Mechanical response prediction and dynamic damage tracking method of semi-confined space building structures

Abstract

The present invention discloses a method for predicting the mechanical response of a semi-enclosed space building structure and for tracking dynamic damage. The method is based on the urban building digital twin technology, combined with a numerical simulation method, to simulate the structural mechanical response, and to construct an initial structural mechanical response database; to construct a spatiotemporal state matrix, and to use a sequential Monte Carlo method, combined with knowledge constraint optimization to achieve future state simulation, and to construct a dynamic structural mechanical response database; to construct a structural mechanical response prediction network, and to combine the dynamic structural mechanical response database to construct a structural mechanical response model through data training and optimization; to construct a service life prediction model, and to use the structural mechanical response model, combined with a damage determination method and a data dynamic reconstruction method, to achieve dynamic damage tracking of semi-enclosed space buildings. The method of the present invention has high efficiency, strong generalization ability, and high precision.

Description

Technical Field

[0001] The invention belongs to the technical field of data processing, and in particular to a method for predicting mechanical response of a semi-enclosed space building structure and for tracking dynamic damage. Background Art

[0002] A semi-enclosed space building refers to a space building that is not completely closed, has one or more sides that are relatively open or semi-open, and has limited air exchange with the external environment, such as tunnels, cabins, underground spaces, etc.

[0003] Dynamic damage refers to the damage caused to an object when it is subjected to dynamic loads. It is a time-varying process. The continuous influence of dynamic forces will cause dynamic changes in stress and strain inside the building structure. When these changes exceed the bearing capacity of the building materials or structures, damage will occur, and the damage will continue to develop with the continuous action of dynamic forces.

[0004] At present, for dynamic damage tracking, as described in the paper "Resilient analysis on tunnel structural serviceability based on lifetime dynamic prediction model" (Dawei Huang, Hehua Zhu, Yi Shen, Zhiguo Yan, Jiann-Wen Woody Ju, Resilient analysis on tunnel structural serviceability based on lifetime dynamic prediction model, Tunnelling and Underground Space Technology, ISSN 0886-7798, https: / / doi.org / 10.1016 / j.tust.2022.104690.), a tunnel structural serviceability elasticity analysis method based on a lifetime dynamic prediction model is disclosed, including: predicting tunnel performance as a combination of long-term aging and short-term events, evaluating long-term aging with a statistical probability model, and dividing the system state with damage indicators; quantifying the performance changes of short-term events with the concept of elasticity, and judging them in stages and based on thresholds. Based on the failure mode and effect analysis, the life performance model is established, the key defects are determined, the displacement settlement is calculated by examples, the threshold is established, the defect elastic process is evaluated, the defect curve is adjusted in combination with the event impact, the life prediction is re-evaluated with the relevant model, and the model is corrected according to the threshold and performance loss classification to achieve dynamic prediction. However, the above method assumes that all conditional factors are independent variables to construct the structural equation to predict the system performance, but in actual tunnel engineering, many factors are interrelated and affect each other, such as the complex coupling relationship between geological conditions and defects such as tunnel settlement and leakage. This assumption may not accurately reflect the actual situation, which makes the model in actual application, may need additional adjustment or correction to adapt to the actual coupling effect, and the analysis is not efficient and accurate. In addition to geological conditions, environmental factors around the tunnel, such as groundwater dynamic changes, vibration of surrounding buildings, etc., will also affect the performance of the tunnel structure, but the model may not fully incorporate the long-term dynamic change process of these factors and their impact on the extrapolated prediction, resulting in the model's weak extrapolation ability.

[0005] In summary, the problems existing in the existing technology are: low efficiency, low accuracy and weak model extrapolation capability in tracking dynamic damage of semi-enclosed space buildings. Summary of the invention

[0006] The purpose of the present invention is to provide a method for predicting mechanical response of semi-enclosed space building structures and dynamic damage tracking, which has high efficiency, strong generalization ability and high precision.

[0007] The technical solution to achieve the purpose of the present invention is:

[0008] On the one hand, a method for predicting the structural mechanical response of a semi-enclosed space building based on a structural mechanical response model is provided, comprising:

[0009] Based on the urban building digital twin technology and combined with numerical simulation methods, structural mechanics response simulation is carried out to build an initial structural mechanics response database;

[0010] Construct a structural mechanics response prediction network, combine it with the initial structural mechanics response database, and construct a structural mechanics response model through data training and optimization;

[0011] Based on the structural mechanics response model, the structural mechanics response of semi-enclosed space buildings under long-term service is predicted.

[0012] Preferably, the initial structural mechanics response database construction step includes:

[0013] Use urban building digital twin technology to generate a real semi-enclosed space digital twin with external environment;

[0014] Obtain the internal structure information of the real semi-confined space digital twin through the semi-confined space building design specifications;

[0015] Based on the neural core surface reconstruction algorithm, a digital twin of a semi-enclosed space area is constructed;

[0016] Combine multiple digital twins of semi-confined space areas to construct a digital twin dataset of semi-confined space areas;

[0017] Construct a digital twin neural network, use the semi-confined space area digital twin dataset for training and optimization, and generate a digital twin generation model;

[0018] With the help of digital twin generation model, the digital twin of the semi-enclosed space area with internal structure and external environment is obtained;

[0019] Through the physical connection network of building monomers, the finite element structural mechanics analysis method is used to construct the mechanical property correlation network of building monomers;

[0020] With the help of the mechanical property association network of the building monomer and considering the local substructure pulse excitation correction, the structural mechanics response simulation is carried out, and an initial structural mechanics response database is constructed with the long-term service conditions of the building monomer as input and the structural response of the building monomer under long-term service as output.

[0021] Preferably, the data features in the digital twin data set of the semi-enclosed space area are: semi-enclosed space building number, area size, semi-enclosed space building type, semi-enclosed space building length, semi-enclosed space building width, semi-enclosed space building height, semi-enclosed space building node number, semi-enclosed space building node X-axis position, semi-enclosed space building node Y-axis position, semi-enclosed space building node Z-axis position;

[0022] The data features in the initial structural mechanics response database are: semi-confined space building number, service time of the semi-confined space building at the current moment, semi-confined space building type, semi-confined space building environment, semi-confined space building quality, semi-confined space building usage intensity, semi-confined space building material characteristics, semi-confined space building node number, whether the semi-confined space building node belongs to the structured grid, the position of the semi-confined space building node on the X-axis, the position of the semi-confined space building node on the Y-axis, the position of the semi-confined space building node on the Z-axis, the external force on the semi-confined space building node on the X-axis, the external force on the semi-confined space building node on the Y-axis, the external force on the semi-confined space building node on the Z-axis, the stress of the semi-confined space building node, the strain of the semi-confined space building node, and the displacement of the semi-confined space building node.

