BIM-based prefabricated component management method and system

Abstract

This application provides a BIM-based method and system for managing prefabricated components, enabling rapid acquisition and family-based generation of simulated prefabricated component models. This method not only flexibly generates multiple target simulated prefabricated component models of the same family based on a first simulated prefabricated component model, but also further constructs multiple simulated prefabricated building models corresponding to the same family of simulated components based on these target models. Furthermore, by accurately determining the key parameters reflecting the model's characteristic distribution (i.e., the first component characteristic parameters and the second component characteristic parameters), the accuracy and applicability of the simulated prefabricated component models are significantly improved, greatly enhancing the design efficiency and flexibility of prefabricated buildings.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method and system for managing prefabricated components based on BIM. Background Art

[0002] In the construction industry, with the continuous development of Building Information Modeling (BIM) technology, its application in building design, construction and management is becoming more and more extensive. BIM technology integrates and shares construction project information by creating three-dimensional digital models, providing strong support for the informatization and intelligence of the construction industry.

[0003] In the field of prefabricated buildings, the application of BIM technology is also of great significance. As a new type of construction method, prefabricated buildings occupy an important position in modern buildings with their advantages such as high efficiency, energy saving and environmental protection. However, there are still some challenges and problems in the design and management of prefabricated buildings.

[0004] On the one hand, the diversity and complexity of prefabricated components increase the difficulty of design and management. Traditional design methods often require manual creation and adjustment of models for each component, which is not only time-consuming and labor-intensive, but also prone to errors. On the other hand, the components of prefabricated buildings need to work closely together to ensure the stability and safety of the overall structure. Therefore, how to efficiently manage and coordinate these components has become an urgent problem to be solved.

[0005] In addition, in the design process of prefabricated buildings, different component models of the same component family often need to be repeatedly modified and adjusted to meet different design requirements. However, this modification and adjustment process often lacks systematicity and standardization, resulting in low design efficiency and uneven model quality. Summary of the invention

[0006] In view of this, the purpose of this application is to provide a BIM-based prefabricated component management method and system.

[0007] According to a first aspect of the present application, a method for managing assembled components based on BIM is provided, which is applied to a BIM service system. The method comprises:

[0008] Acquire a first simulated prefabricated component model from a prefabricated component model library, wherein the prefabricated component model library includes at least one simulated prefabricated component model;

[0009] Based on the first simulated prefabricated component model, generating at least one target simulated prefabricated component model of the same simulated component group;

[0010] For each of the target simulated prefabricated component models, at least one simulated prefabricated building model is generated based on the target simulated prefabricated component model, and the simulated prefabricated building models generated based on each of the target simulated prefabricated component models correspond to the same simulated component group; wherein the step of determining each simulated prefabricated component model in the prefabricated component model library includes:

[0011] Determine a first component characteristic parameter and a second component characteristic parameter, wherein the first component characteristic parameter and the second component characteristic parameter respectively reflect a focused node and a dispersed derivative parameter of a model characteristic distribution corresponding to a group of assembled component models;

[0012] A simulated assembled component model is generated based on the first component characteristic parameter and the second component characteristic parameter.

[0013] According to the second aspect of the present application, a BIM service system is provided, which includes a processor and a readable storage medium, wherein the readable storage medium stores a program, and when the program is executed by the processor, the aforementioned BIM-based prefabricated component management method is implemented.

[0014] According to a third aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are monitored to be executed, the aforementioned BIM-based prefabricated component management method is implemented.

[0015] According to any of the above aspects, the embodiment of the present application realizes the rapid acquisition and group generation of the simulated prefabricated component model by efficiently utilizing the prefabricated component model library. This method can not only flexibly generate multiple target simulated prefabricated component models of the same simulated component group based on the first simulated prefabricated component model, but also further construct multiple simulated prefabricated building models corresponding to the same simulated component group based on these target models. In addition, by accurately determining the key parameters reflecting the distribution of model characteristics (i.e., the first component characteristic parameters and the second component characteristic parameters), the accuracy and applicability of the simulated prefabricated component model are significantly improved, and the design efficiency and flexibility of prefabricated buildings are greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0017] Figure 1A schematic diagram of a process of a BIM-based prefabricated component management method provided in an embodiment of the present application is shown;

[0018] Figure 2 A schematic diagram of the component structure of a BIM service system for implementing the above-mentioned BIM-based prefabricated component management method provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0019] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present application.

[0020] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0021] Figure 1 The flowchart of the BIM-based prefabricated component management method provided by the embodiment of the present application is shown. It should be understood that in other embodiments, the order of some steps in the BIM-based prefabricated component management method can be interchanged according to actual needs, or some steps can be omitted or deleted. The detailed steps of the BIM-based prefabricated component management method are introduced as follows.

[0022] Step S110: acquiring a first simulated prefabricated component model from a prefabricated component model library, wherein the prefabricated component model library includes at least one simulated prefabricated component model.

[0023] Step S120: generating at least one target simulated prefabricated component model of the same simulated component group based on the first simulated prefabricated component model.

