Machine learning based design method for concrete formwork support systems

Abstract

The application discloses a design method of a concrete formwork support system based on machine learning, and relates to the technical field of civil engineering construction. The application obtains three-dimensional cavity topology data of a target formwork system, a concrete pumping pressure sequence and non-Newtonian fluid rheological parameters, inputs the same into a generator in a pre-trained generative adversarial network to output an extreme flow field topology structure; according to this, a local uplift pressure vector and a parallel shear stress vector are extracted, and then input into a discriminator to calculate an asymmetric support deployment scheme containing anti-pull embedded parts and anti-shear inclined rod coordinate parameters; finally, the application verifies and outputs target drawings based on preset safety check rules. The application breaks the prejudice of traditional static compression design, gives the support system dynamic defense capability against floating and shearing, effectively solves the formwork instability problem caused by high-pressure pumping fluid mechanics damage in a special-shaped limited space, and guarantees the compliance of the project.

Description

Technical Field

[0001] This invention relates to the field of civil engineering construction technology, and in particular to a design method for a concrete formwork support system based on machine learning. Background Technology

[0002] In the construction of modern irregular thin-shell buildings (such as large-span curved domes, underwater immersed tunnel linings, etc.), it is usually necessary to use high-pressure pumping to inject high-viscosity self-compacting concrete into the closed or semi-closed formwork cavity.

[0003] Traditional concrete formwork support system design methods are generally based on static gravity mechanics. Existing technologies typically treat the overall weight of the poured concrete as the sole downward uniformly distributed load, thus designing a horizontally and vertically aligned, perfectly symmetrical support grid system. However, when concrete is pumped under high pressure within a confined, irregularly shaped mold cavity, the concrete actually exhibits complex non-Newtonian fluid rheological properties. The high-pressure pumped fluid flows, compresses, and flows back within the narrow mold cavity, generating strong tangential frictional forces (parallel shear stress) along the inner wall of the formwork, and producing enormous vertical lifting forces (local uplift pressure) at local corners or the top plate. Traditional purely compressive symmetrical grid structures, lacking node designs to resist pull (tension) and lateral shear, are highly susceptible to localized bursting or overall floating failure when facing such dynamic fluid loads.

[0004] Furthermore, although some existing technologies employ deep learning for structural topology optimization, the irregular support skeletons generated by these algorithms are often purely mathematical probabilistic fits, lacking the constraints of physical and mechanical conservation laws, and failing to be linked to the mandatory bottom-line requirements of national building construction safety standards. This leads to the predicament of AI-generated solutions failing to pass compliance reviews and posing potential mechanical hazards in the highly conservative field of building construction. Summary of the Invention

[0005] This invention aims to overcome the shortcomings of existing traditional support systems, which are based solely on static gravity design and cannot withstand the hydrodynamic damage caused by high-pressure pumping. It provides a machine learning-based design method for concrete formwork support systems. The core technical problem to be solved is: how to accurately capture the dynamic lateral shear force and uplift pressure generated by high-pressure pumped concrete within a confined space, and based on this, intelligently generate asymmetric pull-out and shear support topology drawings that balance physical and mechanical equilibrium with engineering code compliance.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: a design method for a concrete formwork support system based on machine learning, executed by a processor, comprising the following steps:

[0007] Acquire the three-dimensional internal cavity topology data, concrete pumping pressure sequence, and non-Newtonian fluid rheological parameters of the target template system;

[0008] The three-dimensional internal cavity topology data, the concrete pumping pressure sequence, and the non-Newtonian fluid rheological parameters are input into the generator in a pre-trained generative adversarial network to output the extreme flow field topology.

[0009] Based on the extreme flow field topology, the local uplift pressure vector and parallel shear stress vector are extracted;

[0010] The local uplift pressure vector and the parallel shear stress vector are input into the discriminator in the generative adversarial network to calculate and generate an asymmetric support deployment scheme that includes the coordinate parameters of the anti-uplift embedded part and the coordinate parameters of the anti-shear inclined member;

[0011] The asymmetric support deployment scheme is verified based on preset security verification rules, and the target support system drawing is output after the verification is passed.

[0012] Further, the step of inputting the three-dimensional internal cavity topology data, the concrete pumping pressure sequence, and the non-Newtonian fluid rheological parameters into the generator in a pre-trained generative adversarial network includes:

[0013] A discretized mesh model is constructed based on the three-dimensional internal cavity topology data;

[0014] Based on the discretized grid model, a physical partial differential loss function is introduced into the generator, combined with an inline formula. Calculate the momentum conservation residual, where For density, For the velocity field, For pressure field, Dynamic viscosity;

[0015] The network weights are updated iteratively using the momentum conservation residuals to output the extreme flow field topology that satisfies the mechanism constraints.

[0016] Furthermore, the steps of extracting the local uplift pressure vector and parallel shear stress vector based on the extreme flow field topology include:

[0017] Extract the normal momentum tensor of the extreme flow field topology at the boundary grid, and convert the normal momentum tensor into the upward local uplift pressure vector;

[0018] Based on the Herschel-Barclay rheological model and inline formulas Calculate the wall friction stress matrix, where For yield stress, This is the consistency coefficient. Shear rate, The rheological index;

[0019] The wall friction stress matrix is ​​converted into the parallel shear stress vector.