[0023] Preferably, the digital twin neural network construction steps include:

[0024] The input features in the digital twin data set of the semi-enclosed space area are used as the input of the first multi-layer perceptron neural network, and the output features in the digital twin data set of the semi-enclosed space area are used as the output of the first multi-layer perceptron neural network to construct a generator;

[0025] The discriminator is constructed by taking the output features in the digital twin dataset of the semi-enclosed space area as the input of the second multi-layer perceptron neural network and taking the corresponding authenticity probability as the output of the second multi-layer perceptron neural network;

[0026] The generator and discriminator constitute a generative adversarial network, also known as a digital twin neural network.

[0027] Preferably, it also includes:

[0028] Based on the initial structural mechanics response database, a spatiotemporal state matrix is ​​constructed;

[0029] Use the sequential Monte Carlo method combined with knowledge-constrained optimization to achieve future state simulation and build a dynamic structural mechanics response database;

[0030] Build a structural mechanics response prediction network, combine it with the dynamic structural mechanics response database, and build a structural mechanics response model through data training and optimization.

[0031] Preferably, the step of constructing a dynamic structural mechanics response database includes:

[0032] By sorting the node data of static reference under each time slice according to the time of the same semi-confined space building, and performing two-dimensional interpolation of time and space items, a spatiotemporal state matrix of the semi-confined space building under a relatively continuous time state is constructed; the spatiotemporal state matrix is ​​a data set number matrix with time items as matrix row features, space items as matrix column features, and semi-confined space building node numbers as matrix elements, wherein the space items are obtained according to the distance between the node position and the origin;

[0033] Using the statistical analysis library, the spatiotemporal state matrix is ​​fitted and distributed according to different time dimensions to obtain the parameters of each distribution function;

[0034] Through the expert scoring method and the data fitting model, a knowledge constraint neural network is constructed, and the knowledge constraint neural network is embedded in the regularization term calculation of the sequential Monte Carlo to achieve the combination of sequential Monte Carlo and knowledge constraint optimization, and to build a model combining sequential Monte Carlo and knowledge constraint optimization.

[0035] Based on the parameters of each distribution function obtained, the number of sampling and the step size are set, and by building a model combining sequential Monte Carlo and knowledge-constrained optimization, various possible future states are simulated to achieve future state simulation;

[0036] The simulated future state is used as a new data feature to be spliced ​​into the data features in the space-time state matrix to construct a dynamic structural mechanics response database.

[0037] Preferably, the structural mechanics response prediction network includes a structural mechanics response prediction network encoder, a structural mechanics response prediction network processor and a structural mechanics response prediction network decoder;

[0038] In the structural mechanics response prediction network encoder: the third multi-layer perceptron neural network is used to construct a node feature grid encoder and an edge feature grid encoder respectively, and a hierarchical coding strategy is introduced to form a grid encoder; the first implicit neural network is used to construct a node feature implicit encoder and an edge feature implicit encoder respectively, and a hierarchical coding strategy is introduced to form an implicit encoder; with the help of the first feedforward neural network, dynamic weights are given to the grid encoder and the implicit encoder respectively, and the grid encoder and the implicit encoder are weighted and fused to form a structural mechanics response prediction network encoder;

[0039] In the structural mechanics response prediction network processor: a node structured grid processor and an edge structured grid processor are constructed respectively by using the first three-dimensional convolutional neural network, and a structured grid processor is formed by using a hierarchical processing strategy; a node unstructured grid processor and an edge unstructured grid processor are constructed respectively by using the fourth multilayer perceptron neural network and the second implicit neural network connected in series, and an unstructured grid processor is formed by using a hierarchical processing strategy;

[0040] In the structural mechanics response prediction network decoder: a structured grid processor is constructed by introducing a second three-dimensional convolutional neural network with an attention mechanism combined with an average pooling operation, and an unstructured grid decoder is constructed by introducing a fifth multi-layer perceptron neural network with an attention mechanism combined with an average pooling operation; wherein, the input of the structured grid processor is the output of the structured grid processor, and the input of the unstructured grid decoder is the output of the unstructured grid processor.

[0041] In the second aspect, a method for dynamic damage tracking of semi-enclosed space buildings based on a structural mechanics response model is also provided, including:

[0042] Based on the second use of feed-forward neural network, a service life prediction model is constructed;

[0043] With the help of the structural mechanics response model mentioned above, combined with the damage determination method and the data dynamic reconstruction method, dynamic damage tracking of semi-enclosed space buildings can be achieved.

[0044] Preferably, the steps of dynamic damage tracing of semi-enclosed space buildings include:

[0045] Through real-time acquisition of real data of semi-confined space buildings, the initial / dynamic structural mechanics response database is corrected in real time;

[0046] Using the structural mechanics response model, based on the real-time corrected initial / dynamic structural mechanics response database, the prediction results of the corresponding structural mechanics response are obtained;

[0047] Combined with the damage determination method, node damage is quantified and the node damage quantification value is spliced ​​into the initial / dynamic structural mechanics response database after real-time correction;

[0048] With the help of the spliced ​​initial / dynamic structural mechanical response database, the future state of the semi-confined space building is simulated, the service life of the semi-confined space building is obtained, and a service life prediction data set is constructed;

[0049] Combined with the service life prediction data set, the second feedforward neural network is trained and optimized to build a service life prediction model;

[0050] With the help of the service life prediction model, the service life of semi-enclosed space buildings at any time can be obtained, the moment when the service life changes the most can be found, and then the service condition design that has the greatest impact on the service life can be obtained to achieve dynamic damage tracking.

[0051] Preferably, the data features of the service life prediction data set are: semi-confined space building number, service time of the semi-confined space building at the current moment, semi-confined space building type, semi-confined space building environment, semi-confined space building quality, semi-confined space building usage intensity, semi-confined space building material characteristics, and service life of the semi-confined space building.

