CN-121981455-A - Digital twinning-based full-flow intelligent management and control method and system for prefabricated capping beam construction

CN121981455ACN 121981455 ACN121981455 ACN 121981455ACN-121981455-A

Abstract

The application relates to the technical field of digital twin and civil engineering, and provides an assembled bent cap prefabrication whole-flow intelligent management and control method and system based on digital twin, wherein the method comprises the steps of inputting the prefabricated construction requirement of a bent cap, converting the prefabricated construction requirement into semantic feature vectors, and generating comprehensive semantic feature vectors based on the semantic feature vectors corresponding to different types of construction requirements; the method comprises the steps of mapping comprehensive semantic feature vectors through a first graph neural network to obtain technological parameters, simulating and outputting construction result description indexes in a capping beam prefabricated digital twin three-dimensional scene based on the technological parameters, mapping the construction result description indexes into simulated semantic feature vectors, calculating error vectors of the simulated semantic feature vectors and the comprehensive semantic feature vectors, outputting parameter adjustment amounts through an optimization strategy model based on the error vectors and the technological parameters, judging whether stopping optimization conditions are met or not according to the error vectors and the adjustment amounts, and outputting optimized technological parameters if the stopping optimization conditions are met. The application realizes the efficient optimization of the capping beam prefabrication construction and improves the capping beam prefabrication effect.

Inventors

NAN NAN
XIE ZHANGHUI
LI ZHIMING
XIAO YI

Assignees

湾区(广东)建筑装配科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260115

Claims (9)

1. The digital twinning-based full-flow intelligent management and control method for prefabricated capping beam construction is characterized by comprising the following steps of: Inputting construction requirements of cap girder prefabrication, converting the construction requirements into semantic feature vectors, and generating comprehensive semantic feature vectors based on the semantic feature vectors generated by different types of input; mapping the comprehensive semantic feature vector in a first graph neural network to obtain a capping beam prefabricated process parameter; Carrying out construction process simulation on the technological parameters in a preset digital twin three-dimensional scene of the bent cap prefabrication, and outputting construction result description indexes; Mapping the construction result description index to a semantic space to obtain a simulation semantic feature vector, and calculating the difference between the simulation semantic feature vector and the comprehensive semantic feature vector to obtain an error vector; Establishing an optimization strategy model, inputting the semantic error vector and the technological parameters, and outputting the adjustment quantity of the technological parameters; judging whether the preset stop optimization condition is met or not based on the semantic error vector and the adjustment quantity of the process parameters, and if yes, outputting the optimized process parameters.
2. The digital twinning-based assembled bent cap prefabrication construction full-flow intelligent control method according to claim 1, wherein the first graph neural network is obtained through training: Constructing a bipartite graph G= (V, E), wherein the vertex set V is divided into a semantic node subset Vs and a process parameter node subset Vp, and the edge set E only exists between nodes of Vs and Vp; Initializing a node characteristic value for each semantic node Vs epsilon Vs, wherein the characteristic value of the semantic node corresponds to one dimension of the comprehensive semantic characteristic vector; initializing node characteristic values for each process parameter node VP epsilon VP; A plurality of iterations of messaging are performed, in each iteration: 1) For each process parameter node vp, aggregating messages transmitted by all adjacent semantic nodes; 2) Updating the characteristic value of the technological parameter node according to the aggregated information; the characteristic value of the process parameter node after the last iteration is used as the predicted process parameter to be output; and calculating the loss between the predicted process parameter and the real process parameter, and updating the weight parameter involved in the message transmission process through a back propagation algorithm based on the loss until the training stop condition is met.
3. The digital twinning-based assembled bent cap prefabrication construction full-flow intelligent management and control method according to claim 1, wherein the loss function of the first graph neural network is as follows: ; where L is the total loss value, m is the total number of training samples, k is the sample index, A predicted process parameter vector for the kth sample, Is the actual process parameter vector for the kth sample, For the weight regularization term, W is a weight matrix, Is a regularization coefficient.
4. The digital twinning-based assembled bent cap prefabrication whole-flow intelligent control method is characterized in that the bent cap prefabrication digital twinning three-dimensional scene is constructed by fusing a BIM design model of the bent cap, field three-dimensional scanning data, a material attribute library and historical construction data.
5. The digital twinning-based full-flow intelligent control method for prefabricated capping beam construction of claim 1, wherein the technological parameters comprise one or more of concrete mixing ratio, vibration frequency and duration, steam curing temperature rise/drop rate, dead time, prestress tensioning force and sequence and hoisting path planning.
6. The digital twinning-based full-flow intelligent control method for prefabricated bent cap prefabrication construction of claim 1, wherein the construction result description index comprises one or more of a rigidity index, a strength index and an anti-cracking index.
7. The digital twinning-based assembled bent cap prefabrication construction full-flow intelligent control method according to claim 1, wherein the step of mapping the construction result description index to a semantic space to obtain a simulation semantic feature vector comprises the following steps: constructing a second graph neural network, wherein the second graph neural network comprises semantic label nodes and physical index nodes, and a semantic concept is determined by a plurality of physical indexes together; Training the second graph neural network by adopting historical data, and mapping the construction result description index to a semantic space by utilizing the trained second graph neural network to obtain a simulation semantic feature vector.
8. The digital twinning-based assembled bent cap prefabrication construction full-flow intelligent control method according to claim 1, wherein the stop optimization condition is as follows: Or (b) Or (b) ; Wherein, the ; ; ; Wherein, the Is the mean value of the error of the last w rounds, Is the error convergence threshold value and, Is the minimum valid update threshold value, Importance weights representing semantic feature dimensions i, Representing the error corresponding to the semantic feature dimension i of the current round, n represents the total number of semantic feature dimensions, Representing the average parameter adjustment amplitude of the current wheel, The current j-th dimension technological parameter gradient adjustment quantity, s is the total number of technological parameter dimensions; A variance representing the trend of the error variation; Is a trend stability threshold.
9. Digital twinning-based full-flow intelligent management and control system for prefabricated capping beam construction, which is characterized by comprising: the input module is used for inputting construction requirements of cap beam prefabrication, converting the construction requirements into semantic feature vectors, and generating comprehensive semantic feature vectors based on the semantic feature vectors generated by different types of input; The first mapping module is used for mapping the comprehensive semantic feature vector in a first graph neural network to obtain a capping beam prefabricated process parameter; The simulation module is used for simulating the construction process of the process parameters in a preset digital twin three-dimensional scene of the bent cap prefabrication and outputting construction result description indexes; the second mapping module is used for mapping the construction result description index to a semantic space to obtain a simulation semantic feature vector, and calculating the difference between the simulation semantic feature vector and the comprehensive semantic feature vector to obtain an error vector; The establishing module is used for establishing an optimization strategy model, inputting the semantic error vector and the technological parameters and outputting the adjustment quantity of the technological parameters; And the output module is used for judging whether the preset stop optimization condition is met or not based on the semantic error vector and the adjustment quantity of the process parameters, and if yes, outputting the optimized process parameters.