[0020] Further, the step of inputting the local uplift pressure vector and the parallel shear stress vector into the discriminator in the generative adversarial network to calculate and generate an asymmetric support deployment scheme including the coordinate parameters of the uplift-resistant embedded part and the coordinate parameters of the shear-resistant inclined member, and generating the coordinate parameters of the uplift-resistant embedded part, includes:

[0021] Initialize the initial support topology matrix with orthogonal structure;

[0022] The local uplift pressure vector is mapped to the loaded node of the initial support topology matrix, and the pull-out bearing capacity difference matrix is ​​calculated in combination with the preset ultimate bearing capacity of the embedded parts.

[0023] Extract the target tension nodes with values ​​greater than zero from the tensile bearing capacity difference matrix, and generate the coordinate parameters of the tensile pre-embedded part at the target tension node.

[0024] Furthermore, the step of calculating and generating an asymmetric support deployment scheme further includes:

[0025] The parallel shear stress vector is mapped to the initial support topology matrix, and the lateral displacement tendency component of each node is calculated.

[0026] Identify the target lateral displacement tendency component that exceeds the stiffness deformation threshold, and generate a spatial connection line vector in the adjacent spatial element of the target lateral displacement node.

[0027] The coordinate parameters of the anti-shear inclined member are generated by matching and replacing the spatial connection line vector, and combined with the coordinate parameters of the anti-pull-out embedded part to form the asymmetric support deployment scheme.

[0028] Furthermore, the step of verifying the asymmetric support deployment scheme based on preset security verification rules includes an underlying physical verification process:

[0029] A structural stiffness matrix is ​​constructed based on the asymmetric support deployment scheme, and the local uplift pressure vector and the parallel shear stress vector are combined into an external load column vector.

[0030] Based on inline formula Perform finite element analysis to calculate the deformation displacement field matrix, where Here is the structural stiffness matrix. Let U be the column vector of external loads, and U be the deformation displacement field matrix.

[0031] When the maximum shear strain value of the deformation displacement field matrix is ​​less than a preset limit threshold, a physical verification pass command is generated.

[0032] Furthermore, the verification steps also include a top-level compliance verification process:

[0033] In response to the physical verification pass instruction, the associated local high-risk node stress peak matrix is ​​extracted;

[0034] The actual safety factor matrix is ​​calculated by dividing the preset material ultimate yield strength by each element of the stress peak matrix of the local high-risk nodes.

[0035] When the smallest element of the actual safety factor matrix is ​​greater than the minimum data of the building code, a compliance verification pass instruction is generated.

[0036] Furthermore, the concrete pumping pressure sequence carrying timestamp tag data also includes:

[0037] Locate the local uplift pressure peak coordinates corresponding to each of the timestamp tag data in the extreme flow field topology;

[0038] Based on inline formula Calculate and generate a three-dimensional fluid wavefront propagation time series table, in which For the characteristics of wave propagation speed, S t Let S be the position of the wavefront at time t, and S0 be the initial point;

[0039] The three-dimensional fluid wavefront propagation timing table is mapped to the asymmetric support deployment scheme, and a timing reinforcement early warning file is output.

[0040] Furthermore, the steps for obtaining the rheological parameters of non-Newtonian fluids include:

[0041] Receive the temperature matrix and dynamic slump extension data of the mixture at the target construction site;

[0042] The temperature matrix of the mixture and the dynamic slump expansion data are input into a preset nonlinear rheological support vector machine model;

[0043] The radial basis function is used to perform mapping calculations, and the plastic viscosity property value and yield stress property value are output to be combined into the rheological parameters of the non-Newtonian fluid.

[0044] Furthermore, the steps for outputting the target support system drawings after successful verification include:

[0045] The asymmetric support deployment scheme is analyzed to extract the features of the three-dimensional spatial coordinate scatter point set;

[0046] The features of the three-dimensional spatial coordinate scatter point set are replaced by the mapping of the solid mesh model to generate a visual design model;

[0047] The visualization design model is subjected to orthographic projection processing to generate a two-dimensional target support system drawing with parametric annotations.

[0048] The beneficial effects of this invention are as follows:

[0049] 1. This invention introduces three-dimensional internal cavity topology data, concrete pumping pressure sequence and non-Newtonian fluid rheological parameters, and uses the generator in the generative adversarial network to perform fluid dynamics game calculations. It can accurately capture the extremely hidden local uplift pressure vector and parallel shear stress vector inside the confined mold cavity, and identify the real fluid dynamic root cause of the failure of complex molds from the source.

[0050] 2. This invention utilizes a discriminator to receive the destructive force vector. This method can automatically find weak nodes and generate an asymmetric support deployment scheme containing the coordinate parameters of the anti-pull-out embedded parts and the coordinate parameters of the anti-shear oblique rods in high-risk stress areas. This greatly reduces the risk of formwork rupture and saves material piling in unnecessary areas.

[0051] 3. Before outputting the final drawings, this invention forces the asymmetric support deployment scheme to be verified based on security verification rules, ensuring that the defense line calculated by artificial intelligence not only holds true in the adversarial game network, but also absolutely meets the bottom line restrictions of real physical laws and the mandatory regulations of the construction industry, thus clearing the compliance obstacles of the new design method in the engineering drawing review process. Attached Figure Description

[0052] Figure 1 A flowchart illustrating the design method of a machine learning-based concrete formwork support system provided in an embodiment of the present invention. Detailed Implementation

[0053] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. Identical components are indicated by the same reference numerals.

[0054] It should be noted that the terms front, back, left, right, up, and down used in the following description refer to the directions in the attached diagram, while the terms inside and outside refer to the directions toward or away from the geometric center of a specific component.