[0052] Compared with the prior art, the present invention has the following significant advantages:

[0053] 1. Higher efficiency: Combining multi-layer perceptron neural network, implicit neural network and three-dimensional convolutional neural network to build encoder, processor and decoder, a structural mechanics response prediction network with efficient three-dimensional space-time physical field prediction capability is obtained. With the help of dynamic structural mechanics database, a structural mechanics response model is obtained. By building a service life prediction model and with the help of the structural mechanics response model, efficient dynamic damage tracking of semi-enclosed space buildings can be further achieved;

[0054] 2. Higher accuracy: Build a bidirectional digital twin generation model of internal structure and environment, build a mechanical property association network, build a space-time state matrix, build a dynamic structural mechanical response database with accurate topological relationships, mechanical properties, and space-time dynamic relationships, and further build a structural mechanical response model that considers topological relationships, mechanical properties, and space-time dynamic relationships. Combined with the service life prediction model, a precise prediction method for dynamic damage of semi-enclosed space buildings is constructed;

[0055] 3. Stronger extrapolation capability: Construct a structural mechanics response model that considers topological relationships, mechanical properties, and spatiotemporal dynamic relationships, construct a structural mechanics response neural network with powerful three-dimensional physical field learning and prediction capabilities, train and optimize, and obtain a structural mechanics response model with strong extrapolation capability. Further construct a semi-enclosed space dynamic damage tracking method with strong extrapolation capability.

[0056] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is the main flow chart of the dynamic damage tracing method of semi-enclosed space building based on the structural mechanics response model of the present invention.

[0058] Figure 2 yes Figure 1 Flowchart of the steps for constructing the initial structural mechanics response database.

[0059] Figure 3 This is a schematic diagram of the generative adversarial network architecture.

[0060] Figure 4 yes Figure 1 Flowchart of the steps for building a dynamic structural mechanics response database.

[0061] Figure 5 It is a schematic diagram of the space-time state matrix.

[0062] Figure 6 yes Figure 1 Flowchart of the steps in building the structural mechanics response model.

[0063] Figure 7 It is the network architecture diagram of the structural mechanics response prediction network encoder.

[0064] Figure 8 This is the network architecture diagram of the structural mechanics response prediction network processor, where (a) is a structured grid processor and (b) is an unstructured grid processor.

[0065] Fig. 9 It is the network architecture diagram of the structural mechanics response prediction network decoder, where (a) is the structured grid decoder and (b) is the unstructured grid decoder.

[0066] Fig.10 yes Figure 1 Flowchart of the steps for dynamic damage tracing in semi-confined space buildings. DETAILED DESCRIPTION

[0067] like Figure 1 As shown, the method for dynamic damage tracing of semi-enclosed space buildings based on the structural mechanics response model includes the following steps:

[0068] S10. Construction of initial structural mechanical response database: Construct a digital twin dataset of semi-enclosed space areas with internal structures, combine the digital twin neural network constructed by the generative adversarial network, construct an internal structure-environment bidirectional digital twin generation model, construct a mechanical property association network, combine numerical simulation methods, perform structural mechanical response simulation, and construct an initial structural mechanical response database.

[0069] like Figure 2 As shown, the steps of constructing the initial structural mechanics response database include:

[0070] S11. Construction of internal structure-environment bidirectional digital twin generation model: With the help of urban building digital twin technology, existing semi-enclosed space building design specifications, and neural core surface reconstruction algorithm, a digital twin dataset of semi-enclosed space areas with internal structures is generated, and a generative adversarial network is constructed to generate a digital twin neural network. Through data training and optimization, an internal structure-environment bidirectional digital twin generation model is generated.

[0071] The steps of constructing the internal structure-environment bidirectional digital twin generation model include:

[0072] S111. Construction of digital twin dataset: Use urban building digital twin technology to generate a real semi-enclosed space digital twin with an external environment. Use the existing semi-enclosed space building design specifications to obtain the internal structure information of the digital twin. Use the neural core surface reconstruction algorithm to improve the accuracy of the digital twin. Construct a digital twin dataset of a semi-enclosed space area with internal structure and external environment.

[0073] 1. Urban building digital twin technology: Use drones equipped with high-precision sensors to collect geospatial data, generate three-dimensional point cloud data and models, and import them into the GIS software platform for analysis and visualization. Use SFM and MVS algorithms to realize urban building digital twins.

[0074] 2. Existing semi-confined space building design specifications: Search for information and obtain existing semi-confined space building internal structure design specifications to obtain the internal structure information of the digital twin.

[0075] 3. Neural Kernel Surface Reconstruction Algorithm: The Neural Kernel Surface Reconstruction (NKSR) algorithm is a new algorithm launched by NVIDIA. It is mainly used to reconstruct high-fidelity three-dimensional models from large point cloud data, and is used to build digital twins of semi-enclosed space areas with internal structures and external environments.

[0076] 4. Digital twin dataset: Construct multiple digital twins of semi-confined space areas into a digital twin dataset of semi-confined space areas.

[0077] Features of the digital twin dataset: semi-enclosed space building number, area size, semi-enclosed space building type, semi-enclosed space building length, semi-enclosed space building width, semi-enclosed space building height, semi-enclosed space building node number, semi-enclosed space building node X-axis position, semi-enclosed space building node Y-axis position, semi-enclosed space building node Z-axis position; Among them, the input features of the data are semi-enclosed space building number, area size, semi-enclosed space building type, semi-enclosed space building length, semi-enclosed space building width, semi-enclosed space building height, semi-enclosed space building node number, the number of data input features of the dataset is 7, and the output features of the data are semi-enclosed space building node X-axis position, semi-enclosed space building node Y-axis position, The Z-axis position of the semi-confined space building node. The number of data output features of the data set is 3; the semi-confined space building number is used to distinguish building cases. The unit of area size is square kilometers. Different numbers in the semi-confined space building type represent different semi-confined space building types. For example, 0 represents a tunnel, 1 represents a cabin, and 2 represents a basement. The size of the semi-confined space building is represented by the length of the semi-confined space building, the width of the semi-confined space building, and the height of the semi-confined space building. The node number of the semi-confined space building represents the node number of this data. The location information of the node consists of the X-axis position of the semi-confined space building node in the world coordinate system with the center of the area as the origin, the Y-axis position of the semi-confined space building node, and the Z-axis position of the semi-confined space building node.