[0024] Step S130, for each of the target simulated prefabricated component models, at least one simulated prefabricated building model is generated based on the target simulated prefabricated component model, and the simulated prefabricated building models generated based on each of the target simulated prefabricated component models correspond to the same simulated component group. The step of determining each simulated prefabricated component model in the prefabricated component model library includes:

[0025] Step S101 : determining a first component characteristic parameter and a second component characteristic parameter, wherein the first component characteristic parameter and the second component characteristic parameter respectively reflect a focused node and a dispersed derivative parameter of a model characteristic distribution corresponding to a group of assembled component models.

[0026] Step S102: generating a simulated assembled component model based on the first component characteristic parameter and the second component characteristic parameter.

[0027] In this embodiment, in step S110, a certain architectural design platform is designing a large residential community project, which includes multiple different types of prefabricated buildings. The server maintains a huge prefabricated component model library, which is accumulated and constructed over a long period of time and contains a variety of simulated prefabricated component models, such as prefabricated wall panel models, prefabricated floor slab models, prefabricated beam and column models of different specifications.

[0028] In this community project, the designers have determined that the overall architectural style is modern and simple, and have made preliminary plans for the structure and function of the building. The server starts working according to these requirements. First, step S110 is performed to obtain a first simulated prefabricated component model from the prefabricated component model library.

[0029] For example, the server receives a requirement to build a design plan starting from a prefabricated wall panel model. The server will search the model library for prefabricated wall panel models that match the modern minimalist style and are suitable for residential areas. In the model library, prefabricated wall panel models can be classified in many ways, such as by material (concrete prefabricated wall panels, lightweight prefabricated wall panels, etc.), by size (wall panels of different heights and widths), and by function (insulation type, sound insulation type, etc.). Based on the project requirements, the server gave priority to a commonly used prefabricated wall panel model with a height of 3 meters, a width of 0.6 meters, a concrete material and a certain insulation function as the first simulated prefabricated component model. This model contains various detailed information about the wall panel, such as simulation data on geometry, material properties, connection nodes, etc.

[0030] In step S101, continuing to take the above-mentioned prefabricated wall panel model as an example, for the prefabricated wall panel selected as the first simulated assembled component model, the server needs to determine its first component characteristic parameter and second component characteristic parameter.

[0031] For the geometric aspects of the first component characteristic parameters:

[0032] The overall shape of the wall panel is a rectangle, and its length, width and height are the most basic geometric features. For this prefabricated wall panel, the length dimension determines its coverage in the building structure, while the width and height affect the spatial division and appearance proportion of the building. These basic geometric dimensions are part of the focal nodes in the feature distribution of the geometric shape. For example, the length is 4 meters, which is a clear and relatively stable feature value and is one of the focal points of the model feature distribution in the geometric shape dimension.

[0033] In terms of the location and type of connection nodes, it is assumed that the upper and lower edges of the prefabricated wall panels are connected by a specific type of reserved slots, the left and right edges are connected by bolts, and the location of the connection nodes is evenly distributed at specific locations on the edges. The types and locations of these connection nodes are very critical features when constructing prefabricated buildings, and they constitute the focus nodes of the model feature distribution in the connection structure dimension.

[0034] Material properties:

[0035] The material is concrete, and its strength grade is an important focus node. For example, the strength grade of concrete is C30. This strength grade determines important properties such as the bearing capacity and durability of the wall panel, and is the focus point in the characteristic distribution of material properties.

[0036] The type and thickness of the insulation material is also part of the focus node. Assuming that the insulation material used is polystyrene foam board with a thickness of 5 cm, this plays a key role in the thermal insulation performance of the wall panel.

[0037] For the geometric aspects of the second component characteristic parameters:

[0038] Although the overall shape is rectangular, there may be certain dimensional tolerances during the actual production and installation process. For example, the tolerance range of length is ±5 mm, the tolerance range of width is ±3 mm, and the tolerance range of height is ±2 mm. These tolerances are scattered derivative parameters in the dimensions of geometric shapes, which reflect the variability of model features in actual production and installation.

[0039] There may be some minor unevenness on the surface of the wallboard, which is allowed within a certain range. Assume that the standard deviation of the unevenness is 0.5 mm, which is also a dispersion derivative parameter in the distribution of geometric shape features.

[0040] Material properties:

[0041] Although the mix ratio of concrete is determined according to certain standards, there may be some small fluctuations in actual production. For example, the ratio of cement, sand, gravel and water may fluctuate within a range of ±1%, which is a dispersed derivative parameter in the material property dimension.

[0042] The thermal conductivity of the insulation material may fluctuate within a certain range due to different production batches. Assuming that the fluctuation range of the thermal conductivity is ±0.05W / (m·K), this is also part of the dispersion derived parameters.

[0043] In step S102, the server has determined the first component characteristic parameter and the second component characteristic parameter of the prefabricated wall panel. Then, the server loads these parameters into the target generation network. This target generation network is pre-trained and is specifically used to generate a simulated assembled component model according to the given component characteristic parameters.