[0055] To make the content of this invention easier to understand, the technical solutions in the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings.

[0056] like Figure 1As shown, this application provides a design method for a concrete formwork support system based on machine learning, taking the asymmetric thin-shell lining pouring project of a large underwater curved immersed tunnel as an example. This project requires injecting high-viscosity self-compacting concrete into a closed and curved formwork cavity by high-pressure pumping.

[0057] Step S1: Obtain the three-dimensional internal cavity topology data, concrete pumping pressure sequence, and non-Newtonian fluid rheological parameters of the target template system.

[0058] The purpose of this step is to establish the initial digital environment and physical boundary conditions for the entire design methodology. The three-dimensional internal cavity topology data of the target template system refers to the set of three-dimensional geometric coordinates reflecting the shape and dimensions of the internal space of the mold cavity, typically obtained by reading Building Information Modeling (BIM). The concrete pumping pressure sequence refers to the dynamic thrust data applied to the concrete by construction machinery over time, which can be obtained in real-time through the pressure sensor array at the front end of the pumping equipment or by consulting construction organization design documents. Non-Newtonian fluid rheological parameters refer to physical quantities reflecting the properties of concrete that distinguish it from ordinary Newtonian fluids such as water (e.g., the force required to initiate flow, the viscosity during flow, etc.). The principle behind obtaining these three types of data is that the forces within the confined space are no longer static gravity, but a dynamic system determined by the spatial shape, power source, and fluid material.

[0059] In this embodiment, in the immersed tunnel project, the processor first reads the BIM model of the curved tunnel lining (obtaining three-dimensional internal cavity topology data), then obtains the 15 MPa stepped pressurization plan set by the ground pump (obtaining the concrete pumping pressure sequence), and obtains the initial yield stress and viscosity values ​​of the batch of self-compacting concrete through laboratory testing (obtaining non-Newtonian fluid rheological parameters).

[0060] Step S2: Input the three-dimensional internal cavity topology data, concrete pumping pressure sequence and non-Newtonian fluid rheological parameters into the generator in the pre-trained generative adversarial network, and output the extreme flow field topology structure.

[0061] Explanation: Traditional methods cannot predict how fluids will irregularly flow within complex spaces. This step addresses this gap by introducing a generator from Generative Adversarial Networks (GANs). The generator acts like an attacker searching for system vulnerabilities. It receives the 3D internal cavity topology data, concrete pump pressure sequence, and non-Newtonian fluid rheological parameters from previous steps, and then simulates fluid propulsion in a virtual space. The principle is to use deep learning algorithms to continuously attempt to find the fluid motion state that will cause the greatest damage to the template. The final output extreme flow field topology is a 3D spatial state field containing the extreme distributions of velocity and pressure. The technical effect of this step is to accurately locate hidden stress hazard zones that traditional experience cannot cover.

[0062] Specifically, this step includes the following sub-steps:

[0063] Step S2.1: Construct a discretized mesh model based on the three-dimensional internal cavity topology data.

[0064] Explanation: This process transforms a continuous physical space into a computer-processable digital matrix. Based on the three-dimensional internal cavity topology data, it is divided into countless tiny geometric units, forming a discretized mesh model. The principle is the finite volume method, providing spatial computing nodes for subsequent fluid dynamics calculations. In this embodiment, the processor divides the smooth 3D tunnel model into 1 million tiny cube meshes with sides of 5 cm, forming the discretized mesh model.

[0065] Step S2.2: Based on the discretized mesh model, introduce a physical partial differential loss function into the generator, combined with the inline formula. Calculate the momentum conservation residual, where For density, For the velocity field, For pressure field, This refers to dynamic viscosity.

[0066] Traditional neural networks are black boxes, easily generating data that violates physical laws. This step forcibly implants the Navier-Stokes equations into the generator as a physical partial differential loss function. In the formula, the part within parentheses represents the inertial force and convective acceleration of the fluid. Represents the pressure gradient. This represents viscous drag. The calculated momentum conservation residual represents the degree of deviation between the flow field generated by current artificial intelligence and the laws of reality. The technical effect is to endow deep learning models with physical interpretability. In this embodiment, when calculating the flow of concrete within the grid, the processor substitutes the concrete density. and viscosity Using this formula, it was found that the flow velocity generated by the grid at a certain point violated the law of energy conservation, resulting in a high residual value of momentum conservation.

[0067] Step S2.3: Iteratively update the network weights using momentum conservation residuals to output the extreme flow field topology that satisfies the mechanism constraints.

[0068] The purpose of this sub-step is to correct errors in the artificial intelligence. Using the momentum conservation residuals calculated in the previous step, the generator's internal parameters are continuously corrected through the backpropagation algorithm, forcing the residuals to approach zero. When the residuals are extremely small, the output extreme flow field topology is a true prediction that perfectly conforms to the laws of fluid dynamics.

[0069] In this embodiment, the processor continuously fine-tunes the generator parameters based on the high residual value to re-simulate until the residual is close to 0. At this point, the output extreme flow field topology structure containing the real high-pressure accumulation zone at the tunnel corner is confirmed to be valid.

[0070] In this embodiment, the generator combines the tunnel model, 15 MPa pressure, and concrete viscosity. Through simulation calculations, it is found that when fluid rushes into the top corner of the tunnel, violent fluid compression and backflow will occur. Therefore, it outputs an extreme flow field topology structure that includes the abnormal flow velocity and pressure distribution at the corner.

