep into the foundation to mix with the surrounding soil. This mixture then undergoes a physical and chemical reaction, hardening the soft soil into a high-quality foundation with integrity, water stability, and a certain strength. The construction of deep cement-soil mixing piles requires different process parameters tailored to different soil layers. In engineering practice, only a few survey boreholes are typically drilled within a construction area, and all construction parameters for deep cement-soil mixing piles are adjusted based on the soil layer distribution in these survey boreholes. However, many projects have complex soil layer distributions, and the actual soil layer distribution does not align with the survey boreholes. Consequently, construction parameters are often inappropriate for the specific soil layer, resulting in poor quality or even failure of some pile sections. Utilizing theoretical models of the mixing process to address the limited amount of survey borehole data and compensate for the large training data required for neural networks remains a challenge in intelligent applications. Summary of the Invention

[0003] The purpose of this invention is to propose a data-driven intelligent mixing and quality prediction method for steel bar pre-tightening, which is in urgent need of a scientific and reasonable method for pre-tightening steel bars by comprehensively utilizing modern monitoring technology and intelligent analysis means.

[0004] A data-driven intelligent mixing and quality prediction method comprises the following steps:

[0005] S101 establishes a neural network model driven by soil layer parameter data based on the test parameters of the mixing pile penetration process equipment. It trains the neural network by combining mixing dynamics theory with data-driven training to identify soil layer parameters.

[0006] S102 establishes a neural network model for mapping soil layer parameters and mixing parameters to identify mixing parameters;

[0007] S103 establishes a neural network model that considers the influence of soil layer parameters, mixing parameters, and quality parameters to identify the construction quality of mixing piles;

[0008] S104 reads the state parameters of the mixing equipment in real time, inputs them into the intelligent mixing neural network model, analyzes the cement slurry, water parameters, speed, etc. of the mixing equipment in real time, and controls the output of the mixing equipment to achieve intelligent mixing;

[0009] S105 obtains the actual cement slurry, water parameters, and speed parameters of the mixing machine in real time, inputs them into the quality prediction neural network model, and predicts the quality of the mixing pile.

[0010] Furthermore, in the above step S101, the step of establishing a neural network model driven by soil layer parameter data according to the test parameters of the equipment during the mixing pile penetration process is as follows:

[0011] a) establishing a dynamic model of the mixing tool's penetration process, wherein the dynamic model expressions are shown in equations (1) to (5);

[0012]

[0013] σ=γ w h w +∑γ i h i (5)

[0014] Where θ is the rotation angle of the machine, is the penetration acceleration, Z is the penetration displacement, G is the gravity of the mixer, F L is the top winch tension, which can be measured by the sensor, F n is the downward thrust component generated by the tilted blade, C u is the shear strength, c is the cohesion, is the internal friction angle, S t is the soil sensitivity, F d is the end resistance, F c is the side friction resistance, d is the drill pipe diameter, σ zi is the vertical effective stress of the overlying soil layer i; k si为 Lateral pressure coefficient of the i-th layer of soil, μ si is the friction coefficient between the i-th layer of soil and the drill rod, h i is the thickness of the i-th soil layer, k is the effective torque coefficient after considering the overlapping part; γ i is the density of the i-th layer of soil, h i is the thickness of the i-th soil layer, γ i is the density of water, h w is the depth of water, B1 is the width of the first blade, α1 is the inclination angle of the first blade, d is the diameter of the drill pipe, B2 is the width of other blades, α2 is the inclination angle of other blades, M1 is the moment required to overcome the shear strength of the cylindrical soil, M2 is the moment required to overcome the shear strength of the top and bottom surfaces, M is the total moment to overcome the shear strength of the soil, l is the blade length; B is the blade width, θ is the blade inclination angle, and J is the number of mixing shafts;

[0015] b) Constructing a neural network model driven by soil layer parameter data, wherein the input of the neural network model driven by soil layer parameter data is the tool torque, drill rod elevation, drill rod speed and drill rod rotation speed during the penetration process, and the output of the neural network model driven by soil layer parameter data is the shear strength C of each soil layer parameter. u, cohesion c, stress σ, internal friction angle and soil sensitivity S t , the neural network model driven by soil layer parameter data adopts a fully connected neural network structure, and n hidden layers are set according to the preset number of hidden layers and the number of neurons;

[0016] c) Perform dynamic data-driven solution.