[0044] The internal structure of the target generation network may contain multiple hidden layers and neurons, which perform complex calculations based on the input feature parameters. For example, the neurons in the network construct the basic framework of the prefabricated wall panel model based on the first component feature parameters (such as basic dimensions of the geometric shape, connection node information, material strength grade, etc.). Then, this basic framework is adjusted according to the second component feature parameters (such as dimensional tolerances, material ratio fluctuations, etc.) to simulate the uncertainty in actual production and installation.

[0045] Specifically, when generating the 3D geometric model of the prefabricated wall panel, the network constructs an ideal rectangular wall panel shape based on the basic dimensions of length, width and height. Then, taking into account the tolerance range of length, width and height, based on this ideal shape, the actual size of the model is adjusted by randomly generating values ​​within the tolerance range, making the generated model closer to the actual production situation. For the connection nodes, the initial model of the connection structure is constructed based on the basic information of its type and position, and then taking into account some possible installation deviations (which can be represented by the relevant data in the characteristic parameters of the second component), the position of the connection node is slightly randomly adjusted.

[0046] In terms of material properties, the basic mechanical properties of the material are determined according to the first component characteristic parameters such as the concrete strength grade. Then, according to the fluctuation range of the concrete mix ratio (the second component characteristic parameter), the microstructure parameters inside the material are adjusted to affect the actual performance of the material. For thermal insulation materials, insulation layer models with different performances are generated according to parameters such as its type, thickness, and fluctuation range of thermal conductivity. Finally, all these adjusted geometries, connection structures, and material properties are integrated together to generate a complete simulated assembled component model of prefabricated wall panels.

[0047] In step S120, still taking the above-mentioned prefabricated wall panel as an example, the server generates a target simulated prefabricated component model of the same simulated component group (prefabricated wall panel group) based on the first simulated prefabricated component model (the previously selected prefabricated wall panel model). Assume that according to the design requirements of the community project, 5 target simulated prefabricated component models need to be generated (that is, the first scale of the target simulated prefabricated component models to be generated is determined to be 5).

[0048] The server first constructs five candidate simulated prefabricated component models. These candidate models can be obtained by making some preliminary and regular changes on the basis of the first simulated prefabricated component model. For example, in terms of geometry, the first candidate model can be to increase the height of the prefabricated wall panel by 0.1 meters and keep the width unchanged; the second candidate model can be to reduce the length by 0.2 meters and adjust the spacing of the connection nodes; the third candidate model can be to change the thickness of the wall panel, thereby affecting its weight and structural performance; the fourth candidate model can be to change the surface texture of the wall panel, which does not affect the structural performance but affects the appearance; the fifth candidate model can be to adjust the distribution of the insulation material to form different insulation structures inside the wall panel.

[0049] Next, the server generates five target simulated prefabricated component models of the same simulated component group based on the first simulated prefabricated component model and each candidate simulated prefabricated component model.

[0050] First, the server determines the estimated correlation between each candidate simulated prefabricated component model and the first simulated prefabricated component model. This estimated correlation is calculated based on some pre-set rules and algorithms. For example, for the first candidate model (height increase of 0.1 meters), since the height change is relatively small and within the range allowed by the building structure, the estimated correlation may be higher, assuming 0.8 (each estimated correlation is within the set correlation range, such as 0.6 - 0.9). For the second candidate model (length reduction of 0.2 meters and adjustment of the connection node spacing), because the length change and the adjustment of the connection nodes will have a greater impact on the structure, the estimated correlation may be lower, assuming 0.6.

[0051] Then, the server performs a cyclic regularization conversion on each candidate simulated prefabricated component model. After each round of regularization conversion, the true correlation between the candidate simulated prefabricated component model after each round of regularization conversion and the first simulated prefabricated component model is determined. For example, for the first candidate model, in the first round of regularization conversion, the structural stability analysis of the wall panel after the height increase may be performed, and some related parameters (such as increasing the number of internal reinforcement ribs) are adjusted according to the analysis results, and then the true correlation with the first simulated prefabricated component model is recalculated. Assume that the true correlation becomes 0.75 after the first round of conversion.

[0052] The server determines the correlation error based on each true correlation and each estimated correlation. For the first candidate model, the true correlation is 0.75, the estimated correlation is 0.8, and the correlation difference is 0.8 - 0.75 = 0.05. Then the correlation error is determined based on each correlation difference (this can be done by taking the square root of the sum of all the correlation differences, etc.).

[0053] In the cyclic regularization conversion process, parameter learning is terminated when the correlation error is minimized. It is assumed that after multiple rounds of conversion, the correlation errors of each candidate model are minimized. At this time, the server outputs the candidate simulated prefabricated component models after each round of regularization conversion when parameter learning is terminated as 5 target simulated prefabricated component models of the same simulated component family. These target models have different characteristics while satisfying a certain degree of correlation with the first simulated prefabricated component model, which can meet the diverse needs of different building units for prefabricated wall panels in community projects.