[0071] Step S3: Extract the local uplift pressure vector and parallel shear stress vector based on the extreme flow field topology.

[0072] The purpose of this step is to transform the complex fluid dynamics state into mechanical indices usable for structural design. Based on the extreme flow field topology output in the previous step, the fluid exerts destructive forces on the template in two directions within the confined space: the local uplift pressure vector refers to the force that vertically lifts the template upwards (attempting to make the template float or burst); the parallel shear stress vector refers to the force that, due to viscous friction, the fluid adheres tightly to and pulls parallel to the inner wall of the template during flow (attempting to cause lateral slippage and deformation of the template). The principle of extracting these two sets of vectors is to perform vector decomposition of the flow field pressure field and velocity gradient field. The technical effect is to concretize the abstract fluid topology into engineering mechanical vectors.

[0073] Specifically, this step includes the following sub-steps:

[0074] Step S3.1: Extract the normal momentum tensor of the extreme flow field topology at the boundary grid, and convert the normal momentum tensor into an upward local uplift pressure vector.

[0075] This step further explains the source of the uplift pressure calculation. Connecting to the previous step, after obtaining the extreme flow field topology, we only focus on the fluid in direct contact with the template inner wall (i.e., the boundary mesh). The normal momentum tensor refers to the energy matrix generated by the fluid's vertical impact on the template wall. Through area integration, it is equivalently transformed into a vertically upward force vector (i.e., the local uplift pressure vector). In this embodiment, the processor extracts the normal momentum tensor of the fluid's vertical upward impact at the template mesh at the top of the tunnel, and after calculation, transforms it into a definite 50 kN local uplift pressure vector.

[0076] Step S3.2: Based on the Herschel-Barclay rheological model and inline formulas Calculate the wall friction stress matrix, where For yield stress, This is the consistency coefficient. For shear rate, The rheological index is used; Explanation: This step reveals the calculation principle of non-Newtonian fluid friction. Unlike water, concrete exhibits strong nonlinear characteristics when sliding in pipes. The Herschel-Bulkley rheological model is introduced to calculate the wall friction stress matrix. In the formula, a value greater than [value missing] must be applied. Only when the concrete is subjected to force will it begin to flow; after flowing, its frictional force depends on the current shear rate. and its consistency and rheological index The technical effect of this step is to accurately reproduce the wall-adhesion tensile force unique to high-viscosity materials.

[0077] In this embodiment, the processor inputs the concrete yield stress measured in the laboratory. and consistency Combined with the shear rate at the tunnel sidewall The formula is used to calculate the wall friction stress matrix generated when concrete flows over the sidewall.

[0078] Step S3.3: Convert the wall friction stress matrix into a parallel shear stress vector.

[0079] Explanation: The planar stress distribution (wall friction stress matrix) obtained in the previous step is vector summed along the spatial direction to obtain the concentrated linear force acting on the nodes of the support system, i.e., the parallel shear stress vector.

[0080] In this embodiment, the processor sums the planar frictional stress of the sidewall and converts it into a precise parallel shear stress vector of 30 kN along the tunnel axis.

[0081] In this embodiment, the processor analyzes the extreme flow field topology at the corner and accurately calculates that the concrete in this area exerts a vertical upward lifting force of 50KN on the top formwork (local uplift pressure vector) and a frictional pulling force of 30KN along the tangential direction of the formwork (parallel shear stress vector).

[0082] Step S4: Input the local uplift pressure vector and the parallel shear stress vector into the discriminator in the adversarial network to calculate and generate an asymmetric support deployment scheme that includes the coordinate parameters of the anti-uplift embedded part and the coordinate parameters of the anti-shear inclined member, and generate the coordinate parameters of the anti-uplift embedded part.

[0083] This step embodies the design logic of adversarial game theory. The previous step extracted the destructive forces (local uplift pressure vector and parallel shear stress vector), while the discriminator in this step acts as a defender. After receiving these destructive force vectors, it breaks away from the traditional bias of horizontal and vertical, perfectly symmetrical arrangement in support systems, specifically placing resistive elements at mechanical extreme points. The coordinate parameters of the anti-uplift embedded parts refer to the spatial positions of the mechanical components used to firmly hold the formwork and prevent it from floating; the coordinate parameters of the anti-shear diagonal members refer to the positions of the steel pipes used for diagonal support to counteract parallel friction. The final asymmetric support deployment scheme is physically uneven in density and highly customized.

[0084] Specifically, this step includes the following sub-steps:

[0085] Step S4.1: Initialize the initial support topology matrix with orthogonal structure;

[0086] The purpose of this step is to provide a basic framework for the modification. Before performing asymmetric reinforcement, a standard, horizontally and vertically aligned full-span scaffolding grid is first generated according to traditional experience. This grid is then converted into a digital initial support topology matrix, serving as the foundational environment for subsequent reinforcement simulations by the discriminator. In this embodiment, the processor automatically generates an initial support topology matrix below the tunnel, consisting of standard uprights and horizontal bars spaced 1 meter apart.

[0087] Step S4.2: Map the local uplift pressure vector to the loaded node of the initial support topology matrix, and calculate the pull-out bearing capacity difference matrix in combination with the preset ultimate bearing capacity of the embedded parts.

[0088] The purpose of this sub-step is to find weaknesses in the defense line. The local uplift pressure vector obtained in the previous step is applied to the virtual connection points (load-bearing nodes) corresponding to the initial support topology matrix. Since traditional steel pipes are mainly resistant to compression but not tension (i.e., pull-out), the upward tensile force is subtracted from the preset ultimate bearing capacity of the mechanical connector to obtain the pull-out bearing capacity difference matrix. This matrix reflects the extent to which the system cannot resist the buoyancy of the fluid.

