Concrete structural member prefabrication precision management and control system based on BIM and digital twinning

Abstract

The invention discloses a concrete structural member prefabrication precision management and control system based on BIM and digital twinning. The system comprises a multi-physics field precision prediction module, a precision gene map construction module, an intelligent optimization decision module, a real-time data acquisition module, a digital twinning mapping module and a precision evaluation output module. The system predicts component deformation by establishing a coupling model of a temperature field, a humidity field, a stress field and a chemical field and generates a reverse compensation scheme; constructing a multi-dimensional feature system containing materials, processes, environments and structural genes to realize knowledge reuse; optimizing production parameters by using a genetic algorithm and a deep learning network; the state of the component is monitored in real time through laser scanning and various sensors; constructing a synchronously updated digital twinborn body through point cloud registration and an interpolation algorithm; and calculating six geometric accuracy indexes to carry out quality evaluation and continuously updating the knowledge base. According to the method, the precision control of the whole process from prediction, production to evaluation of the prefabricated component is realized, and the precision control level of the component is remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of building engineering technology, and in particular relates to a precision control system for prefabricated concrete structural components based on BIM and digital twins. Background Technology

[0002] With the rapid development of prefabricated buildings in my country, the application of precast concrete structural components is becoming increasingly widespread. However, there are common problems in the production process of existing precast components: on the one hand, concrete shrinks and deforms during pouring and curing due to factors such as hydration reaction and temperature and humidity changes, resulting in dimensional deviations of the components; on the other hand, traditional quality inspection is mostly done after the fact, which cannot detect and correct precision deviations in a timely manner during the production process.

[0003] Existing technologies mainly rely on manual measurement or simple 3D scanning for finished product inspection, which suffers from low inspection efficiency, inability to predict deformation trends, and a lack of process control methods. Although some technical solutions have introduced BIM technology for design and construction management, they have failed to achieve deep integration with the actual production process and lack the ability to achieve full-process precision control based on digital twins.

[0004] Therefore, there is an urgent need for an intelligent management and control system that can achieve accurate prediction, real-time monitoring, and adaptive adjustment of prefabricated components throughout their entire life cycle. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a prefabrication accuracy control system for concrete structural components based on BIM and digital twins. By constructing a real-time bidirectional mapping mechanism between physical components and digital models, it enables accuracy prediction, monitoring, and adaptive adjustment throughout the prefabrication process.

[0006] This invention is achieved through the following technical solution: A BIM and digital twin-based system for controlling the precision of precast concrete structural components, the system comprising: The multiphysics field accuracy prediction module is used to read the geometric parameters of the component's length, width, and thickness from the BIM model, as well as the reinforcement parameters such as the diameter of the longitudinal and transverse reinforcing bars, the spacing of the longitudinal and transverse reinforcing bars, the thickness of the concrete cover, the cement content, the water-cement ratio, the maximum particle size of the coarse aggregate, and the fineness modulus of the fine aggregate. It establishes a set of coupled equations for the temperature field, humidity field, stress field, and chemical field, and solves them using the finite element method to predict the shrinkage strain and warping deformation of the component at different ages, generating inverse compensation amounts in the length, width, and thickness directions. The precision gene map construction module is used to extract material features, process features, environmental features and structural features that affect precision, construct multi-dimensional feature vectors, reduce dimensionality through principal component analysis, establish a precision classification system using K-means clustering, and use association rule mining to discover strong correlations between parameters and precision, forming a searchable and reusable knowledge base; The intelligent optimization decision module is used to calculate the Euclidean distance between the current component feature vector and historical cases, select similar cases to determine the optimal range of process parameters, use a genetic algorithm to search for the optimal parameter combination through selection, crossover and mutation operations, establish a deep neural network to predict the accuracy deviation, and generate control instructions when the deviation exceeds the threshold. The real-time data acquisition module is used to acquire the three-dimensional point cloud coordinates of the component surface through laser scanning equipment, measure the internal temperature distribution through temperature sensor, measure the relative humidity through capacitive humidity sensor, measure the strain value through fiber optic grating sensor, measure the vibration acceleration through piezoelectric accelerometer, and dynamically adjust the acquisition frequency according to the production stage. The digital twin mapping module is used to register point cloud and BIM model using iterative nearest point algorithm, calculate normal deviation distribution, generate continuous deviation field through Kriging interpolation, generate temperature and humidity field distribution through inverse distance weighted interpolation, and construct digital twin that is updated synchronously with physical components. The accuracy evaluation output module is used to calculate six geometric accuracy indicators: length deviation, width deviation, thickness deviation, diagonal difference, flatness, and perpendicularity. It calculates the individual scores through piecewise functions, and obtains the comprehensive score by weighted summation. It determines the quality level, generates an inspection report containing a three-dimensional deviation cloud map, and feeds the production data back to the knowledge base.

[0007] Furthermore, in the multiphysics field accuracy prediction module, the temperature field is described by a three-dimensional unsteady-state heat conduction equation, and the hydration heat generation rate is related to the cement dosage and degree of hydration; the humidity field is described by Fick's diffusion law, and the moisture diffusion coefficient is related to the aggregate parameters; the stress field establishes the stress-strain relationship through constitutive equations, considering the influence of reinforcement ratio on stiffness; the chemical field describes the hydration reaction rate through the Arrhenius equation, and the activation energy is related to the water-cement ratio; the shrinkage strain includes three components: autogenous shrinkage, drying shrinkage, and carbonation shrinkage.