[0054] In step S130, taking one of the five previously generated target simulated prefabricated component models (assuming that it is the target model of the prefabricated wall panel with a height increase of 0.1 meters) as an example, the server generates at least one simulated prefabricated building model based on this target model. First, the server determines at least one random perturbation model. These random perturbation models can be models that simulate various uncertain factors in the construction process. For example, a random perturbation model can simulate the impact of wind force on the installation accuracy of prefabricated wall panels at the construction site; another random perturbation model can simulate the impact of construction workers' operation errors on the firmness of wall panel connections.

[0055] For each random perturbation model, a predefined number of iteration optimizations (assuming 3 times) is performed on the target simulated prefabricated component model based on the random perturbation model.

[0056] First round of optimization:

[0057] Assume that a random perturbation model is used to simulate the influence of wind on the construction site. Based on this model, the target simulated prefabricated component model (prefabricated wall panel) is randomly perturbed and adjusted. For example, the wind may cause a certain horizontal offset of the wall panel during the installation process. The server calculates the offset of the wall panel according to the magnitude and direction of the wind (which are parameters in the random perturbation model) and generates a random perturbation adjustment model. Then, the target simulated prefabricated component model (the wall panel before the perturbation) is structurally adapted, for example, considering possible offsets, adjusting the connection method with adjacent components, and generating a structural feature model that carries the simulated component group characteristics of the simulated component group (prefabricated wall panel group) corresponding to the first simulated prefabricated component model. Finally, the random perturbation adjustment model and the structural feature model generated in this round of the cycle are fused to generate the target adjustment model.

[0058] Second round of optimization:

[0059] Continue to use this random perturbation model to perform random perturbation adjustment on the target adjustment model generated in the first cycle. For example, the continued action of wind force may further increase the offset of the wall panel, and the server recalculates the offset and adjusts the model. Then, perform structural adaptability adjustment on the target adjustment model generated in the first cycle again (because the structure may need to be readjusted as the wall panel further deviates) to generate a new structural feature model. The random perturbation adjustment model of this round and the new structural feature model are merged to generate a new target adjustment model.

[0060] The third round of optimization:

[0061] Similarly, the third random disturbance adjustment and structural adaptability adjustment are performed to generate the final target component model sequence.

[0062] Based on the target component model sequence generated by the last round of cyclic optimization, a simulated prefabricated building model is generated. For example, the prefabricated wall panel that has been adjusted many times is placed in the corresponding position of the building. At the same time, according to the overall design plan of the building, other prefabricated components (such as prefabricated floor slabs, prefabricated beams and columns, etc.) are added to construct a complete simulated prefabricated building model. In the same way, for each target simulated prefabricated component model, multiple different simulated prefabricated building models can be generated according to different random perturbation models. These building models all belong to the same simulated component group (prefabricated wall panel group) and can simulate various situations that may be encountered during actual construction and use, providing a more comprehensive reference for building design and construction.

[0063] Based on the above steps, the embodiment of the present application realizes the rapid acquisition and group generation of the simulated prefabricated component model by efficiently utilizing the prefabricated component model library. This method can not only flexibly generate multiple target simulated prefabricated component models of the same simulated component group based on the first simulated prefabricated component model, but also further construct multiple simulated prefabricated building models corresponding to the same simulated component group based on these target models. In addition, by accurately determining the key parameters reflecting the distribution of model characteristics (i.e., the first component characteristic parameters and the second component characteristic parameters), the accuracy and applicability of the simulated prefabricated component model are significantly improved, and the design efficiency and flexibility of prefabricated buildings are greatly improved.

[0064] In a possible implementation, step S102 includes: loading the first component characteristic parameter and the second component characteristic parameter into a target generation network to generate a simulated assembly component model.

[0065] The training step of the target generation network includes:

[0066] Step A110: determining a sample learning data sequence, wherein the sample learning data sequence includes a plurality of real prefabricated building models.

[0067] Step A120: Load each of the real prefabricated building models into the basic component parsing unit of the basic generation network to generate a real prefabricated component model.

[0068] Step A130, for each of the real prefabricated component models, the real prefabricated component model is loaded into the basic component feature encoding unit of the basic generation network, a first sample component feature parameter and a second sample component feature parameter are generated, and the first sample component feature parameter and the second sample component feature parameter are loaded into the initial generation unit of the basic generation network to generate a simulated prefabricated component model corresponding to the real prefabricated component model. The first sample component feature parameter and the second sample component feature parameter respectively reflect the focus node and the dispersed derivative parameter after the real prefabricated component model is mapped to the model feature distribution.

[0069] Step A140, determining a global training error based on each of the real prefabricated component models and the corresponding simulated prefabricated component models, performing cyclic parameter learning on the basic generation network based on the sample learning data sequence, terminating the parameter learning when the global training error is minimized, and outputting the initial generation unit when the parameter learning is terminated as the target generation network.

[0070] In the architectural design project mentioned above, for the prefabricated wall panel, the server has determined its first component characteristic parameters (such as basic geometric dimensions, connection node type and position, material strength grade, etc.) and second component characteristic parameters (such as dimensional tolerance, material ratio fluctuation, etc.). Now we need to load these parameters into the target generation network to generate a simulated prefabricated component model.