[0017] Furthermore, the steps of performing dynamic data driven solution are:

[0018] a) constructing the i-th soil layer distance functional function of the mixing process according to the initialization, the formula of the i-th soil layer distance functional function is:

[0019]

[0020] F=Ψ(M,P) (7)

[0021] P = ∫F n -F d -F c dz (8)

[0022]

[0023] Where, α i , β i , γ i and η i is the preset parameter, is the constitutive function, F(C u ,S ti ) is a calculation function;

[0024] b) Derivative the distance functional function of the i-th soil layer to obtain the iterative formula as shown in formulas (10) to (15):

[0025]

[0026]

[0027] c) Using the formulas (10) to (15), C u,i To perform derivation, other Perform similar processing and eliminate the coefficient λ i , establish the iterative equations as (16) to (20),

[0028]

[0029] d) According to the iterative equation, the iterative formula for the i-th layer is established as formula (21) to (24)

[0030]

[0031] e) Set k = 0, randomly select any group of data near the current construction pile foundation from the soil layer database, and interpolate to obtain a group of states according to the current construction pile position and depth Soil layer parameters are used as virtual driving data;

[0032] f) Substitute into the iterative formula to calculate

[0033] g) Randomly calculate and interpolate data within any set distance near the current construction pile according to position and depth. Take the minimum difference as

[0034] h) Repeat steps f to g to obtain the optimal parameters;

[0035] i) Carry out the above steps for all soil layers to obtain the parameters c of all soil layers u ,c,σ, S t .

[0036] Furthermore, in the above S101, the step of training the neural network by combining stirring dynamics theory with data driving is:

[0037] a) Initialization of the training set. The training set contains data on the torque of the machine, drill rod elevation, drill rod speed, and drill rod rotation speed at all times during the construction of a certain number of piles, as well as soil parameter data for different depths of a certain number of exploration test holes in the construction area.

[0038] b) Establishing an optimization function, the expression of the optimization function is formula (25),

[0039]

[0040] Where N represents the total number of soil layers, the superscript p represents the neural network prediction value, and the superscript * represents the dynamic equation data driven calculation value. They represent the regularization parameters of the corresponding parameters, and a1~a5 are preset combination coefficients;

[0041] c) Establishing a neural network;

[0042] d) Establishing a parameter vector for all the calculation parameters of the neural network, wherein the expression of the parameter vector is formula (26),

[0043] x=[x1,...,x j ,..,x n] (26)

[0044] e) According to the preset parameters x j The range of variation and the set of data points are randomly selected as the initial data point matrix x s,j , n is the number of degrees of freedom, s = 1 ~ S, as the initial value, and input into the neural network as the initial parameters;

[0045] f) Start looping for each mixing pile penetration construction process data in the training set;

[0046] g) Run the data-driven solution of the dynamic equation to obtain all soil layer parameter data;

[0047] h) Run the neural network model, using the pile data as input, to obtain the parameters c of all soil layers u ,c,σ, S t result;

[0048] i) Calculate the sth and and its corresponding and is the minimum value of all optimization functions of the sth data point, is the minimum value of the optimization function of all S data points;

[0049] j) Calculate the following formula (27):

[0050]

[0051] Where ω is the weight, c1 and c2 are preset coefficients, and r1 and r2 are random numbers that can be different in calculation;

[0052] k) Calculate the parameter matrix update using equation (28);

[0053]

[0054] l) If the difference between the calculated optimization function MSE and the last calculated value is less than the preset value or ratio, or the number of calculations reaches the preset number of calculations, the calculation is terminated, otherwise the parameter The matrix is updated into the neural network, and the process jumps to step 7 to continue.