[0008] Furthermore, the material characteristics of the precision gene mapping construction module include cement type, dosage, water-cement ratio, aggregate gradation, admixture type and dosage; process characteristics include vibration time, frequency, rod diameter, insertion depth, point spacing, pouring layer thickness, speed, temperature, demolding time and strength; environmental characteristics include ambient temperature, curing temperature, temperature variation range, ambient humidity, curing humidity, wind speed and atmospheric pressure; structural characteristics include component dimensions, reinforcement parameters, protective layer thickness, reinforcement ratio and embedded part volume.

[0009] Furthermore, the genetic algorithm of the intelligent optimization decision-making module adopts real number encoding, the fitness function comprehensively considers accuracy deviation, production time and cost constraints, and the population size and number of generations are set according to the complexity of the problem; the deep neural network includes an input layer, multiple hidden layers and an output layer, adopts the ReLU activation function, and is trained through the backpropagation algorithm.

[0010] Furthermore, the digital twin mapping module achieves point cloud registration through three steps: feature extraction, coarse registration, and fine registration, with the registration error controlled at the millimeter level; Kriging interpolation determines spatial correlation based on a semi-variogram function, and the weight of inverse distance weighted interpolation is inversely proportional to the power of the distance.

[0011] Furthermore, the quality grade judgment criteria of the accuracy evaluation output module are as follows: a comprehensive score of 90 or above is considered excellent, 60 to 90 is considered qualified, and less than 60 is considered unqualified; the deviation cloud map uses color coding, with blue indicating negative deviation, green indicating qualified, and red indicating positive deviation.

[0012] Furthermore, the workflow of the system includes: The first step is for the multiphysics accuracy prediction module to read the BIM model and material parameters, solve the coupled equations, predict the deformation, and generate a compensation scheme. The second step involves the precision gene map construction module extracting feature parameters, constructing feature vectors, and establishing a knowledge base. The third step is for the intelligent optimization decision-making module to search for similar cases and optimize process parameters; Fourth, the real-time data acquisition module acquires point cloud and sensor data; Fifth step: The digital twin mapping module registers the data, updates the digital twin, and returns to the third step if the deviation exceeds the threshold; The sixth step involves the accuracy evaluation output module calculating the accuracy index, assigning a rating, and updating the knowledge base accordingly. Beneficial effects

[0013] 1. Pre-control of accuracy throughout the entire process: Through multi-physics field coupling prediction and reverse compensation design, the accuracy problem can be prevented in advance, and deformation can be controlled from the source; 2. Knowledge-driven optimization: Based on high-precision genomic maps and historical experience, intelligent parameter optimization and knowledge reuse are achieved; 3. Real-time transparent monitoring: Real-time monitoring and visualization of the production process are achieved through multi-source sensors and digital twin technology; 4. Adaptive closed-loop control: Automatically adjusts process parameters based on deviation analysis to achieve dynamic control of accuracy; 5. Continuous improvement mechanism: Through data feedback and model updates, the system performance is continuously improved with use. Attached Figure Description

[0014] Figure 1 A schematic diagram of the system architecture described in this invention is shown; Figure 2 A flowchart illustrating the working steps of the system described in this invention is shown. Detailed Implementation

[0015] Exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0016] Combination Figure 1 This invention provides a BIM and digital twin-based prefabrication accuracy control system for concrete structural components. Through the collaborative work of six functional modules, the system forms a complete closed-loop control system from design to production and from monitoring to evaluation.

[0017] The multiphysics field accuracy prediction module, serving as the system's prediction engine, predicts the deformation patterns of components throughout the entire production process by establishing and solving a set of coupled multiphysics equations. This module extracts the component's geometric information from the BIM model data interface, including length, width, and thickness, as well as reinforcement information, including longitudinal and transverse steel bar diameters, longitudinal bar spacing, transverse bar spacing, and concrete cover thickness. It also reads concrete mix proportion parameters from the material database, including cement content, water-cement ratio, maximum coarse aggregate size, and fineness modulus of fine aggregate.

[0018] Based on the read parameters, the module establishes a finite element model, and the mesh size is determined according to the rebar spacing: Where: Δx is the grid size; This refers to the spacing between longitudinal reinforcement bars; is the spacing of the horizontal reinforcement bars; H is the thickness of the component; min is the function to take the minimum value.

[0019] The temperature field follows a three-dimensional unsteady-state heat conduction equation:

[0020] Where: ρ is the density of concrete; t is the specific heat capacity of concrete; T is the temperature; t is the time; k is the thermal conductivity. For gradient operators; • For divergence operators; The rate of heat of hydration generation per unit volume.

[0021] Concrete density is calculated based on the mix proportions:

[0022] Where: C represents the amount of cement used; Water usage; This refers to the amount of coarse aggregate used; This refers to the amount of fine aggregate used.