[0071] The target generation network is a complex neural network structure that is pre-trained to generate accurate prefabricated component models based on input feature parameters. When the server loads the first component feature parameters and the second component feature parameters into the target generation network, each neuron in the network begins to calculate according to a predetermined algorithm.

[0072] For example, in the network's input layer, different neurons receive different feature parameter values. A dedicated group of neurons is responsible for processing parameters related to geometric shapes. The network determines the general outline of the wall panel based on parameters of basic dimensions such as length, width and height. Parameters of the type and location of connecting nodes are received and processed by another group of neurons to determine how the connection structure is constructed in the 3D model.

[0073] When processing material property parameters, information such as material strength grade is used to determine the neural calculations related to the basic mechanical properties of the material. Data such as the fluctuation range in the characteristic parameters of the second component are used to adjust these basic calculation results to simulate the uncertainty in actual production.

[0074] The entire target generation network, through the connection and calculation between multiple layers of neurons, finally outputs a complete simulated assembly component model of prefabricated wall panels. This model contains information such as precise geometric shape, connection structure and material properties, and can accurately simulate the performance of real prefabricated wall panels in buildings.

[0075] In order to train the target generation network, the server first needs to determine the sample learning data sequence. In the historical data of building design and construction, the server collects multiple real prefabricated building models. These real prefabricated building models come from different construction projects and cover various types of buildings, such as residential, commercial, industrial, etc.

[0076] For example, in the area of ​​residential buildings, there are real prefabricated building models of multi-story and high-rise residential buildings. These models contain detailed building structure information, including the specific specifications, installation locations, and connection methods of various prefabricated components such as prefabricated wall panels, prefabricated floor slabs, and prefabricated beams and columns.

[0077] The real prefabricated building model of a commercial building may include prefabricated components with larger spans and a combination of prefabricated components corresponding to a more complex building space layout. For industrial buildings, their real prefabricated building models may have special structural requirements, such as higher load-bearing capacity and prefabricated component designs corresponding to special functional requirements (such as fire protection and explosion protection).

[0078] These real prefabricated building models of different types and functional requirements together constitute the sample learning data sequence, providing a rich data foundation for the training of the target generation network.

[0079] Next, the server starts to process each real prefabricated building model in the sample learning data sequence. Taking a real prefabricated building model of a multi-story residential building as an example, the server loads this building model into the basic component parsing unit of the basic generation network.

[0080] The task of the basic component analysis unit is to separate and analyze each prefabricated component model from the complex building model. For this multi-story residential building model, the basic component analysis unit will identify the prefabricated wall panels, prefabricated floor slabs, prefabricated beams and columns.

[0081] For prefabricated wall panels, it will parse out basic information such as the geometric shape, material type, and connection method of the wall panels to generate a real prefabricated component model. For example, it is parsed that a prefabricated wall panel is 3 meters high, 0.6 meters wide, made of concrete, and uses a specific connection node method. This information constitutes a complete real prefabricated component model. Similarly, for prefabricated floor slabs and prefabricated beams and columns, similar parsing operations will be performed to generate their corresponding real prefabricated component models.

[0082] The server continues to process the previously generated real assembled component model. Taking the previously parsed real assembled component model of the prefabricated wall panel as an example, this model is loaded into the basic component feature encoding unit of the basic generation network.

[0083] The basic component feature encoding unit will feature encode the real prefabricated component model of the prefabricated wall panel. For this prefabricated wall panel, the first sample component feature parameters (focused nodes) may include its precise geometric dimensions (such as 3 meters in height and 0.6 meters in width), standard connection node locations and types, standard strength grades of concrete, etc. These parameters are the key features of the prefabricated wall panel in an ideal state.

[0084] The second example component characteristic parameters (dispersed derived parameters) may include some variability factors in the actual production and installation process, such as the allowable tolerance range of geometric dimensions (height ±2 mm, width ±3 mm), the possible fluctuation range of concrete strength grade (assuming ±5%), etc.

[0085] Then, the first sample component feature parameters and the second sample component feature parameters are loaded into the initial generation unit of the basic generation network. Based on these parameters, the initial generation unit generates a simulated prefabricated component model corresponding to the real prefabricated component model (prefabricated wall panel). This simulation model retains the key features of the real component model and takes into account the uncertainty factors in actual production and installation, which is closer to the real world.

[0086] Next, the server determines the global training error for each real prefabricated component model and its corresponding simulated prefabricated component model. Taking the prefabricated wall panel as an example, the first network learning cost is first determined.

[0087] For the first network learning cost, it reflects the characteristic distance between each real prefabricated component model and the corresponding simulated prefabricated component model. For example, in terms of geometry, the height of the real prefabricated wall panel is 3 meters, while the height of the generated simulated prefabricated wall panel may have a certain deviation due to factors such as tolerance. The server determines the characteristic distance in the geometry dimension by calculating the size of this deviation (for example, using methods such as Euclidean distance). In terms of material properties, the concrete strength grade of the real wall panel and the simulated wall panel may differ due to the fluctuation range, and the characteristic distance in the material property dimension is also determined by the corresponding calculation method. The characteristic distances of various dimensions are combined to obtain the first network learning cost.