[0089] In this embodiment, the processor applies a local uplift pressure vector of 50KN to the virtual node at the top of the tunnel and finds that the standard fastener can only withstand a tensile force of 10KN. Thus, it calculates that there is a dangerous gap of 40KN in the difference matrix of the pull-out bearing capacity of the node.

[0090] Step S4.3: Extract the target tension nodes with values ​​greater than zero from the tensile bearing capacity difference matrix, and generate the coordinate parameters of the tensile embedded parts at the target tension nodes.

[0091] Based on the results of the previous step, if the difference is greater than zero, it means that the target tension node will be pulled apart by the fluid. Therefore, the discriminator accurately issues instructions to add tensile defense lines at these coordinates where fracture is imminent, generating coordinate parameters for the pull-out pre-embedded part. In this embodiment, for the target tension node with a 40KN gap, the processor accurately captures its spatial coordinates X, Y, Z, and generates coordinate parameters for the pull-out pre-embedded part using a special tension bolt on these coordinates.

[0092] Step S4.4: Map the parallel shear stress vector to the initial support topology matrix and calculate the lateral displacement tendency component of each node.

[0093] This sub-step corresponds to the above-described resistance to buoyancy; this step is used to resist parallel friction. The parallel shear stress vector obtained in the previous step is applied to the initial support topology matrix. Since the friction force is lateral, it will push the entire support frame to translate. The distance that each intersection point attempts to move in the horizontal direction, i.e., the lateral displacement tendency component, is calculated through mechanical transfer. In this embodiment, the processor applies a 30KN parallel shear stress vector to the matrix and calculates the lateral displacement tendency component of the uprights near the sidewall, which causes a 3 cm rightward tilt.

[0094] Step S4.5: Identify the target lateral displacement tendency component that exceeds the stiffness deformation threshold, and generate a spatial connection line vector in the adjacent spatial element of the target lateral displacement node.

[0095] Explanation: When the aforementioned lateral displacement tendency component exceeds the maximum deformation limit allowed by national standards (i.e., the stiffness deformation threshold), the location is determined to be a target lateral displacement node (i.e., a point about to collapse). In this case, in the computer model, a virtual diagonal line is drawn from this point to a stable foundation or main node, generating a spatial connection vector to construct a geometric triangle resisting lateral thrust. In this embodiment, a 3 cm offset is found to exceed the standard threshold of 1.5 cm. The processor immediately locks it as a target lateral displacement node and draws a diagonal line (i.e., a spatial connection vector) between it and the adjacent foundation node.

[0096] Step S4.6: Match and replace the spatial connection line vector to generate the coordinate parameters of the shear-resistant oblique member, and combine them with the coordinate parameters of the pull-out embedded part to form an asymmetric support deployment scheme.

[0097] This step is crucial for drawing generation. The virtual geometric lines (spatial connection vectors) generated in the previous step are matched with actual standard-length steel pipes (e.g., finding the closest 2.5-meter or 3-meter long diagonal tie rods), resulting in the actual coordinate parameters of the shear-resistant diagonal members. Subsequently, the tensile and shear design instructions are combined to formally form the final output of the adversarial game—the asymmetric support deployment scheme.

[0098] In this embodiment, the processor finds a 3-meter diagonal bar matching the diagonal length in the inventory database, generates shear-resistant diagonal bar coordinate parameters, and combines them with the aforementioned tie bolt positions to ultimately form an asymmetric support deployment scheme that includes details of local reinforcement.

[0099] In this embodiment, after the discriminator receives a 50KN upward and a 30KN parallel destructive force, it automatically adds three tensile anchors at the top corner of the tunnel on the virtual drawing (generating anti-pull-out embedded part coordinate parameters) and adds two rows of diagonal steel pipes at a 45-degree angle on the side (generating anti-shear diagonal rod coordinate parameters), thus forming an asymmetric support deployment scheme with uneven density.

[0100] Step S5: Verify the asymmetric support deployment scheme based on the preset security verification rules, and output the target support system drawings after the verification is passed.

[0101] The purpose of this step is to ensure that the AI-generated solution not only theoretically succeeds in resisting structural errors but also conforms to real-world physical laws and national building safety standards. The asymmetric support deployment scheme generated in the previous steps is reviewed using pre-set safety verification rules. This is achieved by performing secondary calculations using classical mechanics formulas and parameters from mandatory national standards to eliminate any potential algorithmic illusions. The technical effect is to guarantee the absolute engineering safety and compliance of the final target support system drawings.

[0102] Security verification rules include a low-level physical verification process and a high-level compliance verification process, specifically including the following sub-steps:

[0103] Step S5.1: Construct the structural stiffness matrix based on the asymmetric support deployment scheme, and synthesize the local uplift pressure vector and the parallel shear stress vector into an external load column vector.

[0104] This step begins the underlying physical verification of the safety verification rules. All material elastic moduli and geometric properties of the previously output asymmetric support deployment scheme (i.e., the reinforced skeleton) are extracted and combined into a large structural stiffness matrix. Simultaneously, the destructive forces generated by the fluid (local uplift pressure vector and parallel shear stress vector) are superimposed to form a unified external load column vector. In this embodiment, the processor extracts all physical properties of the tunnel support system reinforced with diagonal braces and ground anchors into a structural stiffness matrix, and combines the 50kN uplift pressure and 30kN shear force into an external load column vector.