[0023] The thermal conductivity k is related to the moisture content and temperature.

[0024] in: The reference thermal conductivity; Here, W is the influence coefficient of moisture content; W is the moisture content. is the temperature influence coefficient; T is the temperature.

[0025] The rate of heat of hydration is related to the amount of cement used and the degree of hydration.

[0026] Where: C represents the amount of cement used; Heat of hydration per unit mass of cement; This represents the rate of change in hydration degree.

[0027] The rate of change in hydration degree is described by the Arrhenius equation: Where: A is the frequency factor; f(α) is the hydration degree function; exp is the natural exponential function; is the activation energy; R is the ideal gas constant; T is the absolute temperature.

[0028] Activation energy is related to the water-cement ratio:

[0029] Where: Ea is the activation energy; w / c is the water-cement ratio.

[0030] The degree of hydration function takes into account the influence of cement type:

[0031] in: α is the cement type coefficient; n and m are the current degree of hydration; and m and m are the reaction order.

[0032] The boundary conditions of the temperature field include convective heat transfer with the environment and thermal conduction with the template. Convective heat transfer boundary conditions:

[0033] in: The surface normal temperature gradient; The convective heat transfer coefficient; Surface temperature; The ambient temperature.

[0034] The convective heat transfer coefficient is related to the wind speed:

[0035] in: The convective heat transfer coefficient; This refers to wind speed.

[0036] The humidity field describes the migration and consumption of moisture within the concrete. The initial moisture content is determined by the mix proportions.

[0037] in: This represents the initial volumetric water content. C represents the amount of water used; C represents the amount of cement used. This refers to the amount of coarse aggregate used; This refers to the amount of fine aggregate used.

[0038] Moisture diffusion follows Fick's Law:

[0039] Where: W is the volumetric water content; This is the partial derivative of moisture content with respect to time; The water diffusion coefficient; This refers to the source of water consumption.

[0040] The moisture diffusion coefficient is related to aggregate parameters: in: The water diffusion coefficient; The reference diffusion coefficient; This represents the maximum particle size of the coarse aggregate. It is the fineness modulus of fine aggregate.

[0041] Water consumption sources include hydration consumption and evaporation loss: in: Total consumption; Consumed for hydration; This is due to evaporation loss.

[0042] Hydration consumption is directly proportional to the rate of change in hydration degree:

[0043] in: dα / dt is the hydration consumption; C is the cement dosage; dα / dt is the hydration degree change rate; 0.25 is the theoretical water demand coefficient for the hydration reaction.

[0044] The boundary conditions of the humidity field take into account the influence of the protective layer thickness: Where: ∂W / ∂n is the surface normal humidity gradient; The mass transfer coefficient; Surface moisture content; This represents the equilibrium moisture content corresponding to the ambient humidity.

[0045] The mass transfer coefficient is inversely proportional to the thickness of the protective layer:

[0046] in: The mass transfer coefficient; is the reference mass transfer coefficient; c is the thickness of the protective layer.

[0047] Stress field analysis needs to consider the combined effects of temperature stress, shrinkage stress, and constraint stress. The stress-strain relationship is described by constitutive equations:

[0048] Where: σ is the stress tensor; ε is the elasto-plastic stiffness matrix; ":" is the tensor double dot product operator; ε is the total strain tensor; For contraction strain tensor; For temperature strain tensor; Let be the creep strain tensor.

[0049] Elastic-plastic stiffness matrix considering the influence of reinforcement:

[0050] in: This is the equivalent stiffness matrix; Here is the concrete stiffness matrix; ρ is the reinforcement ratio. The stiffness matrix contributed by the reinforcing steel.

[0051] The reinforcement ratio is calculated using the steel reinforcement parameters:

[0052] Where: ρ is the reinforcement ratio; π is the ratio of pi; The diameter of the longitudinal reinforcing bars; This refers to the spacing between longitudinal reinforcement bars; The diameter of the transverse reinforcing bar; This refers to the spacing between the horizontal reinforcing bars.

[0053] Shrinkage strain includes three components: autogenous shrinkage, drying shrinkage, and carbonization shrinkage.

[0054] in: This represents the total contraction strain; For self-contraction strain; This is due to drying shrinkage strain; This is for carbonization shrinkage strain.

[0055] Self-shrinkage strain is generated by chemical shrinkage caused by cement hydration: in: (t) represents the self-shrinkage strain at age t; This represents the final self-shrinking strain value; This is the self-shrinkage rate coefficient; exp is the self-contraction time exponent; exp is the natural exponential function.

[0056] The final autogenous shrinkage strain value is related to the cement dosage and water-cement ratio: in: is the material coefficient; C is the cement content; w / c is the water-cement ratio.

[0057] Drying shrinkage strain takes into account the effects of component size and ambient humidity: in: The drying shrinkage strain at age t; This represents the final drying shrinkage strain value; The relative humidity influence coefficient; This is the dimensional influence coefficient; t is the start drying time; h is the theoretical thickness.

[0058] The theoretical thickness is calculated based on the component's surface area ratio: Where: h is the theoretical thickness; V is the volume; A is the exposed surface area; L is the length; W is the width; and H is the thickness.