[0088] Then, the learning cost of the second network is determined. The learning cost of the second network reflects the characteristic distance between the feature distribution of the model generated by mapping the real prefabricated component model and the feature distribution of the prior model. Assuming that a certain standard feature distribution of prefabricated wall panels is specified in the feature distribution of the prior model, the feature distribution of the simulated prefabricated component model generated according to the real prefabricated component model may have deviations. For example, in terms of the distribution density of the connection nodes, the prior model has a standard distribution requirement, while the generated simulation model may be different due to various factors. The learning cost of the second network is determined by calculating the size of this deviation.

[0089] Based on the first network learning cost and the second network learning cost, a global training error is determined (for example, it may be a weighted sum of the two).

[0090] Then, the basic generative network is subjected to cyclic parameter learning based on the sample learning data sequence. In each round of learning, the server adjusts various parameters in the basic generative network (such as the connection weights between neurons) to reduce the global training error. As the number of cycles increases, the global training error gradually decreases.

[0091] When the global training error is minimized, the server terminates parameter learning. At this time, the initial generation unit when parameter learning is terminated is output as the target generation network. After such a training process, this target generation network can accurately generate a simulated assembly component model that meets the requirements based on the input component feature parameters.

[0092] In a possible implementation, step A140 includes:

[0093] Step A141: determining a first network learning cost based on each of the real prefabricated component models and the corresponding simulated prefabricated component models, wherein the first network learning cost reflects a characteristic distance between each of the real prefabricated component models and the corresponding simulated prefabricated component models.

[0094] Step A142, based on the first sample component feature parameters, the second sample component feature parameters, the focused nodes and the dispersed derivative parameters of the prior model feature distribution, determine the second network learning cost, wherein the second network learning cost reflects the feature distance between the model feature distribution generated by mapping the real prefabricated component model and the prior model feature distribution.

[0095] Step A143: determining a global training error based on the first network learning cost and the second network learning cost.

[0096] In a possible implementation manner, the correlation between any two simulated prefabricated component models in the prefabricated component model library is less than a set threshold value.

[0097] In this embodiment, when the server processes the prefabricated wall panel assembly component model, it is necessary to determine a first network learning cost for each real prefabricated wall panel component model and its corresponding simulated prefabricated wall panel component model.

[0098] For the real prefabricated wall panel, its height is 3 meters, width is 0.6 meters, and thickness is 0.2 meters. In the simulated prefabricated wall panel, due to the error simulation in the production process, the height is actually 3.005 meters, the width is 0.598 meters, and the thickness is 0.202 meters.

[0099] The server uses the Euclidean distance formula to calculate the characteristic distance in the geometric dimension. For height, the distance is |3 - 3.005| = 0.005 meters; for width, the distance is |0.6 - 0.598| = 0.002 meters; for thickness, the distance is |0.2 - 0.202| = 0.002 meters. Then these distance values ​​are squared and added and then taken the square root to get the characteristic distance value in the geometric dimension. Assume that the calculated value is 0.0054 meters.

[0100] The spacing between the upper and lower edge connection nodes of the real prefabricated wall panel is 1 meter, and the spacing between the left and right edge connection nodes is 0.5 meter. In the simulated prefabricated wall panel, the spacing between the upper and lower edge connection nodes is 1.01 meter, and the spacing between the left and right edge connection nodes is 0.49 meter.

[0101] The distance between the upper and lower edge connection nodes is calculated to be |1 - 1.01| = 0.01 meters, and the distance between the left and right edge connection nodes is |0.5 - 0.49| = 0.01 meters. Similarly, the characteristic distance value in the connection node dimension is calculated using a certain method (such as adding squares and taking the square root). Assume that the value obtained is 0.0141 meters.

[0102] The concrete strength grade of the real prefabricated wall panel is C30, and the strength grade of the simulated prefabricated wall panel is equivalent to C29.5 due to consideration of material fluctuations. The server converts the difference in strength grade into characteristic distance according to the preset calculation method. Assume that according to the relationship curve between strength grade and performance, the characteristic distance in the material property dimension is calculated to be 0.02 (the 0.02 here is the relative distance value obtained according to the specific conversion relationship).

[0103] The server integrates the characteristic distance values ​​calculated in various dimensions such as geometric shape dimension, connection node dimension and material property dimension. For example, a weighted average method can be used, assuming that the weight of the geometric shape dimension is 0.3, the weight of the connection node dimension is 0.3, and the weight of the material property dimension is 0.4. Then the first network learning cost is 0.3×0.0054 + 0.3×0.0141+ 0.4×0.02 = 0.0123 (this is just an example calculation result). This first network learning cost reflects the overall characteristic distance between each real prefabricated component model and the corresponding simulated prefabricated component model in terms of multiple important features.

[0104] The focus node of the prior model feature distribution specifies a standard size range for the geometric shape of the precast wall panel. For example, the height should be between 2.95 meters and 3.05 meters, and the width should be between 0.58 meters and 0.62 meters. For the real precast wall panel, its height is 3 meters and its width is 0.6 meters, which meets the requirements of the focus node of the prior model.