[0055] Furthermore, in the above step S105, the steps of establishing the quality prediction neural network model are:

[0056] a) Set the input of the quality prediction neural network model as soil layer parameters, cement slurry parameters, shotcrete volume, and water spray volume;

[0057] b) setting the output of the quality prediction neural network model to be the mixing pile strength;

[0058] c) configuring a fully connected neural network based on the preset number of hidden layers and neurons;

[0059] d) The training set of the neural network contains a certain number of cement slurry parameters, shotcrete volume, water spray volume, and corresponding mixing pile strength test results at all time points during the construction process of the mixing pile. The neural network is trained using the training set data and saved after completion.

[0060] The present invention has the following beneficial effects: by establishing a neural network model between soil layer parameters, mixing parameters and quality parameters, a deep understanding of soil layer characteristics during the construction process can be achieved, thereby improving the accuracy of mixing pile construction; based on real-time reading of mixer state parameters, the system can timely analyze cement slurry, water parameters, rotation speed, etc. during the mixing process, and intelligently control the mixing equipment to ensure construction quality; the use of a neural network that combines data drive and dynamics can effectively reduce manual intervention, improve the level of construction intelligence, and reduce the risks brought by differences in human operations; by establishing a quality prediction neural network model, the construction quality of the mixing pile can be accurately predicted based on real-time data parameters such as cement slurry and shotcrete volume, thereby ensuring the overall quality of the project. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 This is a flow chart of a data-driven intelligent stirring and quality prediction method of the present invention. DETAILED DESCRIPTION

[0062] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments given here are only used to illustrate and explain the present invention and cannot be used to limit the present invention.

[0063] The following is a specific embodiment of a data-driven intelligent mixing and quality prediction method.

[0064] S101 establishes a neural network model driven by soil layer parameter data based on the test parameters of the mixing pile penetration process equipment. It trains the neural network by combining mixing dynamics theory with data-driven training to identify soil layer parameters.

[0065] S102 establishes a neural network model for mapping soil layer parameters and mixing parameters to identify mixing parameters;

[0066] S103 establishes a neural network model that considers the influence of soil layer parameters, mixing parameters, and quality parameters to identify the construction quality of mixing piles;

[0067] S104 reads the state parameters of the mixing equipment in real time, inputs them into the intelligent mixing neural network model, analyzes the cement slurry, water parameters, speed, etc. of the mixing equipment in real time, and controls the output of the mixing equipment to achieve intelligent mixing;

[0068] S105 obtains the actual cement slurry, water parameters, and speed parameters of the mixing machine in real time, inputs them into the quality prediction neural network model, and predicts the quality of the mixing pile.

[0069] Furthermore, in the above step S101, the step of establishing a neural network model driven by soil layer parameter data according to the test parameters of the equipment during the mixing pile penetration process is as follows:

[0070] a) establishing a dynamic model of the mixing tool's penetration process, wherein the dynamic model expressions are shown in equations (1) to (5);

[0071]

[0072]

[0073] σ=γ w h w +∑γ i h i (5)

[0074] Where θ is the rotation angle of the machine, is the penetration acceleration, Z is the penetration displacement, G is the gravity of the mixer, F L is the top winch tension, which can be measured by the sensor, F n is the downward thrust component generated by the tilted blade, C u is the shear strength, c is the cohesion, is the internal friction angle, S t is the soil sensitivity, F d is the end resistance, F c is the side friction resistance, d is the drill pipe diameter, σ zi is the vertical effective stress of the overlying soil layer i; k si为 Lateral pressure coefficient of the i-th layer of soil, μ si is the friction coefficient between the i-th layer of soil and the drill rod, h i is the thickness of the i-th soil layer, k is the effective torque coefficient after considering the overlapping part; γ i is the density of the i-th layer of soil, h i is the thickness of the i-th soil layer, γ i is the density of water, h wis the depth of water, B1 is the width of the first blade, α1 is the inclination angle of the first blade, d is the diameter of the drill pipe, B2 is the width of other blades, α2 is the inclination angle of other blades, M1 is the moment required to overcome the shear strength of the cylindrical soil, M2 is the moment required to overcome the shear strength of the top and bottom surfaces, M is the total moment to overcome the shear strength of the soil, l is the blade length; B is the blade width, θ is the blade inclination angle, and J is the number of mixing shafts;

[0075] b) Constructing a neural network model driven by soil layer parameter data, wherein the input of the neural network model driven by soil layer parameter data is the tool torque, drill rod elevation, drill rod speed and drill rod rotation speed during the penetration process, and the output of the neural network model driven by soil layer parameter data is the shear strength C of each soil layer parameter. u , cohesion c, stress σ, internal friction angle and soil sensitivity S t , the neural network model driven by soil layer parameter data adopts a fully connected neural network structure, and n hidden layers are set according to the preset number of hidden layers and the number of neurons;

[0076] c) Perform dynamic data-driven solution.