[0105] Step S5.2: Based on inline formulas Perform finite element analysis to calculate the deformation displacement field matrix, where Here is the structural stiffness matrix. Let U be the column vector of external loads, and U be the deformation displacement field matrix.

[0106] This is the classic finite element method for solving solid mechanics problems. By solving this system of linear or nonlinear equations, the forces acting on the system are derived. Under the action of, it has stiffness How much three-dimensional deformation will the system produce? (i.e., the deformation displacement field matrix). In this embodiment, the processor substitutes the system stiffness K and the external force F into the formula to solve for the deformation displacement field matrix composed of the specific displacement values ​​of each node of the reinforced support frame.

[0107] Step S5.3: When the maximum shear strain value of the deformation displacement field matrix is ​​less than the preset limit threshold, a physical verification pass instruction is generated.

[0108] The purpose of this sub-step is to confirm the physical defense line. It extracts the most severe torsional deformation (maximum shear strain value) from the deformation displacement field matrix of the previous step. If this deformation does not break the material (less than a preset limit threshold), it proves that the AI-generated solution is mechanically robust, and the system issues a physical verification pass command. In this embodiment, extraction revealed that due to the addition of diagonal bracing, the maximum torsional deformation was only 0.5 mm, far less than the material fracture threshold, and the system immediately generated a physical verification pass command.

[0109] Step S5.4: In response to the physical verification pass command, extract the associated local high-risk node stress peak matrix.

[0110] This step involves compliance verification. Mechanical non-fracture does not necessarily imply compliance with engineering specifications. After receiving the previous physical verification pass instruction, the focus is on extracting the internal stress values ​​(i.e., the peak stress matrix of locally high-risk nodes) of the node group experiencing the most extreme stress in the model. In this embodiment, after the physical verification passes, the processor specifically retrieves the stress data (i.e., the peak stress matrix of locally high-risk nodes) of the node connecting the tensile anchor, which reaches as high as 200 MPa.

[0111] Step S5.5: Calculate the actual safety factor matrix by dividing the preset material ultimate yield strength by each element of the stress peak matrix of the local high-risk nodes.

[0112] Explanation: Engineering designs must allow for contingencies. The actual safety factor matrix is ​​obtained by dividing the theoretical breaking strength (ultimate yield strength) of the steel by the actual peak stress extracted in the previous step. In this embodiment, the ultimate strength of the steel pipe is 400 MPa, which, divided by the actual stress of 200 MPa, yields an actual safety factor of 2.0 for this node.

[0113] Step S5.6: When the smallest element of the actual safety factor matrix is ​​greater than the minimum data of the building code, generate a compliance verification pass instruction.

[0114] The purpose of this step is to finalize legal compliance. If the safety factor of the weakest link in the system (the smallest element of the actual safety factor matrix) still meets the minimum requirements stipulated by the national housing and construction department (the baseline data of building codes, such as requiring a minimum safety factor), then this step is necessary. If the condition is met, then the application is completely approved, and a compliance verification pass instruction is generated.

[0115] In this embodiment, the processor comparison found that the calculated minimum security factor of 2.0 is greater than the mandatory requirement of 1.5 in national standards, confirming that the scheme is completely legal and compliant, and officially generating a compliance verification pass instruction.

[0116] Step S5.7: Analyze the asymmetric support deployment scheme to extract the features of the three-dimensional spatial coordinate scatter point set.

[0117] This sub-step initiates the final stage of drawing rendering. The previously generated asymmetric support deployment scheme is essentially just a bunch of cold numbers and coordinate matrices. This step extracts these, forming a pure three-dimensional spatial coordinate scatter set feature, which is then readable by the graphics engine. In this embodiment, the processor extracts a skeleton matrix (i.e., a three-dimensional spatial coordinate scatter set feature) consisting of hundreds of thousands of X, Y, Z coordinate points from a complex security verification file.

[0118] Step S5.8: Retrieve the features of the 3D spatial coordinate scatter set by mapping and replacing the entity mesh model, and generate a visual design model.

[0119] The purpose of this sub-step is to materialize the abstract coordinates. From a pre-defined library (component database), 3D model components with realistic material appearances (such as metallic sheen and specific pipe diameters) are retrieved and placed one by one onto the coordinate points extracted in the previous step, transforming the skeleton matrix into a visually perceptible 3D design model. In this embodiment, the processor automatically retrieves real 3D models of steel pipes and bases from the library, inserts them one by one into the matrix coordinates, and generates a realistic 3D visual design model of tunnel support containing complex tie rods.

[0120] Step S5.9: Perform orthographic projection processing on the visualized design model to generate a two-dimensional target support system drawing with parametric annotations.

[0121] The purpose of this sub-step is to generate the final engineering deliverable. Using orthographic projection algorithms in computer graphics, the 3D visualization design model is flattened into traditional planar views (such as top and side views), and dimensions and material specifications are automatically annotated, generating the final output—the target support system drawings. In this embodiment, the processor cuts and projects the 3D model from different perspectives, ultimately outputting tunnel support construction plan and elevation drawings (i.e., target support system drawings) that workers can understand and that are marked with accurate dimensions and angles.

[0122] In this embodiment, the processor uses structural mechanics formulas and the "Formwork Engineering Construction Specification" to verify the asymmetric support deployment scheme, confirming that the added diagonal braces and ground anchors do not exceed the yield strength of the steel. After verification, the target support system drawings for workers to construct are officially printed out.