[0078] For example, [0, 50, 0, 5000, 2.5, 5, 0, 1250, 0, 0] represents a semi-enclosed space building numbered 0, whose area size is 50 square kilometers. The semi-enclosed space building is a tunnel, which is 5000m long, 2.5m wide, and 5m high. This data represents the data of the node numbered 0, and the node is located at (1250m, 0m, 0m).

[0079] S112, digital twin neural network construction: using data interpolation, expand the digital twin data set case, generate a new digital twin data set, multilayer perceptron neural network 1, construct the new digital twin data set input as the neural network input, the new digital twin data set output as the neural network output, construct the generator, multilayer perceptron neural network 2, the new digital twin data set output as the neural network input, the one-dimensional scalar, that is, the authenticity probability of the digital twin data as the output, construct the discriminator, generate the adversarial network, such as Figure 3 As shown, the construction of digital twin neural network is realized.

[0080] 1. Multilayer perceptron neural network 1: Multilayer perceptron neural network 1 is constructed by 1 input layer, 3 hidden layers, and 1 output layer. The number of neurons in the input layer is the number of input features of the new digital twin dataset. The number of neurons in the three hidden layers is the same, which is a number greater than 64 and a multiple of 2. The number of neurons in the output layer is the number of output features of the new digital twin dataset, and the nonlinear activation function RELU is used as the activation function.

[0081] 2. Multilayer Perceptron Neural Network 2: The architecture of Multilayer Perceptron Neural Network 2 is the same as that of Multilayer Perceptron Neural Network 1, and it also uses the nonlinear activation function RELU as the activation function.

[0082] 3. Generative adversarial network: Generative adversarial network is used to generate digital twin data that is closer to real data and more accurate.

[0083] 4. How to obtain the authenticity probability: By comparing with the results of the real digital twin, the authenticity judgment of the data output by the generator in the generative adversarial network is obtained, the real data is obtained, and the authenticity probability of the data is obtained by the ratio of the real data to all the data.

[0084] S113. Construction of digital twin generation model: The constructed digital twin dataset is combined with the digital twin neural network to build a digital twin generation model through neural network training and optimization.

[0085] S12. Construction of initial structural mechanical response database: With the help of the digital twin generation model, a digital twin of a semi-enclosed space area with internal structure and external environment is obtained, a physical connection network is constructed, a mechanical property association network is constructed, and numerical simulation methods are combined to consider local substructure pulse excitation correction to perform structural mechanical response simulation and construct an initial structural mechanical response database.

[0086] S121. Construction of mechanical property association network: With the help of digital twin generation model, obtain the digital twin of semi-enclosed space area with internal structure and external environment. Through the connection parts and detailed construction information between buildings, build the basic physical connection network of building units. Use finite element structural mechanics analysis method to construct the mechanical properties of building units under long-term service, so as to construct a mechanical property network.

[0087] S122. Structural mechanics response simulation: With the help of the constructed mechanical property network of the building unit, the numerical simulation method is used, the local substructure pulse excitation correction is considered, and the high-precision structural mechanics response simulation of the building unit is achieved through parameter adjustment and optimization.

[0088] S123. Construction of initial structural mechanical response database: The long-term service conditions of building units are used as database input, and the structural responses of building units under long-term service are used as database output to construct the initial structural mechanical response database.

[0089] Initial structural mechanics response database: The data features of the initial structural mechanics response database are: semi-confined space building number, service time of the semi-confined space building at the current moment, semi-confined space building type, semi-confined space building environment, semi-confined space building quality, semi-confined space building usage intensity, semi-confined space building material characteristics, semi-confined space building node number, whether the semi-confined space building node belongs to the structured grid, the position of the semi-confined space building node on the X-axis, the position of the semi-confined space building node on the Y-axis, the position of the semi-confined space building node on the Z-axis, the external force exerted on the semi-confined space building node on the X-axis, the external force exerted on the semi-confined space building node on the Y-axis, the external force exerted on the semi-confined space building node on the Z-axis, the stress of the semi-confined space building node, the strain of the semi-confined space building node, and the displacement of the semi-confined space building node.

[0090] Among them, the semi-confined space building number, the service time of the semi-confined space building, the semi-confined space building type, the semi-confined space building environment, the semi-confined space building quality, the use intensity of the semi-confined space building, the semi-confined space building material characteristics, the semi-confined space building node number, whether the semi-confined space building node belongs to the structured grid as a long-term service condition, represents the input characteristics of the data, and the number of input characteristics of the data is 8; the external force exerted on the semi-confined space building node on the X-axis, the external force exerted on the semi-confined space building node on the Y-axis, the external force exerted on the semi-confined space building node on the Z-axis, the stress of the semi-confined space building node, the strain of the semi-confined space building node, and the displacement of the semi-confined space building node as the structural response of the building node under long-term service represent the output characteristics of the data, and the number of output characteristics of the data is 6.

[0091] At the current moment, the service time of the semi-confined space building is measured in days. The environment of the semi-confined space building refers to the comprehensive environmental score of the semi-confined space building based on the geological environment and climatic conditions, which is obtained after being scored by experts. For example, 0 means an extremely bad environment, while 10 means an excellent environment. The quality of the semi-confined space building is obtained based on the building quality inspection when the semi-confined space building is completed. The weights are designed according to the importance of the sub-items. The sub-item quality assessments of all the sub-items at the time of acceptance are combined with the weights to obtain the building quality through weighted summation. The use intensity of the semi-confined space building is obtained based on the ratio between the current overload use time and the service life of the building. The characteristics of the semi-confined space building materials are sorted according to the type of the semi-confined space building and the fatigue resistance of the materials, and then numbered. For example, 0 represents the material with the strongest fatigue resistance for this type of semi-confined building, 1 represents the material with the strongest fatigue resistance, and 2 represents the material with the strongest fatigue resistance. It represents the material with strong fatigue resistance for this type of semi-enclosed building. Whether the semi-enclosed space building node belongs to the structured grid is obtained by means of the finite element structural mechanics analysis method. When the semi-enclosed space building node belongs to the structured grid, the value is 0, and when the semi-enclosed space building node does not belong to the structured grid, the value is 1. The semi-enclosed space building node in the X-axis position, the semi-enclosed space building node in the Y-axis position, and the semi-enclosed space building node in the Z-axis position represent the node position. The external force on the semi-enclosed space building node in the X-axis, the external force on the semi-enclosed space building node in the Y-axis, and the external force on the semi-enclosed space building node in the Z-axis are obtained through the mechanical property network, which represents the stress condition of the node under long-term service. The stress of the semi-enclosed space building node, the strain of the semi-enclosed space building node, and the displacement of the semi-enclosed space building node represent the structural mechanics response of the node.