[0059] Relative humidity influence coefficient:

[0060] in: RH represents the humidity effect coefficient; RH represents the ambient relative humidity.

[0061] Size influence factor:

[0062] in: is the dimensional influence coefficient; h is the theoretical thickness.

[0063] Carbonation shrinkage strain is caused by the reaction of carbon dioxide with cement hydration products: in: The carbonization shrinkage strain at age t; This represents the final carbonization shrinkage strain value; This is the carbonization rate coefficient.

[0064] Temperature strain is caused by temperature changes:

[0065] in: For temperature strain; ΔT is the coefficient of thermal expansion; ΔT is the temperature change.

[0066] The coefficient of thermal expansion is related to the type of aggregate:

[0067] in: The coefficient of thermal expansion of concrete; The coefficient of thermal expansion of cement paste; The coefficient of thermal expansion of the aggregate; This represents the aggregate volume fraction.

[0068] Creep strain describes the time-varying deformation of concrete under sustained stress: in: σ(t0) represents the creep strain from loading at time t0 to time t; σ(t0) represents the stress at time t0; E(t0) represents the elastic modulus at time t0; and φ(t,t0) represents the creep coefficient.

[0069] The creep coefficient is expressed as an exponential function: in: This is the final creep coefficient; This is the creep rate coefficient.

[0070] The chemical field describes the process of cement hydration reaction, and the development of hydration degree follows a kinetic equation: Where: α(t) is the degree of hydration at time t; The maximum degree of hydration; It is the hydration rate constant; It represents the hydration reaction order.

[0071] The maximum degree of hydration is related to the water-cement ratio:

[0072] in: is the maximum degree of hydration; w / c is the water-cement ratio; min is the function to take the minimum value.

[0073] By solving the above four-field coupled equations, the multiphysics precision prediction module obtains the temperature, humidity, stress, and strain distributions of the component at different ages and locations. Based on the prediction results, the module calculates the total deformation in each direction:

[0074] Where: Δ The total deformation is along the length; L is the length of the component. , , These are shrinkage, temperature, and creep strain, respectively.

[0075] To compensate for the predicted shrinkage deformation, the module generates a reverse compensation scheme: in: Template size; For design dimensions; Δ To predict the amount of deformation.

[0076] The compensation amounts in both the width and thickness directions are calculated using the same method. The compensation scheme also considers the non-uniformity of deformation; for areas with large warping deformation, a differentiated compensation strategy is adopted.

[0077] The precision gene map construction module is responsible for systematically encoding the multidimensional factors affecting precision, forming a searchable and reusable knowledge system. This module defines feature vectors containing four categories of features: materials, processes, environment, and structure, each of which is closely related to precision control.

[0078] The coding of material characteristics covers key parameters of concrete mix design. Cement type is identified through coding; different types of cement have different hydration and shrinkage properties. Cement dosage directly affects the total heat of hydration and the magnitude of shrinkage. The water-cement ratio is a key parameter controlling concrete strength and shrinkage. The maximum particle size of coarse aggregate affects concrete workability and shrinkage. The fineness modulus of fine aggregate affects concrete density. Fly ash and mineral powder admixtures, as mineral admixtures, can improve concrete performance. The type and dosage of water-reducing agents affect concrete fluidity and shrinkage. The dosage of expansive agents is used to compensate for shrinkage.

[0079] The coding of process characteristics reflects the control parameters of the production process. Vibration parameters include vibration time (tv), vibration frequency, vibrator diameter, insertion depth, and spacing between vibration points; these parameters collectively determine the density of the concrete. Pouring parameters include pouring layer thickness, pouring speed, and pouring temperature, which affect the initial state and temperature field distribution of the concrete. Demolding parameters include demolding time and demolding strength, which determine the state of the component upon demolding.

[0080] Environmental characteristics encode the conditions of the production environment. Ambient temperature during pouring affects the initial temperature of the concrete. The temperature of the curing chamber and its daily variation determine the temperature conditions during curing. Ambient relative humidity and curing humidity affect the development of drying shrinkage. Wind speed affects surface heat transfer and mass transfer. Atmospheric pressure affects the rate of moisture evaporation.

[0081] Structural features encode the geometric and reinforcement information of the components, including component dimensions L, W, H, reinforcement parameters d1, d2, s1, s2, protective layer thickness c, reinforcement ratio ρ, and embedded part volume. .

[0082] Feature vector standardization is a prerequisite for similarity calculation; the z-score standardization method is used. in: These are the standardized features; Original features; Let be the mean of the i-th feature; Let be the standard deviation of the i-th feature.

[0083] The standardized eigenvectors are then subjected to dimensionality reduction using principal component analysis. First, the covariance matrix is ​​constructed: Where: C is the covariance matrix; n is the number of samples; X is the standardized feature matrix.

[0084] Find the eigenvalues ​​and eigenvectors of the covariance matrix: in: Let i be the i-th eigenvector; These are the corresponding eigenvalues.

[0085] The eigenvalues ​​are sorted from largest to smallest, and the top m principal components whose cumulative contribution rate reaches a set threshold are selected: Where: m is the number of principal components selected; d is the original feature dimension; η is the contribution rate threshold, usually taken as 0.9.