[0123] In addition, the present invention also includes the following optional embodiments:

[0124] Implementation of the time-series reinforcement early warning function: Since the concrete pumping pressure sequence carries timestamp tag data, the method of the present invention further includes the following steps:

[0125] Step S6.1: Locate the local uplift pressure peak coordinates corresponding to each timestamp tag data in the extreme flow field topology.

[0126] The purpose of this step is to upgrade static drawings into a four-dimensional dynamic construction early warning system. Since pumping is a process that occurs over time, the dangerous stress locations (local uplift pressure peak coordinates) of the preceding extreme flow field topology are extracted at different time points (i.e., timestamp tag data). In this embodiment, the processor backtracks the flow field model and finds that the uplift pressure peak is at the lower left corner of the tunnel at the 10-minute mark, and shifts to the tunnel arch at the 40-minute mark.

[0127] Step S6.2: Based on inline formulas Calculate and generate a three-dimensional fluid wavefront propagation time series table, in which This describes the wave propagation velocity characteristics. This step explains the tracking principle of dynamic concealed risks. In the formula, Representing fluid in The distance or location of propagation at any given moment. Starting from point A, the integral part represents the accumulation of velocity over time. Using this formula, the system can accurately depict how the concrete fluid front (wavefront) propagates forward like an ocean wave within the mold cavity over time, forming a three-dimensional fluid wavefront propagation time series.

[0128] In this embodiment, the processor uses the integral formula to calculate the specific advance coordinates of the concrete liquid surface inside the tunnel every 5 minutes, and compiles them into a three-dimensional fluid wavefront propagation timing table.

[0129] Step S6.3: Map the three-dimensional fluid wavefront propagation time series table to the asymmetric support deployment scheme and output the time series hardening early warning file.

[0130] The technical benefit of this step is that it directly guides on-site workers in their practical operations. By integrating the time ripple calculated in the previous step with the asymmetric support deployment scheme generated earlier, it tells on-site personnel when to tighten which screws. The output time-series reinforcement early warning file gives the support system dynamic response capabilities.

[0131] In this embodiment, the system integrates the timetable with the design drawings, and the output time-series reinforcement early warning file prompts the workers: please go to the tunnel arch C area immediately after the 35th minute after the start of pouring to carry out secondary torque reinforcement on the pre-arranged anti-pull-out embedded parts.

[0132] Implementation methods for obtaining rheological parameters of non-Newtonian fluids:

[0133] Step S7.1: Receive the temperature matrix and dynamic slump extension data of the mixture at the target construction site.

[0134] This step details how to obtain non-Newtonian fluid parameters. Because concrete is greatly affected by the on-site environment, it is essential to collect the temperature distribution of the freshly mixed concrete (mixture temperature matrix) and its area spread naturally on the ground (dynamic slump spread data). In this embodiment, the sensor installed on the mixer truck transmitted a current concrete temperature of 25 degrees Celsius (mixture temperature matrix), and the on-site workers measured its spread diameter to be 600 millimeters (dynamic slump spread data).

[0135] Step S7.2: Input the temperature matrix of the mixture and the dynamic slump extension data into the preset nonlinear rheological support vector machine model.

[0136] Explanation: Temperature and spread area are merely surface phenomena; machine learning is needed to map them to essential physical parameters. This step inputs the two surface data points into an algorithm specifically designed for nonlinear classification and regression (i.e., a nonlinear rheological support vector machine model). In this embodiment, the processor packages the values ​​of 25 degrees and 600 millimeters and inputs them into a cloud-trained nonlinear rheological support vector machine model for inference.

[0137] Step S7.3: Perform mapping calculations using the radial basis kernel function, and output the plastic viscosity property value and yield stress property value to combine them into non-Newtonian fluid rheological parameters.

[0138] This step reveals the internal execution principle of the machine learning model. Since the relationship between temperature, slump, and fluid dynamics parameters is extremely complex, the radial basis function (RBF) is used as a mathematical tool to project low-dimensional surface data into a high-dimensional space for decoding. The final output is a core indicator reflecting rheological properties (plastic viscosity and yield stress). Combining these two indicators yields the desired non-Newtonian fluid rheological parameters.

[0139] In this embodiment, the model successfully transformed the surface data into a plastic viscosity of 50 Pa·s and a yield stress of 80 Pa through high-dimensional kernel function calculation. These two parameters are combined to form non-Newtonian fluid rheological parameters for use in generative adversarial network simulation.

[0140] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A design method for a concrete formwork support system based on machine learning, characterized in that, Executed by the processor, including the following steps: Acquire the three-dimensional internal cavity topology data, concrete pumping pressure sequence, and non-Newtonian fluid rheological parameters of the target template system; The three-dimensional internal cavity topology data, the concrete pumping pressure sequence, and the non-Newtonian fluid rheological parameters are input into the generator in a pre-trained generative adversarial network to output the extreme flow field topology. Based on the extreme flow field topology, the local uplift pressure vector and parallel shear stress vector are extracted; The local uplift pressure vector and the parallel shear stress vector are input into the discriminator in the generative adversarial network to calculate and generate an asymmetric support deployment scheme that includes the coordinate parameters of the anti-uplift embedded part and the coordinate parameters of the anti-shear inclined member, and generate the coordinate parameters of the anti-uplift embedded part; The asymmetric support deployment scheme is verified based on preset security verification rules, and the target support system drawing is output after the verification is passed.