[0092] For example, [0, 1850, 0, 0, 85, 0.1, 0, 12507, 125000, 0, 0, 0, 0, 50000, 500, 0.01, 1] represents a semi-enclosed space building numbered 0, which has been in service for 1850 days. The building type of the building is a tunnel, and its service environment is excellent. When completed, the comprehensive quality of the building is 85 points, and its overload service days are 185 days (1850×0.1=185). The material properties used in the building are excellent. This data is the data of node 12507 of the building. The node is located at (125000mm, 0mm, 0mm), and the force magnitude of the node on the X-axis, Y-axis, and Z-axis is (0N, 0N, 50000N), the stress of the node is 500Mpa, the strain is 0.01, and the displacement is 1mm.

[0093] S20. Construction of dynamic structural mechanics response database: Construct a spatiotemporal state matrix, use the sequential Monte Carlo method combined with knowledge-constrained optimization to achieve future state simulation, and construct a dynamic structural mechanics response database.

[0094] like Figure 4 As shown, the steps of constructing the dynamic structural mechanics response database include:

[0095] S21, construction of space-time state matrix: by sorting the node data of static reference under each time slice according to the same semi-confined space building through time, and by performing two-dimensional interpolation of time and space items, the space-time state matrix of the semi-confined space building under the relatively continuous time state is constructed;

[0096] 1. If Figure 5 As shown, the spatiotemporal state matrix is ​​a data set number matrix representation with time terms as matrix row features, space terms as matrix column features, and semi-closed building node numbers as matrix elements, where the space terms are obtained based on the distance between the node position and the origin.

[0097] 2. Data feature changes after two-dimensional interpolation: whether the semi-confined space building node belongs to the structured grid, the position of the semi-confined space building node on the X axis, the position of the semi-confined space building node on the Y axis, the position of the semi-confined space building node on the Z axis, the external force on the semi-confined space building node on the X axis, the external force on the semi-confined space building node on the Y axis, the external force on the semi-confined space building node on the Z axis, the stress of the semi-confined space building node, the strain of the semi-confined space building node, and the displacement of the semi-confined space building node are the main interpolation features. After interpolation, the value of whether the semi-confined space building node belongs to the structured grid is no longer 0 or 1, but 0, any decimal between 0 and 1, or 1. When the value of whether the semi-confined space building node belongs to the structured grid is greater than 0.5, the value of whether the semi-confined space building node belongs to the structured grid is rewritten as 1, and the values ​​of whether the remaining semi-confined space building nodes belong to the structured grid are 0; other features used for interpolation refer to the main interpolation features other than whether the semi-confined space building node belongs to the structured grid, such as the external force on the semi-confined space building node on the X-axis, etc. These features are not specially processed after interpolation.

[0098] S22. Future state simulation: Through the estimation of distribution function parameters, the sequential Monte Carlo method is used in combination with knowledge-constrained optimization to achieve future state simulation.

[0099] S221. Parameter estimation: Use the statistical analysis library to fit the distribution of data features in the spatiotemporal state according to different time dimensions, and calculate the parameters of each distribution function by trying different probability distribution functions.

[0100] For example, for a set of time series data, by comparing the likelihood function values ​​under different distribution functions, the distribution function with the largest likelihood function value is selected as the fitting distribution of the data set, and its corresponding parameters are determined, such as mean, variance, etc. These parameters can accurately describe the statistical characteristics of the data in different time dimensions and provide important basic parameters for subsequent simulations.

[0101] S222. Construction of a combined model of Sequential Monte Carlo and knowledge constrained optimization: Through the expert scoring method and with the help of the data fitting model, a knowledge constrained neural network is constructed, and the knowledge constrained neural network is embedded in the regularization term calculation of Sequential Monte Carlo to achieve the combination of Sequential Monte Carlo and knowledge constrained optimization, and to construct a combined model of Sequential Monte Carlo and knowledge constrained optimization.

[0102] The knowledge constraint term is added as a regularization term to the process of sequential Monte Carlo. By adjusting the model training process, the model can meet the constraints of physical laws while learning data features, avoid model overfitting, improve the model's generalization ability and prediction accuracy, and achieve knowledge constraint optimization.

[0103] S223, Future state simulation: Based on the estimated parameters, by setting the appropriate number of samplings and sampling step size, with the help of the combined model of Sequential Monte Carlo and Knowledge Constrained Optimization, large-scale simulation calculations are carried out to simulate various state changes that may occur in the future, generate multiple possible future trajectories, and realize future state simulation.

[0104] S23. Construction of dynamic structural mechanics response database: The generated future state is spliced ​​as a new data feature into the data features in the space-time state matrix to construct a dynamic structural mechanics response database.

[0105] S30. Construction of structural mechanics response model: By constructing a structural mechanics response prediction network, combined with a dynamic structural mechanics response database, and through data training and optimization, a structural mechanics response model is constructed.

[0106] like Figure 6 As shown, the steps of constructing the structural mechanics response model include:

[0107] S31. Construction of structural mechanics response prediction network: An encoder is constructed by combining a multi-layer perceptron neural network and an implicit neural network, and a processor and a decoder are constructed by combining a multi-layer perceptron neural network and a three-dimensional convolutional neural network to obtain a structural mechanics response prediction network.

[0108] S311, encoder construction: structural mechanics response prediction network encoder such as Figure 7As shown. The node grid encoder and edge grid encoder are constructed by using the multilayer perceptron neural network 3, the hierarchical coding strategy is introduced to construct the grid encoder, the implicit neural representation is introduced, the activation function is replaced by the sine function, and the implicit neural network 1 is constructed by combining the multilayer perceptron network architecture. The node and edge implicit encoders are constructed respectively, the hierarchical coding strategy is introduced to construct the implicit encoder, the dynamic weight is given by the feedforward network 1, the grid encoder and the implicit encoder are weightedly fused, and the structural mechanics response prediction network encoder is constructed.