[0105] In the first sample component characteristic parameters (i.e., the focus node mapped from the real prefabricated wall panel), the height is 3 meters and the width is 0.6 meters. In the second sample component characteristic parameters of the simulated prefabricated wall panel, the height is 3.005 meters and the width is 0.598 meters.

[0106] The relationship between the deviation in calculating the height and the range specified by the prior model is (3.005 - 3) / 0.1 = 0.05 (here 0.1 is the difference between the height range of 3.05 - 2.95 specified by the prior model); the relationship between the deviation in calculating the width and the range specified by the prior model is (0.598 - 0.6) / 0.04 = - 0.05 (here 0.04 is the difference between the width range of 0.62 - 0.58 specified by the prior model).

[0107] The dispersion-derived parameters of the prior model characteristic distribution specify the tolerance range of the precast wall panel dimensions. For example, the tolerance of the height is ±0.01 meters and the tolerance of the width is ±0.02 meters.

[0108] In the second example component feature parameter, the tolerance for height is ±0.005 meters, and the tolerance for width is ±0.01 meters. Calculating the difference in the tolerance range, for height, |0.01 - 0.005| = 0.005 meters; for width, |0.02 - 0.01| = 0.01 meters.

[0109] The server calculates the deviations of the focused nodes and the decentralized derived parameters to obtain the second network learning cost. For example, the deviation values ​​of the focused nodes and the decentralized derived parameters are weighted and summed according to certain weights (assuming that the weight of the focused node deviation is 0.6 and the weight of the decentralized derived parameter deviation is 0.4). Assume that the second network learning cost calculated is 0.03 (here is an example result). This second network learning cost reflects the overall feature distance between the model feature distribution generated by mapping the real prefabricated component model and the prior model feature distribution in terms of key features.

[0110] The server has calculated that the first network learning cost is 0.0123 and the second network learning cost is 0.03. Assume that the method for determining the global training error is a simple weighted sum, the weight of the first network learning cost is 0.4, and the weight of the second network learning cost is 0.6.

[0111] The global training error is 0.4×0.0123+0.6×0.03 = 0.02192. This global training error comprehensively considers the direct feature distance between the real prefabricated component model and the simulated prefabricated component model and the feature distance between the feature distribution of the prior model. It is an important indicator for evaluating the training effect of the basic generative network.

[0112] In the prefabricated component model library maintained by the server, there are multiple simulated prefabricated component models of prefabricated wall panels. Assume that the threshold value is set to 0.1 (this value is set according to actual needs and is used to control the differences between models).

[0113] For two simulated assembly component models A and B of prefabricated wall panels, their correlation is calculated. When calculating the correlation, multiple aspects need to be considered, such as geometric shape, material properties, and connection nodes.

[0114] In terms of geometry, if the height of model A is 3 meters and the width is 0.6 meters, and the height of model B is 3.02 meters and the width is 0.59 meters. Calculate the difference ratio of height and width. For height, the difference ratio is |3 - 3.02| / 3 = 0.0067; for width, the difference ratio is |0.6 - 0.59| / 0.6 = 0.0167. The difference ratio of geometry is combined (assuming that weighted average or other methods are used) to obtain the partial value of the degree of association in geometry.

[0115] In terms of material properties, if the concrete strength grade of model A is C30 and the concrete strength grade of model B is C30.5, the difference ratio of strength grades is calculated and the partial value of the correlation in terms of material properties is obtained.

[0116] In terms of connection nodes, the difference ratios such as the distance and type of connection nodes are calculated to obtain the partial value of the association degree in terms of connection nodes.

[0117] Finally, the correlation between models A and B is obtained by combining the partial values ​​of the correlation in terms of geometry, material properties, and connection nodes (assuming that weighted summation and other methods are used). Assume that the correlation is calculated to be 0.08, which is less than the set threshold value of 0.1. Such a requirement ensures that the various simulated prefabricated component models in the model library have sufficient differences. In practical applications, when the server obtains different models from the model library for combination or analysis, this difference can provide more diverse choices to avoid models that are too similar to accurately simulate different building scenes or meet different design requirements. For example, when designing a complex prefabricated building, different types of prefabricated wall panels are required to meet different functional areas (such as load-bearing areas, non-load-bearing areas, areas with different insulation requirements, etc.). The low correlation of models in the model library can ensure that there are enough model types to choose from.

[0118] In a possible implementation, step S120 includes:

[0119] Step S121 : determining a first scale of a target simulated prefabricated component model to be generated, and constructing a candidate simulated prefabricated component model of the first scale.

[0120] Step S122: generating a first-scale target simulated prefabricated component model of the same simulated component group based on the first simulated prefabricated component model and each of the candidate simulated prefabricated component models.

[0121] In a possible implementation, step S122 includes:

[0122] Step S1221, determining an estimated correlation between each candidate simulated prefabricated component model and the first simulated prefabricated component model, each estimated correlation being within a set correlation range, and each candidate simulated prefabricated component model being used to generate a target simulated prefabricated component model.