[0077] Furthermore, the steps of performing dynamic data driven solution are:

[0078] a) constructing the i-th soil layer distance functional function of the mixing process according to the initialization, the formula of the i-th soil layer distance functional function is:

[0079]

[0080] F=Ψ(M,P) (7)

[0081] P = ∫F n -F d -F c dz (8)

[0082]

[0083] Where, α i , β i , γ i and η i is the preset parameter, is the constitutive function, F(C u ,S ti ) is a calculation function;

[0084] b) Derivative the distance functional function of the i-th soil layer to obtain the iterative formula as shown in formulas (10) to (15):

[0085]

[0086] c) Using the formulas (10) to (15), C u,i To perform derivation, other Perform similar processing and eliminate the coefficient λ i , establish the iterative equations as (16) to (20),

[0087]

[0088]

[0089] d) According to the iterative equation, the iterative formula for the i-th layer is established as formula (21) to (24)

[0090]

[0091] e) Set k = 0, randomly select any group of data near the current construction pile foundation from the soil layer database, and interpolate to obtain a group of states according to the current construction pile position and depth Soil layer parameters are used as virtual driving data;

[0092] f) Substitute into the iterative formula to calculate

[0093] g) Randomly calculate and interpolate data within any set distance near the current construction pile according to position and depth. Take the minimum difference as

[0094] h) Repeat steps f to g to obtain the optimal parameters;

[0095] i) Carry out the above steps for all soil layers to obtain the parameters c of all soil layers u ,c,σ, S t .

[0096] Furthermore, in the above S101, the step of training the neural network by combining stirring dynamics theory with data driving is:

[0097] a) Initialization of the training set. The training set contains data on the torque of the machine, drill rod elevation, drill rod speed, and drill rod rotation speed at all times during the construction of a certain number of piles, as well as soil parameter data for different depths of a certain number of exploration test holes in the construction area.

[0098] b) Establishing an optimization function, the expression of the optimization function is formula (25),

[0099]

[0100] Where N represents the total number of soil layers, the superscript p represents the neural network prediction value, and the superscript * represents the dynamic equation data driven calculation value. They represent the regularization parameters of the corresponding parameters, and a1~a5 are preset combination coefficients;

[0101] c) Establishing a neural network;

[0102] d) Establishing a parameter vector for all the calculation parameters of the neural network, wherein the expression of the parameter vector is formula (26),

[0103] x=[x1,...,x j ,..,x n ] (26)

[0104] e) According to the preset parameters x j The range of variation and the set of data points are randomly selected as the initial data point matrix x s,j , n is the number of degrees of freedom, s = 1 ~ S, as the initial value, and input into the neural network as the initial parameters;

[0105] f) Start looping for each mixing pile penetration construction process data in the training set;

[0106] g) Run the data-driven solution of the dynamic equation to obtain all soil layer parameter data;

[0107] h) Run the neural network model, using the pile data as input, to obtain the parameters c of all soil layers u ,c,σ, S t result;

[0108] i) Calculate the sth and and its corresponding and is the minimum value of all optimization functions of the sth data point, is the minimum value of the optimization function of all S data points;

[0109] j) Calculate the following formula (27):

[0110]

[0111] Where ω is the weight, c1 and c2 are preset coefficients, and r1 and r2 are random numbers that can be different in calculation;

[0112] k) Calculate the parameter matrix update using equation (28);

[0113]

[0114] l) If the difference between the calculated optimization function MSE and the last calculated value is less than the preset value or ratio, or the number of calculations reaches the preset number of calculations, the calculation is terminated, otherwise the parameter The matrix is updated into the neural network, and the process jumps to step 7 to continue.