Description

Digital twinning-based full-flow intelligent management and control method and system for prefabricated capping beam construction Technical Field The application relates to the technical field of digital twinning and civil engineering, in particular to an assembled bent cap prefabrication whole-flow intelligent management and control method and system based on digital twinning. Background In bridge engineering construction, the capping beam is used as an important bearing member, and the prefabrication construction quality of the capping beam directly influences the safety and durability of the whole structure. The traditional capping beam prefabrication process relies on manual experience to set parameters and control the process, and has the following problems: (1) The technological parameters depend on experience, namely constructors set parameters such as template installation, concrete proportioning, maintenance conditions and the like according to experience, so that scientificization and standardization are difficult to realize; (2) The adjustment efficiency is low, once quality deviation occurs, repeated test adjustment is needed, the period is long, and the cost is high; (3) The lack of closed loop feedback, namely the lack of a quantitative feedback mechanism between the construction process and the design intention, is difficult to realize dynamic optimization; (4) The traditional system can not understand the construction requirement in the modes of text description, drawing, field image and the like, and can not convert the construction requirement. Disclosure of Invention Aiming at the technical problems, the application aims to provide an assembled bent cap prefabrication whole-flow intelligent management and control method and system based on digital twinning, and aims to solve the technical problems. In a first aspect, an embodiment of the present application provides a digital twinning-based full-flow intelligent management and control method for prefabricated capping beams, where the method includes: Inputting construction requirements of cap girder prefabrication, converting the construction requirements into semantic feature vectors, and generating comprehensive semantic feature vectors based on the semantic feature vectors generated by different types of input; mapping the comprehensive semantic feature vector in a first graph neural network to obtain a capping beam prefabricated process parameter; Carrying out construction process simulation on the technological parameters in a preset digital twin three-dimensional scene of the bent cap prefabrication, and outputting construction result description indexes; Mapping the construction result description index to a semantic space to obtain a simulation semantic feature vector, and calculating the difference between the simulation semantic feature vector and the comprehensive semantic feature vector to obtain an error vector; Establishing an optimization strategy model, inputting the semantic error vector and the technological parameters, and outputting the adjustment quantity of the technological parameters; judging whether the preset stop optimization condition is met or not based on the semantic error vector and the adjustment quantity of the process parameters, and if yes, outputting the optimized process parameters. Further, the first graph neural network is obtained through training of the following steps: Constructing a bipartite graph G= (V, E), wherein the vertex set V is divided into a semantic node subset Vs and a process parameter node subset Vp, and the edge set E only exists between nodes of Vs and Vp; Initializing a node characteristic value for each semantic node Vs epsilon Vs, wherein the characteristic value of the semantic node corresponds to one dimension of the comprehensive semantic characteristic vector; initializing node characteristic values for each process parameter node VP epsilon VP; A plurality of iterations of messaging are performed, in each iteration: 1) For each process parameter node vp, aggregating messages transmitted by all adjacent semantic nodes; 2) Updating the characteristic value of the technological parameter node according to the aggregated information; the characteristic value of the process parameter node after the last iteration is used as the predicted process parameter to be output; and calculating the loss between the predicted process parameter and the real process parameter, and updating the weight parameter involved in the message transmission process through a back propagation algorithm based on the loss until the training stop condition is met. Further, the loss function of the first graph neural network is: ; where L is the total loss value, m is the total number of training samples, k is the sample index, A predicted process parameter vector for the kth sample,Is the actual process parameter vector for the kth sample,For the weight regularization term, W is a weight matrix,Is a regu