[0086] The dimensionality-reduced features are clustered using the K-means algorithm, grouping components with similar precision features into one class. The objective function for clustering is: Where: J is the clustering objective function; k is the number of clusters; For the j-th category; For the sample; It is the center of class j.

[0087] The formula for updating cluster centers is: in: Centered on the j-th class; | | represents the number of samples in the j-th class.

[0088] Association rule mining is used to discover the intrinsic relationship between parameters and accuracy. For rule A⇒B, the support is defined as: Where: Support is the support level; A is the set of parameters of the preceding term; B is the accuracy result of the following term; count(A∪B) is the number of samples that simultaneously contain both A and B; N is the total number of samples.

[0089] Confidence level is defined as: Where: Confidence is the confidence level; count(A) is the number of samples containing A.

[0090] Lift is used to evaluate the effectiveness of rules: Where: Lift is the lift degree; when Lift>1, it indicates that the rule has a positive correlation.

[0091] The intelligent optimization decision-making module performs parameter optimization and deviation prediction based on a precision gene map. When a new production task arrives, the module first calculates the similarity between the current component and historical cases in the knowledge base. The similarity is measured using a weighted Euclidean distance metric. in: Let be the distance between component i and component j; The weight of the k-th feature; Let i be the k-th feature of component i; Let be the k-th feature of component j; m is the feature dimension.

[0092] Feature weights are determined by information gain:

[0093] in: IG represents the weight of the k-th feature; D represents the information gain; and IG represents the dataset. Let H(D) be the k-th feature; H(D) be the entropy of the dataset; H(D| ) represents the conditional entropy.

[0094] Select the K most similar cases with the smallest distance, analyze the distribution of their process parameters, and determine the optimization range of the parameters:

[0095] in: , These are the lower and upper bounds of the parameter range; The mean of the parameters; is the standard deviation of the parameter; k is the range coefficient, usually taken as 2.

[0096] Within a defined parameter range, a genetic algorithm is used to search for the optimal parameter combination. The chromosomes in the genetic algorithm are encoded using real numbers.

[0097] Where: C represents chromosome; For the vibration time; This refers to the vibration frequency; For maintenance temperature; To maintain humidity; Demolding time; This refers to the demolding strength.

[0098] The initial population is generated uniformly and randomly:

[0099] in: For the j-th gene of the i-th individual; , The range of the j-th parameter; rand() is a uniformly distributed random number in the range [0,1].

[0100] The fitness function comprehensively considers accuracy requirements, production efficiency, and cost constraints: Where: F represents fitness; The weight of the k-th accuracy index; β is the accuracy deviation of the k-th term; β is the time penalty coefficient; Production time; γ is the cost penalty coefficient; For production costs.

[0101] The selection operation uses a roulette wheel method, where the probability of an individual being selected is directly proportional to its fitness. in: Let be the probability that the i-th individual is selected; Let M be the fitness of the i-th individual; M is the population size.

[0102] The crossover operation uses arithmetic crossover to generate two offspring: in: , For offspring; , α is the parent generation; α is the cross coefficient, which is randomly generated in the interval [0,1].

[0103] The mutation operation uses Gaussian mutation:

[0104] in: The mutated gene; Original genes; () has a mean of 0 and a variance of The Gaussian distribution.

[0105] Variation intensity Adaptive adjustment with each generation of evolution:

[0106] in: (g) represents the mutation intensity of the g-th generation; λ is the initial mutation intensity; λ is the decay coefficient; G is the maximum number of generations.

[0107] After a set number of generations of evolution, the algorithm outputs the individual with the highest fitness as the optimal parameter combination.

[0108] Deep neural networks are used to establish a non-linear mapping relationship between parameters and accuracy. The network adopts a fully connected structure, containing an input layer, multiple hidden layers, and an output layer. The input layer receives a standardized feature vector, and the output layer outputs the accuracy prediction value.

[0109] During the forward propagation process, the output of each layer is calculated using the following formula: in: This is the output of the l-th layer; Here is the weight matrix for the l-th layer; This is the output of the (l-1)th layer; σ is the bias vector of the l-th layer; σ is the activation function.

[0110] The activation function uses the ReLU function:

[0111] Where: x is the input; max is the function to find the maximum value.

[0112] The network training uses mean squared error as the loss function: Where: L is the loss function; N is the number of training samples; This is a predicted value; This is the actual value.

[0113] Weight updates are performed using gradient descent.

[0114] in: The updated weights; η represents the current weights; η is the learning rate. This represents the gradient of the loss function with respect to the weights.

[0115] The learning rate uses an adaptive adjustment strategy, decreasing with each training epoch.

[0116] Where: η(t) is the learning rate in the t-th round; α is the initial learning rate; α is the decay coefficient; t is the number of training rounds.

[0117] The gradient is calculated using the backpropagation algorithm:

[0118] in: / Let be the partial derivative of the loss function with respect to the weights of the l-th layer; / Let be the partial derivative of the loss function with respect to the output of the l-th layer; / This is the partial derivative of the output of the l-th layer with respect to the weights.