2. The design method for a machine learning-based concrete formwork support system according to claim 1, characterized in that, The step of inputting the three-dimensional cavity topology data, the concrete pumping pressure sequence, and the non-Newtonian fluid rheological parameters into the generator in a pre-trained generative adversarial network includes: A discretized mesh model is constructed based on the three-dimensional internal cavity topology data; Based on the discretized grid model, a physical partial differential loss function is introduced into the generator, combined with an inline formula. Calculate the momentum conservation residual, where For density, For the velocity field, For pressure field, Dynamic viscosity; The network weights are updated iteratively using the momentum conservation residuals to output the extreme flow field topology that satisfies the mechanism constraints.

3. The design method for a machine learning-based concrete formwork support system according to claim 1, characterized in that, The steps for extracting the local uplift pressure vector and parallel shear stress vector based on the extreme flow field topology include: Extract the normal momentum tensor of the extreme flow field topology at the boundary grid, and convert the normal momentum tensor into the upward local uplift pressure vector; Based on the Herschel-Balclay rheological model and inline formulas Calculate the wall friction stress matrix, where For yield stress, This is the consistency coefficient. For shear rate, The rheological index; The wall friction stress matrix is ​​converted into the parallel shear stress vector.

4. The design method for a machine learning-based concrete formwork support system according to claim 1, characterized in that, The step of inputting the local uplift pressure vector and the parallel shear stress vector into the discriminator in the generative adversarial network to calculate and generate an asymmetric support deployment scheme that includes the coordinate parameters of the uplift-resistant embedded part and the coordinate parameters of the shear-resistant inclined member, and generating the coordinate parameters of the uplift-resistant embedded part, includes: Initialize the initial support topology matrix with orthogonal structure; The local uplift pressure vector is mapped to the loaded node of the initial support topology matrix, and the pull-out bearing capacity difference matrix is ​​calculated in combination with the preset ultimate bearing capacity of the embedded parts. Extract the target tension nodes with values ​​greater than zero from the tensile bearing capacity difference matrix, and generate the coordinate parameters of the tensile pre-embedded part at the target tension node.

5. The design method for a machine learning-based concrete formwork support system according to claim 4, characterized in that, The step of calculating and generating an asymmetric support deployment scheme further includes: The parallel shear stress vector is mapped to the initial support topology matrix, and the lateral displacement tendency component of each node is calculated. Identify the target lateral displacement tendency component that exceeds the stiffness deformation threshold, and generate a spatial connection line vector in the adjacent spatial element of the target lateral displacement node. The coordinate parameters of the anti-shear inclined member are generated by matching and replacing the spatial connection line vector, and combined with the coordinate parameters of the anti-pull-out embedded part to form the asymmetric support deployment scheme.

6. The design method for a machine learning-based concrete formwork support system according to claim 1, characterized in that, The steps for verifying the asymmetric support deployment scheme based on preset security verification rules include the underlying physical verification process: Based on the asymmetric support deployment scheme, a structural stiffness matrix is ​​constructed, and the local uplift pressure vector and the parallel shear stress vector are combined into an external load column vector. Based on inline formula Perform finite element analysis to calculate the deformation displacement field matrix, where Here is the structural stiffness matrix. Let U be the column vector of external loads, and U be the deformation displacement field matrix. When the maximum shear strain value of the deformation displacement field matrix is ​​less than a preset limit threshold, a physical verification pass command is generated.

7. The design method for a machine learning-based concrete formwork support system according to claim 6, characterized in that, The verification process also includes a top-level compliance verification procedure: In response to the physical verification pass instruction, the associated local high-risk node stress peak matrix is ​​extracted; The actual safety factor matrix is ​​calculated by dividing the preset material ultimate yield strength by each element of the stress peak matrix of the local high-risk nodes. When the smallest element of the actual safety factor matrix is ​​greater than the minimum data of the building code, a compliance verification pass instruction is generated.

8. The design method for a machine learning-based concrete formwork support system according to claim 1, characterized in that, The concrete pumping pressure sequence, carrying timestamp-labeled data, also includes: Locate the local uplift pressure peak coordinates corresponding to each of the timestamp tag data in the extreme flow field topology; Based on inline formula Calculate and generate a three-dimensional fluid wavefront propagation time series table, in which For the characteristics of wave propagation speed, S t Let S be the position of the wavefront at time t, and S0 be the initial point; The three-dimensional fluid wavefront propagation timing table is mapped to the asymmetric support deployment scheme, and a timing reinforcement early warning file is output.

9. The design method for a machine learning-based concrete formwork support system according to claim 1, characterized in that, The steps for obtaining rheological parameters of non-Newtonian fluids include: Receive the temperature matrix and dynamic slump extension data of the mixture at the target construction site; The temperature matrix of the mixture and the dynamic slump expansion data are input into a preset nonlinear rheological support vector machine model; The radial basis function is used to perform mapping calculations, and the plastic viscosity property value and yield stress property value are output to be combined into the rheological parameters of the non-Newtonian fluid.

10. The design method for a machine learning-based concrete formwork support system according to claim 1, characterized in that, The steps for outputting the target support system drawings after successful verification include: The asymmetric support deployment scheme is analyzed to extract the features of the three-dimensional spatial coordinate scatter point set; The features of the three-dimensional spatial coordinate scatter point set are replaced by the mapping of the solid mesh model to generate a visual design model; The visualization design model is subjected to orthographic projection processing to generate a two-dimensional target support system drawing with parametric annotations.

Images