[0109] 1. Multilayer Perceptron Neural Network 3: Multilayer Perceptron Neural Network 3 is constructed by 1 input layer, 3 hidden layers, and 1 output layer. The number of neurons in the input layer is the number of input features of data in the dynamic structure response mechanics database. The number of neurons in the 3 hidden layers is the same, and the number of neurons is a number greater than 256 and a multiple of 2. The number of neurons in the output layer is the number of output features of data in the dynamic structure response database, and the nonlinear activation function RELU is used as the activation function.

[0110] 2. Implicit neural network 1: Implicit neural network 1 uses the same multilayer perceptron neural network architecture as multilayer perceptron neural network 3, with the input layer, hidden layer, and output layer, but changes the original nonlinear activation function RELU to a sine function.

[0111] 3. Edge: Use the K-nearest neighbor algorithm to determine neighbor nodes and build edges.

[0112] 4. Hierarchical coding strategy: First, edge coding is performed, and then point coding is performed according to the neighbor node information corresponding to the edge to implement a hierarchical coding strategy.

[0113] 5. Feedforward neural network 1: Feedforward neural network 1 is constructed by 1 input layer, 1 hidden layer and 1 output layer. The number of neurons in the input layer is obtained by the sum of the number of neurons in the output layer of the grid encoder and the implicit encoder. The number of neurons in the hidden layer is a multiple of 2 greater than 64. The number of neurons in the output layer is obtained by the input features of the data in the dynamic structure response database.

[0114] S312, processor construction: Structural mechanics response prediction network processor network architecture such as Figure 8 As shown in (a) and (b) in the figure. A three-dimensional convolutional neural network 1 is constructed, a node structured grid processor and an edge structured grid processor are constructed respectively, a structured grid processor is constructed with the help of a hierarchical processing strategy, a multilayer perceptron neural network 4 and an implicit neural network 2 are connected in series, a node unstructured grid processor and an edge unstructured grid processor are constructed respectively, an unstructured grid processor is constructed with the help of a hierarchical processing strategy, and a processor is constructed.

[0115] The data flow direction is determined according to whether the semi-confined space building node belongs to a structured grid. When the semi-confined space building node belongs to a structured grid, the data will be processed by a structured grid processor. When the semi-confined space building node belongs to an unstructured grid, the data will be processed by an unstructured grid processor.

[0116] 1. Multilayer Perceptron Neural Network 4: Multilayer Perceptron Neural Network 4 is constructed by 1 input layer, 3 hidden layers, and 1 output layer, where the number of neurons in the input layer is consistent with the number of neurons in the output layer of the grid encoder or implicit encoder, the number of neurons in the 3 hidden layers is the same, and the number of neurons is a number greater than 256 and a multiple of 2, the number of neurons in the output layer is consistent with the number of neurons in the hidden layer, and the nonlinear activation function RELU is used as the activation function.

[0117] 2. Three-dimensional convolutional neural network 1: construct an input layer, construct multiple convolutional layers and pooling layers, in which convolutional layers and pooling layers are constructed alternately, construct fully connected layers and output layers, and construct a three-dimensional convolutional neural network. The input dimension of the three-dimensional convolutional neural network is consistent with the number of neurons in the output layer of the grid encoder or implicit encoder, and the output dimension of the three-dimensional convolutional neural network is consistent with the number of neurons in the output layer of the multilayer perceptron neural network 4.

[0118] 3. Implicit neural network 2: Implicit neural network 2 uses the same multilayer perceptron neural network architecture as multilayer perceptron neural network 4, with the input layer, hidden layer, and output layer, but changes the original nonlinear activation function RELU to a sine function.

[0119] S313, decoder construction: Structural mechanics response prediction network decoder network architecture is as follows Fig. 9 As shown in (a) and (b) in Fig. 2, a three-dimensional convolutional neural network 2 is constructed as a decoder for structured grids, and a multi-layer perceptron neural network 5 is constructed as a decoder for unstructured grids. The attention mechanism is introduced in the structured and unstructured grid decoders in combination with the average pooling operation to calculate the channel weight score. The data finally output by the decoder is sorted according to the grid node number to achieve data decoding and restoration, and to construct a decoder.

[0120] The structured grid data is encoded by the encoder, processed by the structured grid processor, and finally decoded and restored by the structured decoder; the unstructured grid data is encoded by the encoder, processed by the unstructured grid processor, and finally decoded and restored by the unstructured decoder; the data is sorted according to the semi-enclosed space building number and the semi-enclosed space building node number, combined with the space-time state matrix, to achieve data decoding and restoration.

[0121] 1. Three-dimensional convolutional neural network 2: construct an input layer, construct multiple convolutional layers and pooling layers, in which convolutional layers and pooling layers are constructed alternately, construct fully connected layers and output layers, and construct a three-dimensional convolutional neural network. The input dimension of the three-dimensional convolutional neural network is consistent with the output dimension of the output layer of the three-dimensional convolutional neural network 1, and the output dimension of the three-dimensional convolutional neural network is consistent with the number of output features of the data in the dynamic structural mechanics response database.

[0122] 2. Multilayer perceptron neural network 5: Multilayer perceptron neural network 5 is constructed by 1 input layer, 3 hidden layers, and 1 output layer. The number of neurons in the input layer is consistent with the number of neurons in the output layer of implicit neural network 2, the number of neurons in the 3 hidden layers is the same, and the number of neurons is a number greater than 256 and a multiple of 2. The number of neurons in the output layer is consistent with the number of output features of data in the dynamic structural mechanics response database, and the nonlinear activation function RELU is used as the activation function.

[0123] S314. Construction of structural mechanics response prediction network: By connecting the encoder, processor, and decoder in series, the data flows from the encoder to the decoder to build a structural mechanics response prediction network.

[0124] S32. Construction of structural mechanics response model: Substitute the data in the dynamic structural mechanics response database into the structural mechanics response prediction network, and obtain the structural mechanics response model through data training and optimization.

[0125] S40. Dynamic damage tracking of semi-confined space buildings: Build a service life prediction model, use the structural mechanics response model, combine the existing damage determination method and data dynamic reconstruction method to achieve dynamic damage assessment, real-time damage correction, service life prediction, damage condition tracking, and complete dynamic damage tracking of semi-confined space buildings.

[0126] The damage condition refers to the service condition that has the greatest impact on the service life at a certain moment. The data input features of the dynamic structural mechanics response database of node damage are spliced, and the node damage is spliced ​​in the dynamic structural mechanics response database as a new data output feature.