[0119] The real-time data acquisition module is responsible for acquiring multi-dimensional status information of components during the production process. This module works in concert with multiple sensors to achieve comprehensive monitoring of the component's shape and internal state.

[0120] Laser scanning equipment uses the principle of line structured light, reconstructing three-dimensional shapes by projecting laser stripes and capturing the deformed image. Based on the principle of triangulation, the coordinates of a point in space are calculated using the following formula: Where: z is the depth coordinate; f is the camera focal length; b is the baseline length; d is the parallax.

[0121] The horizontal and vertical coordinates are calculated using the camera imaging model: Where: x and y are the horizontal and vertical coordinates; u and v are the pixel coordinates; , z is the principal point coordinate; z is the depth; f is the focal length.

[0122] Scan density verification ensures sufficient point cloud data is acquired:

[0123] in: For scan density; This represents the number of point clouds; The scanned area is [area].

[0124] Point cloud preprocessing includes denoising, downsampling, and coordinate transformation. Denoising employs statistical filtering to remove outliers.

[0125] in: For the set of outliers; Let i be the i-th point; The distance from the point to the centroid of its neighborhood; The mean distance; denoted as the standard deviation of the distance; k is the threshold coefficient.

[0126] Downsampling uses a voxel grid method to unify the point cloud density: in: V represents the downsampled points; V is a voxel; |V| is the number of points within a voxel.

[0127] The temperature sensor uses a PT100 platinum resistance thermometer, and its resistance value is related to temperature as follows: in: The resistance value at temperature T; A represents the resistance at 0℃, with a standard value of 100Ω; A is the primary temperature coefficient, with a value of 3.9083 × 10⁻³℃⁻¹; B is the secondary temperature coefficient, with a value of -5.775 × 10⁻¹. 7 ℃⁻².

[0128] Temperature sensors are arranged according to a spatial grid to ensure the capture of the spatial distribution of the temperature field. The sensor positions are determined based on the finite element mesh nodes, and the arrangement density is proportional to the temperature gradient. For surface areas with large gradients, the sensors are arranged more densely; for interior areas with relatively uniform temperatures, the sensors are arranged more sparsely.

[0129] The humidity sensor uses a capacitive principle, and the relationship between relative humidity and capacitance value is as follows: Where: C is the capacitance value; The reference capacitor; This is the primary humidity coefficient; RH is the secondary humidity coefficient; RH is the relative humidity.

[0130] The capacitance value is converted into a voltage signal by a measuring circuit.

[0131] in: C is the output voltage; C is the sensor capacitance. Reference capacitor; This is the reference voltage.

[0132] Fiber Bragg grating sensors are used to measure strain, and their working principle is based on Bragg reflection: in: For the Bragg wavelength; Λ represents the effective refractive index; Λ represents the grating period.

[0133] When the component experiences strain, the grating period changes, causing a Bragg wavelength shift: Where: Δ For wavelength variation; The initial wavelength; ε is the effective photoelastic coefficient; ε is the strain.

[0134] Temperature compensation for strain is achieved using the dual-grating method: in: For practical response; For strain grating wavelength variation; To compensate for temperature-induced grating wavelength changes.

[0135] The vibration sensor uses a piezoelectric accelerometer, and its output voltage is proportional to the acceleration. Where: V is the output voltage; 'a' is the sensitivity coefficient; 'a' is the acceleration. This is the bias voltage.

[0136] The quality of compaction is evaluated by analyzing the spectral characteristics of the acceleration signal: in: A(f) represents the vibration quality index; A(f) is the spectrum of the acceleration signal. , This refers to the effective frequency range.

[0137] The digital twin mapping module integrates real-time collected physical data with the BIM model to construct a digital twin that is updated synchronously with the physical components.

[0138] Point cloud registration is a crucial step in achieving data fusion. The registration process consists of two stages: coarse registration and fine registration. Coarse registration achieves initial alignment through feature matching. Feature point extraction employs the curvature thresholding method. Where: F is the set of feature points; Points in a point cloud; For point The curvature; This is the curvature threshold.

[0139] Curvature was calculated using analysis of covariance:

[0140] Where: κ is the curvature; , , These are the three eigenvalues ​​of the covariance matrix.

[0141] The feature descriptor uses Fast Point Feature Histogram (FPFH): Where: FPFH(p) is the feature descriptor of point p; N(p) is the neighborhood of p; q represents the weights; SPFH(q) is the simplified point feature histogram.

[0142] Feature matching employs nearest neighbor search, with paired feature points used to compute the initial transformation matrix. Fine registration utilizes the Iterative Closest Point (ICP) algorithm, which solves for the optimal transformation by minimizing the point-to-point distance error. Where: R is the rotation matrix; t is the translation vector; As weight; As the source point; is the target point; N is the number of corresponding point pairs.

[0143] The weights are adaptively adjusted based on the distance between point pairs: in: Let be the weight of the i-th pair of points; The distance between points; This represents the standard deviation of the distance.

[0144] Registration convergence criterion:

[0145] in: Let be the transformation matrix for the k-th iteration; ‖·‖F is the Frobenius norm; This is the convergence threshold.

[0146] After registration, calculate the normal deviation for each measuring point:

[0147] in: For normal deviation; These are the actual measurement points; The closest point on the model; It is the normal vector.