[0115] Furthermore, in the above step S105, the steps of establishing the quality prediction neural network model are:

[0116] a) Set the input of the quality prediction neural network model as soil layer parameters, cement slurry parameters, shotcrete volume, and water spray volume;

[0117] b) setting the output of the quality prediction neural network model to be the mixing pile strength;

[0118] c) configuring a fully connected neural network based on the preset number of hidden layers and neurons;

[0119] d) The training set of the neural network contains a certain number of cement slurry parameters, shotcrete volume, water spray volume, and corresponding mixing pile strength test results at all time points during the construction process of the mixing pile. The neural network is trained using the training set data and saved after completion.

[0120] In the above embodiment, the present invention discloses a data-driven intelligent mixing and quality prediction method, which is used for soil layer parameter identification and quality prediction during the intelligent mixing pile construction process, improves construction accuracy and quality control, realizes real-time monitoring and dynamic optimization of the construction process, reduces the risks brought by human factors, and can be widely used in the field of mixing pile construction.

[0121] The above description is a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A data-driven intelligent stirring and quality prediction method, characterized in that: The following steps are involved: S101 establishes a neural network model driven by soil layer parameter data based on the test parameters of the mixing pile penetration process equipment. It trains the neural network by combining mixing dynamics theory with data-driven training to identify soil layer parameters. S102 establishes a neural network model for mapping soil layer parameters and mixing parameters to identify mixing parameters; S103 establishes a neural network model that considers the influence of soil layer parameters, mixing parameters, and quality parameters to identify the construction quality of mixing piles; S104 reads the state parameters of the mixing equipment in real time, inputs them into the intelligent mixing neural network model, analyzes the cement slurry, water parameters, speed, etc. of the mixing equipment in real time, and controls the output of the mixing equipment to achieve intelligent mixing; S105 obtains the actual cement slurry, water parameters, and speed parameters of the mixing machine in real time, inputs them into the quality prediction neural network model, and predicts the quality of the mixing pile.

2. A data-driven intelligent stirring and quality prediction method according to claim 1, characterized in that: In step S101, the steps of establishing a neural network model driven by soil layer parameter data according to the test parameters of the equipment during the mixing pile penetration process are as follows: a) establishing a dynamic model of the mixing tool's penetration process, wherein the dynamic model expressions are shown in equations (1) to (5); σ=γ w h w +∑γ i h i (5) Where θ is the rotation angle of the machine, is the penetration acceleration, Z is the penetration displacement, G is the gravity of the mixer, F L is the top winch tension, which can be measured by the sensor, F n is the downward thrust component generated by the tilted blade, C u is the shear strength, c is the cohesion, is the internal friction angle, S t is the soil sensitivity, F d is the end resistance, F c is the side friction resistance, d is the drill pipe diameter, σ zi is the vertical effective stress of the overlying soil layer i; k si为 Lateral pressure coefficient of the i-th layer of soil, μ si is the friction coefficient between the i-th layer of soil and the drill rod, h i is the thickness of the i-th soil layer, k is the effective torque coefficient after considering the overlapping part; γ i is the density of the i-th layer of soil, h i is the thickness of the i-th soil layer, γ i is the density of water, h w is the depth of water, B1 is the width of the first blade, α1 is the inclination angle of the first blade, d is the diameter of the drill pipe, B2 is the width of other blades, α2 is the inclination angle of other blades, M1 is the moment required to overcome the shear strength of the cylindrical soil, M2 is the moment required to overcome the shear strength of the top and bottom surfaces, M is the total moment to overcome the shear strength of the soil, l is the blade length; B is the blade width, θ is the blade inclination angle, and J is the number of mixing shafts; b) Constructing a neural network model driven by soil layer parameter data, wherein the input of the neural network model driven by soil layer parameter data is the tool torque, drill rod elevation, drill rod speed and drill rod rotation speed during the penetration process, and the output of the neural network model driven by soil layer parameter data is the shear strength C of each soil layer parameter. u , cohesion c, stress σ, internal friction angle and soil sensitivity S t , the neural network model driven by soil layer parameter data adopts a fully connected neural network structure, and n hidden layers are set according to the preset number of hidden layers and the number of neurons; c) Perform dynamic data-driven solution.