[0127] like Fig.10 As shown in the figure, the steps of dynamic damage tracing of semi-enclosed space buildings include:

[0128] S41. Dynamic damage assessment: With the help of the dynamic structural mechanics response database, combined with the existing damage determination method, node damage is quantified according to the structural mechanics response, and the node damage is spliced ​​into the dynamic structural mechanics response database to achieve dynamic damage assessment of semi-enclosed space buildings in different spaces at different times.

[0129] Existing damage determination methods: slight damage, the deformation is usually at the millimeter level or even smaller, the stress is within the lower proportion of the allowable stress of the material, and the node damage is represented by 0.25; moderate damage, for metal structural mechanics, it may reach the order of several millimeters to more than ten millimeters, concrete structural mechanics may have cracks, the crack width is 0.5 mm to 2 mm or even larger, the stress is about 30% - 60% of the allowable stress of the material, and the node damage is represented by 0.5; severe damage, the deformation of metal structural mechanics may be tens of millimeters or even larger, concrete structural mechanics may have large-scale cracks and breakage, the stress may reach more than 80% of its yield strength or even completely yield, and the node damage is represented by 0.75; complete damage, the structural mechanics unit completely loses its original shape and structure, and suffers severe distortion, fracture, collapse, etc., the deformation can no longer be measured by conventional size concepts, and the node damage is represented by 1.

[0130] S42. Real-time correction of damage: Through real-time collection of real data and with the help of the dynamic structural mechanics response database construction method, real-time correction of the dynamic structural mechanics response database data is carried out. With the help of the structural mechanics response model and the input characteristics of the corrected database data, the real-time corrected node structural mechanics response data is obtained. With the help of the existing damage determination method, the node damage is quantified and the real-time corrected damage data is obtained.

[0131] S43. Service life prediction: With the help of the dynamic structural response database of the damage data that has been spliced ​​with real-time corrections, the future state simulation is re-performed, and the service life of the semi-enclosed space building is obtained at the moment when the damage of all key nodes is 1. A service life prediction data set is constructed, and a feedforward neural network 2 is used to obtain a service life prediction model through data training and optimization. The input features of the data in the real case service life prediction data set are constructed to achieve service life prediction.

[0132] The key node refers to the node of the local substructure in the local substructure pulse excitation correction, and the local substructure is obtained from the physical connection network of the building monomer.

[0133] 1. Service life prediction data set: The data features of the service life prediction data set are: semi-confined space building number, current service time of semi-confined space building, semi-confined space building type, semi-confined space building environment, semi-confined space building quality, semi-confined space building usage intensity, semi-confined space building material characteristics, and semi-confined space building service life; among them, semi-confined space building number, current service time of semi-confined space building, semi-confined space building type, semi-confined space building environment, semi-confined space building quality, semi-confined space building usage intensity, and semi-confined space building material characteristics are the input features of the data in the service life prediction data set, and the service life of semi-confined space building is the output feature of the data in the service life prediction data set.

[0134] For example, [0, 1850, 0, 0, 85, 0.1, 0, 16987] represents a semi-enclosed space building numbered 0, which has been in service for 1850 days. The building type of the building is a tunnel, and its service environment is excellent. When completed, the comprehensive quality of the building is 85 points, and its overload service days are 185 days (1850×0.1=185). The materials used in the building have excellent properties. In the current state, the total service life of the building is 16987 years.

[0135] 2. Feedforward neural network 2: Feedforward neural network 2 is constructed by 1 input layer, 1 hidden layer and 1 output layer. The number of neurons in the input layer is consistent with the number of input features of the data in the service life prediction dataset. The number of neurons in the hidden layer is a multiple of 2 greater than 256, and the number of neurons in the output layer is 1.

[0136] S44, damage condition tracking: With the help of the service life prediction model, the service life of the semi-enclosed space building under evaluation at any time is obtained, the time derivative of the service life at each time in the past is calculated, and the time when the service life changes the most is obtained. With the help of the space-time state matrix, the service condition design with the greatest impact on the service life is obtained, and dynamic tracking of damage conditions is achieved;

[0137] The service condition design that has the greatest impact on the service life refers to the condition corresponding to the moment when the service life changes the most.

[0138] S45, Dynamic Damage Tracking: Accurate dynamic damage assessment is achieved through dynamic damage assessment and real-time damage correction, and damage condition tracking is achieved through service life prediction to achieve dynamic damage tracking.

[0139] The above embodiments are only for illustrating the technical idea of ​​the present invention, and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for predicting the structural mechanical response of semi-enclosed space buildings based on a structural mechanical response model, characterized in that: include: Based on the urban building digital twin technology and combined with numerical simulation methods, structural mechanics response simulation is carried out to build an initial structural mechanics response database; Construct a structural mechanics response prediction network, combine it with the initial structural mechanics response database, and construct a structural mechanics response model through data training and optimization; Based on the structural mechanics response model, the structural mechanics response of semi-enclosed space buildings under long-term service can be predicted; The steps to construct the initial structural mechanics response database include: Using urban building digital twin technology, a real semi-enclosed space digital twin with external environment is generated; Obtain the internal structure information of the real semi-confined space digital twin through the semi-confined space building design specifications; Based on the neural core surface reconstruction algorithm, a digital twin of a semi-enclosed space area is constructed; Combine multiple digital twins of semi-confined space areas to construct a digital twin dataset of semi-confined space areas; Construct a digital twin neural network, use the semi-confined space area digital twin dataset for training and optimization, and generate a digital twin generation model; With the help of digital twin generation model, the digital twin of the semi-enclosed space area with internal structure and external environment is obtained; Through the physical connection network of building monomers, the finite element structural mechanics analysis method is used to construct the mechanical property correlation network of building monomers; With the help of the mechanical property correlation network of the building monomer, the structural mechanical response simulation is carried out considering the local substructure pulse excitation correction, and an initial structural mechanical response database is constructed with the long-term service conditions of the building monomer as input and the structural response of the building monomer under long-term service as output; The structural mechanics response prediction network includes a structural mechanics response prediction network encoder, a structural mechanics response prediction network processor and a structural mechanics response prediction network decoder; In the structural mechanics response prediction network encoder: the third multi-layer perceptron neural network is used to construct a node feature grid encoder and an edge feature grid encoder respectively, and a hierarchical coding strategy is introduced to form a grid encoder; the first implicit neural network is used to construct a node feature implicit encoder and an edge feature implicit encoder respectively, and a hierarchical coding strategy is introduced to form an implicit encoder; with the help of the first feedforward neural network, dynamic weights are given to the grid encoder and the implicit encoder respectively, and the grid encoder and the implicit encoder are weighted and fused to form a structural mechanics response prediction network encoder; In the structural mechanics response prediction network processor: a node structured grid processor and an edge structured grid processor are constructed respectively by using the first three-dimensional convolutional neural network, and a structured grid processor is formed by using a hierarchical processing strategy; a node unstructured grid processor and an edge unstructured grid processor are constructed respectively by using the fourth multilayer perceptron neural network and the second implicit neural network connected in series, and an unstructured grid processor is formed by using a hierarchical processing strategy; In the structural mechanics response prediction network decoder: a structured grid processor is constructed by introducing a second three-dimensional convolutional neural network with an attention mechanism combined with an average pooling operation, and an unstructured grid decoder is constructed by introducing a fifth multi-layer perceptron neural network with an attention mechanism combined with an average pooling operation; wherein, the input of the structured grid processor is the output of the structured grid processor, and the input of the unstructured grid decoder is the output of the unstructured grid processor.