[0148] The spatial distribution of the deviation is extended from discrete measurement points to the entire component surface using Kriging interpolation. Kriging interpolation is based on the spatial correlation assumption, which holds that points that are closer together have more similar deviations. A semi-variogram describes this spatial correlation:

[0149] Where: γ(h) is the semivariogram; h is the lag distance; Let δ be the number of point pairs with a distance of h; ) represents the deviation of position xi.

[0150] The semivariogram is typically fitted using a spherical model: in: C is the nugget value; C is the base value; a is the range.

[0151] Kriging weights are obtained by solving a system of linear equations: in: The semivariogram values ​​between measurement points i and j; The semivariogram value between measurement point i and the prediction point; is the Kriging weight; μ is the Lagrange multiplier.

[0152] Interpolation bias is calculated as follows: Where: δ(x0) is the deviation of the predicted point x0; For weights; δ( ) represents the deviation of the measuring point xi.

[0153] Interpolation variance is used to assess the uncertainty of predictions. Where: σ²(x0) is the prediction variance; As weight; is the semivariogram value; μ is the Lagrange multiplier.

[0154] The interpolation of the temperature and humidity fields uses an inverse distance weighting method, which is computationally simple and suitable for regularly distributed sensor data. Where: T(x) is the interpolated temperature at position x; The temperature at measurement point i; As weight.

[0155] The weight is inversely proportional to the power of the distance:

[0156] in: As weight; denoted as distance; p is the exponent, usually taken as 2.

[0157] To avoid singularity when the distance is zero, a modified weighting function is used:

[0158] Where ε is a small positive number to prevent division by zero errors.

[0159] The reconstruction of the strain field considers the tensor properties of strain and employs the moving least squares method: Where: ε(x) is the strain at position x; p(x) is the basis function vector; a(x) is the coefficient vector.

[0160] The coefficients are obtained by minimizing the weighted error: Where: a(x) is the coefficient; For weighting functions; For measuring the strain at the point; p( ) represents the basis function at the measurement point.

[0161] The update frequency of the digital twin is dynamically adjusted according to the production stage. During the pouring stage, when changes are drastic, the update frequency is set to once per minute; in the early curing stage, the update frequency is once per hour; and in the later curing stage, when changes are slower, the update frequency decreases to once every 4 hours. The update strategy combines event-driven and timed updates. in: For the update time; The update time is scheduled. This refers to the moment the event is triggered.

[0162] Event triggering conditions include, but are not limited to, deviations exceeding a threshold, sudden temperature changes, and abnormal vibrations. Where: Event is the triggering event; The maximum deviation; ΔT is the deviation threshold; ΔT is the temperature change; Δ The temperature change threshold; It is the vibration acceleration; This is the acceleration threshold.

[0163] The accuracy evaluation output module performs a comprehensive quality evaluation after the component production is completed. The evaluation is based on six geometric accuracy indicators, each with a clear measurement method and evaluation standard.

[0164] Length deviation is obtained by measuring the distance between the center points of the two end faces of the component: Where: ΔL is the length deviation; , , ), ( , , () represents the coordinates of the centers of the two end faces; For the design length.

[0165] The center of the end face is determined by fitting a plane using the least squares method: Where: a, b, c, and d are plane parameters; , , () represents a point on the end face.

[0166] Width and thickness deviations are measured using a similar method. Diagonal difference reflects the squareness of the component. Where: ΔD is the difference between the diagonals; , The lengths of the two diagonals.

[0167] Flatness was measured using a 2-meter straightedge method. Multiple test locations were selected on the component surface, and the maximum gap between the straightedge and the surface was recorded. Where: P represents flatness; Let be the gap at the i-th measuring point; The number of measurement points; max indicates taking the maximum value.

[0168] Verticality reflects the degree of tilt of a component: Where: V is verticality, in mm / m; Δx and Δy are the offsets of the top relative to the bottom; H ​​is the height of the component.

[0169] The individual accuracy score is calculated based on the ratio of deviation to tolerance:

[0170] in: The deviation ratio; This represents the measured deviation. To allow for deviation.

[0171] The scoring function is in piecewise linear form: The overall score is calculated by weighted summation:

[0172] in: For the overall score; Let be the weighting coefficients, satisfying Σ =1.

[0173] The weighting coefficients are determined based on the importance of each indicator. Generally, the weights of length, width, and thickness deviations are relatively large, followed by flatness and verticality, while the weight of diagonal difference is relatively small.

[0174] The quality level is determined based on the overall score: The generated quality inspection report includes a 3D deviation cloud map, statistical charts of various accuracy indicators, and a quality grade certificate. The deviation cloud map uses color coding to visually display the deviation distribution: blue indicates negative deviation, green indicates acceptable areas, and red indicates positive deviation.

[0175] The evaluation results and data from the entire production process form a complete quality record, which is stored in a historical database. This data is used to update predictive models, optimization algorithms, and knowledge bases, enabling continuous system improvement.