3. A data-driven intelligent stirring and quality prediction method according to claim 2, characterized in that: The steps of performing dynamic data driven solution are: a) Constructing the distance functional function of the i-th soil layer in the mixing process according to the initialization, the formula of the distance functional function of the i-th soil layer is as follows: F=Ψ(M,P) (7) P=∫F n -F d -F c dz (8) Where, α i , β i , γ i and η i is the preset parameter, is the constitutive function, F(C u ,S ti ) is a calculation function; b) Derivative the distance functional function of the i-th soil layer to obtain the iterative formula as shown in formulas (10) to (15): c) Using the formulas (10) to (15), C u,i To perform derivation, other Perform similar processing and eliminate the coefficient λ i , establish the iterative equations as (16) to (20), d) According to the iterative equation, the iterative formula for the i-th layer is established as formula (21) to (24) e) Set k = 0, randomly select any group of data near the current construction pile foundation from the soil layer database, and interpolate to obtain a group of states according to the current construction pile position and depth Soil layer parameters are used as virtual driving data; f) Substitute into the iterative formula to calculate and get g) Randomly calculate and interpolate data within any set distance near the current construction pile according to position and depth. Take the minimum difference as h) Repeat steps f to g to obtain the optimal parameters; i) Carry out the above steps for all soil layers to obtain the parameters c of all soil layers u ,c,σ, S t .

4. A data-driven intelligent stirring and quality prediction method according to claim 1, characterized in that: In S101, the step of performing neural network training by combining stirring dynamics theory with data driving is as follows: a) Initialization of the training set. The training set contains data on the torque of the machine, drill rod elevation, drill rod speed, and drill rod rotation speed at all times during the construction of a certain number of piles, as well as soil parameter data for different depths of a certain number of exploration test holes in the construction area. b) Establishing an optimization function, the expression of the optimization function is formula (25), Where N represents the total number of soil layers, the superscript p represents the neural network prediction value, and the superscript * represents the dynamic equation data driven calculation value. They represent the regularization parameters of the corresponding parameters, and a1~a5 are preset combination coefficients; c) Establishing a neural network; d) Establishing a parameter vector for all the calculation parameters of the neural network, wherein the expression of the parameter vector is formula (26), x=[x1,...,x j ,...,x n ] (26) e) According to the preset parameters x j The range of variation and the set of data points are randomly selected as the initial data point matrix x s,j , n is the number of degrees of freedom, s = 1 ~ S, as the initial value, and input into the neural network as the initial parameters; f) Start looping for each mixing pile penetration construction process data in the training set; g) Run the data-driven solution of the dynamic equation to obtain all soil layer parameter data; h) Run the neural network model, using the pile data as input, to obtain the parameters c of all soil layers u ,c,σ, S t result; i) Calculate the sth and and its corresponding and is the minimum value of all optimization functions of the sth data point, is the minimum value of the optimization function of all S data points; j) Calculate the following formula (27): Where ω is the weight, c1 and c2 are preset coefficients, and r1 and r2 are random numbers that can be different in calculation; k) Calculate the parameter matrix update using equation (28); l) If the difference between the calculated optimization function MSE and the last calculated value is less than the preset value or ratio, or the number of calculations reaches the preset number of calculations, the calculation is terminated, otherwise the parameter The matrix is updated into the neural network, and the process jumps to step 7 to continue.

5. A data-driven intelligent stirring and quality prediction method according to claim 1, characterized in that: In S105, the steps of establishing the quality prediction neural network model are: a) Set the input of the quality prediction neural network model as soil layer parameters, cement slurry parameters, shotcrete volume, and water spray volume; b) setting the output of the quality prediction neural network model to be the mixing pile strength; c) configuring a fully connected neural network based on the preset number of hidden layers and neurons; d) The training set of the neural network contains a certain number of cement slurry parameters, shotcrete volume, water spray volume, and corresponding mixing pile strength test results at all time points during the construction process of the mixing pile. The neural network is trained using the training set data and saved after completion.

Images