2. The prediction method according to claim 1, characterized in that: The data features in the semi-confined space area digital twin dataset are: semi-confined space building number, area size, semi-confined space building type, semi-confined space building length, semi-confined space building width, semi-confined space building height, semi-confined space building node number, semi-confined space building node X-axis position, semi-confined space building node Y-axis position, semi-confined space building node Z-axis position; The data features in the initial structural mechanics response database are: semi-confined space building number, service time of the semi-confined space building at the current moment, semi-confined space building type, semi-confined space building environment, semi-confined space building quality, semi-confined space building usage intensity, semi-confined space building material characteristics, semi-confined space building node number, whether the semi-confined space building node belongs to the structured grid, the position of the semi-confined space building node on the X-axis, the position of the semi-confined space building node on the Y-axis, the position of the semi-confined space building node on the Z-axis, the external force on the semi-confined space building node on the X-axis, the external force on the semi-confined space building node on the Y-axis, the external force on the semi-confined space building node on the Z-axis, the stress of the semi-confined space building node, the strain of the semi-confined space building node, and the displacement of the semi-confined space building node.

3. The prediction method according to claim 1, characterized in that: The steps to build a digital twin neural network include: The input features in the digital twin data set of the semi-enclosed space area are used as the input of the first multi-layer perceptron neural network, and the output features in the digital twin data set of the semi-enclosed space area are used as the output of the first multi-layer perceptron neural network to construct a generator; The discriminator is constructed by taking the output features in the digital twin dataset of the semi-enclosed space area as the input of the second multi-layer perceptron neural network and taking the corresponding authenticity probability as the output of the second multi-layer perceptron neural network; The generator and discriminator constitute a generative adversarial network, also known as a digital twin neural network.

4. The prediction method according to claim 1, characterized in that: Also includes: Based on the initial structural mechanics response database, a spatiotemporal state matrix is ​​constructed; Use the sequential Monte Carlo method combined with knowledge-constrained optimization to achieve future state simulation and build a dynamic structural mechanics response database; Build a structural mechanics response prediction network, combine it with the dynamic structural mechanics response database, and build a structural mechanics response model through data training and optimization.

5. The prediction method according to claim 4, characterized in that: The steps to construct the dynamic structural mechanics response database include: By sorting the node data of static reference under each time slice according to the time of the same semi-confined space building, and performing two-dimensional interpolation of time and space items, a spatiotemporal state matrix of the semi-confined space building under a relatively continuous time state is constructed; the spatiotemporal state matrix is ​​a data set number matrix with time items as matrix row features, space items as matrix column features, and semi-confined space building node numbers as matrix elements, wherein the space items are obtained according to the distance between the node position and the origin; Using the statistical analysis library, the spatiotemporal state matrix is ​​fitted and distributed according to different time dimensions to obtain the parameters of each distribution function; Through the expert scoring method and the data fitting model, a knowledge constraint neural network is constructed, and the knowledge constraint neural network is embedded in the regularization term calculation of the sequential Monte Carlo to achieve the combination of sequential Monte Carlo and knowledge constraint optimization, and to build a model combining sequential Monte Carlo and knowledge constraint optimization. Based on the parameters of each distribution function obtained, the number of sampling and the step size are set, and by building a model combining sequential Monte Carlo and knowledge-constrained optimization, various possible future states are simulated to achieve future state simulation; The simulated future state is used as a new data feature to be spliced ​​into the data features in the space-time state matrix to construct a dynamic structural mechanics response database.

6. A dynamic damage tracking method for semi-enclosed space buildings based on a structural mechanics response model, characterized in that: include: Based on the second feedforward neural network, a service life prediction model is constructed; With the help of the structural mechanics response model as described in any one of claims 1 to 5, combined with a damage determination method and a data dynamic reconstruction method, dynamic damage tracking of semi-enclosed space buildings can be achieved.

7. The damage tracing method according to claim 6, characterized in that: The steps of dynamic damage tracing of semi-confined space buildings include: Through real-time acquisition of real data of semi-confined space buildings, the initial / dynamic structural mechanics response database is corrected in real time; Using the structural mechanics response model, based on the real-time corrected initial / dynamic structural mechanics response database, the prediction results of the corresponding structural mechanics response are obtained; Combined with the damage determination method, node damage is quantified and the node damage quantification value is spliced ​​into the initial / dynamic structural mechanics response database after real-time correction; With the help of the spliced ​​initial / dynamic structural mechanical response database, the future state of the semi-confined space building is simulated, the service life of the semi-confined space building is obtained, and a service life prediction data set is constructed; Combined with the service life prediction data set, the second feedforward neural network is trained and optimized to build a service life prediction model; With the help of the service life prediction model, the service life of semi-enclosed space buildings at any time can be obtained, the moment when the service life changes the most can be found, and then the service condition design that has the greatest impact on the service life can be obtained to achieve dynamic damage tracking.

8. The damage tracing method according to claim 7, characterized in that: The data features of the service life prediction data set are: semi-confined space building number, current service time of semi-confined space building, semi-confined space building type, semi-confined space building environment, semi-confined space building quality, semi-confined space building usage intensity, semi-confined space building material characteristics, and semi-confined space building service life.

Images