[0176] Combination Figure 2 The system's workflow includes: The first step is that the multiphysics accuracy prediction module reads the BIM model data and material parameters, predicts the component deformation by solving the four-field coupling equation system, and generates template compensation parameters. The second step is for the precision gene map construction module to extract the feature parameters of the current component, construct a feature vector, and store it in the knowledge base. The third step is for the intelligent optimization decision-making module to search for similar cases in the knowledge base and determine the optimal process parameters through a genetic algorithm. Fourth, the real-time data acquisition module acquires point cloud and sensor data according to a preset strategy; Fifth step: The digital twin mapping module registers the collected data with the BIM model to generate a digital twin. When the deviation exceeds the threshold, it returns to the third step for parameter adjustment. The sixth step involves the accuracy evaluation output module calculating the accuracy index and assessing the quality level, then feeding the results back to the knowledge base in the second step.

[0177] Through the cyclical execution of the above six steps, the system achieves a complete closed loop from prediction to evaluation. Step six feeds back to step two to update the knowledge base, and step five triggers step three to adjust parameters. The collaborative work of each module ensures effective control of accuracy.

[0178] In summary, this invention achieves precision control throughout the entire lifecycle of precast components through the collaborative work of six modules: multiphysics field coupling prediction, precision gene map construction, intelligent parameter optimization, real-time data acquisition, digital twin mapping, and precision evaluation feedback. It transforms precision control from post-inspection to prediction, monitoring, and regulation throughout the entire process, effectively solving the problem of precision control for precast components.

[0179] The above description is only 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 within the protection scope of the present invention.

Claims

1. A prefabrication accuracy control system for concrete structural components based on BIM and digital twins, characterized in that, include: The multiphysics field accuracy prediction module is used to read the component's geometric parameters and reinforcement parameters from the BIM model, read the mix proportion parameters from the material database, establish a set of coupled equations for temperature field, humidity field, stress field and chemical field, solve the predicted shrinkage strain and warping deformation of the component at different ages using the finite element method, and generate inverse compensation amounts in the length, width and thickness directions based on the predicted deformation. The precision gene map construction module is used to extract material features, process features, environmental features and structural features to construct multi-dimensional feature vectors. It reduces dimensionality through principal component analysis, establishes a precision classification system using clustering algorithms, and uses association rules to mine the correspondence between parameters and precision. The intelligent optimization decision-making module is used to calculate the feature similarity between the current component and historical cases, select similar cases to determine the range of process parameters, use a genetic algorithm to search for the optimal parameter combination, establish a deep neural network to predict the accuracy deviation, and generate a graded control scheme. The real-time data acquisition module is used to acquire surface point clouds through laser scanning and internal state parameters through temperature, humidity, strain, and vibration sensors, and dynamically adjusts the acquisition frequency according to the production stage. The digital twin mapping module is used to register point clouds with BIM models using the iterative nearest point algorithm, calculate normal deviation, generate continuous field distribution through interpolation algorithm, and construct synchronously updated digital twins. The accuracy evaluation output module is used to calculate six indicators: length deviation, width deviation, thickness deviation, diagonal difference, flatness, and perpendicularity, to evaluate the quality level, generate an inspection report, and update the knowledge base.

2. The system according to claim 1, characterized in that, In the multiphysics field accuracy prediction module, the temperature field is solved by the heat conduction equation, and the hydration heat generation rate is proportional to the product of cement dosage and hydration degree change rate; the humidity field is solved by the diffusion equation, and the diffusion coefficient is related to the aggregate parameters; the stress field is established by the constitutive equation, and the stiffness matrix takes into account the reinforcement ratio; the chemical field is described by the hydration kinetic equation, and the activation energy is linearly related to the water-cement ratio.

3. The system according to claim 1, characterized in that, The precision gene map construction module performs principal component analysis through eigenvalue decomposition of the covariance matrix, selects principal components whose cumulative contribution rate reaches a set threshold, uses the K-means algorithm to minimize the sum of squares within a class for clustering, and establishes association rules based on three indicators: support, confidence, and lift.

4. The system according to claim 1, characterized in that, The genetic algorithm of the intelligent optimization decision-making module uses real-number encoded chromosomes, and the fitness function is the product of the inverse of the precision deviation and the time cost penalty term. It is iteratively optimized through roulette wheel selection, arithmetic crossover, and Gaussian mutation.

5. The system according to claim 1, characterized in that, The laser scanning of the real-time data acquisition module is performed at four points: completion of casting, demolding, curing, and finished product delivery; temperature sensors are arranged in a spatial grid; and fiber optic grating sensors measure strain through Bragg wavelength drift.

6. The system according to claim 1, characterized in that, The point cloud registration of the digital twin mapping module includes coarse registration based on curvature features and fine registration based on minimizing point-to-point distance; the bias field is generated by Kriging interpolation, and the weights are determined by a semi-variogram function; the temperature and humidity field is generated by inverse distance weighted interpolation.

7. The system according to claim 1, characterized in that, The accuracy evaluation output module calculates the individual scores using a piecewise linear function. A full score is awarded when the deviation ratio is less than 0.3, decreasing linearly between 0.3 and 1.0, and continuing to decrease between 1.0 and 2.

0. A score of zero is awarded when the deviation ratio is greater than 2.

0. The overall score is the weighted sum of the scores for each item.

Images