[0123] Step S1222, performing a cyclic regularization conversion on each of the candidate simulated prefabricated component models, and after each round of regularization conversion, determining the true correlation between the candidate simulated prefabricated component model after each round of regularization conversion and the first simulated prefabricated component model, and determining the correlation error based on each of the true correlations and each of the estimated correlations.

[0124] Step S1223, during the cyclic regularization conversion process, the parameter learning is terminated when the correlation error is minimized, and the candidate simulated prefabricated component models after each round of regularization conversion when the parameter learning is terminated are output as the first-scale target simulated prefabricated component models of the same simulated component family.

[0125] In this embodiment, when the server processes the operations related to the simulation component group of the prefabricated wall panel, it first determines the first scale of the target simulation prefabricated component model to be generated. Assume that according to the needs of the construction project, for example, a large residential community construction project, it is necessary to provide multiple types of prefabricated wall panels for different building units. After analysis, it is determined that the first scale is 5 target simulation prefabricated component models.

[0126] Then, the server starts to build the first scale (i.e., 5) of candidate simulation prefabricated component models. For prefabricated wall panels, the server builds them based on existing knowledge and design rules.

[0127] For the first candidate model, in terms of geometry, based on the first simulated prefabricated component model (assuming it is a standard prefabricated wall panel with a height of 3 meters and a width of 0.6 meters), the height of this candidate model is increased by 0.1 meters to 3.1 meters, and the width remains unchanged. Such a change may be to meet the needs of some building units for higher space, or to adapt to special structural layouts.

[0128] In terms of material properties, the material remains concrete, but the mix of concrete is adjusted so that its strength grade changes from C30 to C35. This may be because higher load-bearing capacity is required in certain areas, such as the ground floor of a building or where load-bearing walls are located.

[0129] In terms of connection nodes, the connection node type of the upper and lower edges was changed from slot connection to mortise and tenon connection. At the same time, the spacing of the connection nodes was adjusted from the original uniform distribution to a distribution with slightly smaller spacing at both ends and slightly larger spacing in the middle to adapt to the new geometric shape and structural requirements.

[0130] For the second candidate model, the length was reduced from 4 meters to 3.8 meters, the width was increased to 0.62 meters, and the shape of the wall panel was changed from a rectangular to a slightly curved shape. This change may be to meet the special design requirements of the building's appearance or to adapt to a specific building curve structure.

[0131] In terms of material properties, some fiber reinforcement was added to the concrete to improve the crack resistance of the wall panels. At the same time, the type of insulation material was adjusted from polystyrene foam board to rock wool board to improve fire resistance.

[0132] In terms of connection nodes, the number of connection nodes has been increased, and the shape of the connection nodes has been changed to make them more suitable for the new geometric shape and material properties.

[0133] For the third candidate model, for the geometry, the thickness of the wall panels was increased from 0.2m to 0.25m, and some grooves and bumps were added to the surface of the wall panels, which may be to increase the friction between the connections between the wall panels or for aesthetics.

[0134] In terms of material properties, the color additives of concrete were changed to make the wall panels present different colors to meet the color matching requirements of the building appearance. In addition, the early strength development curve of concrete was adjusted so that it can reach higher strength in a shorter time to speed up the construction progress.

[0135] In terms of connection nodes, a new type of magnetic connection node is used as an auxiliary connection method, combined with the original mechanical connection method to improve the reliability of the connection.

[0136] For the fourth candidate model, in terms of geometry, the wall panel is cut into two parts, the upper part is trapezoidal and the lower part is rectangular, and an adjustable connection structure is set between the two parts so that it can be adjusted according to actual conditions at the construction site.

[0137] In terms of material properties, a self-healing concrete material was used, which can automatically repair tiny cracks in the wall panels. At the same time, the thickness distribution of the insulation material was adjusted so that it has different thicknesses in different parts of the wall panels to optimize the insulation effect.

[0138] In terms of connection nodes, a rotatable connection node is designed so that the wall panels can have a certain angle adjustment range during installation, increasing the flexibility of installation.

[0139] For the fifth candidate model, in terms of geometry, the four corners of the wall panel are designed to be rounded, and some reinforcing ribs are added to the edge of the wall panel to improve the overall strength of the wall panel.

[0140] In terms of material properties, a high-performance concrete with lower permeability and higher durability is used. At the same time, a waterproof layer is wrapped around the insulation material to prevent moisture from penetrating into the insulation layer.

[0141] In terms of connection nodes, the sealing design of the connection nodes has been optimized, and a special sealant has been used to improve the waterproof performance of the connection nodes.

[0142] Next, the server starts to determine the estimated correlation between each candidate simulated prefabricated component model and the first simulated prefabricated component model. The estimated correlation is calculated based on a series of pre-set rules and algorithms.

[0143] For the first candidate model:

[0144] In terms of geometric shape, since the height increases by 0.1 meters, the width remains unchanged, and the length remains unchanged, the change in geometric shape is relatively small compared with the first simulated prefabricated component model. The server calculates the partial value of the correlation in the geometric shape dimension as 0.8 according to the preset geometric shape correlation calculation rules (for example, calculated according to the change ratio of height, width, and length and the preset weights).

[0145] In terms of material properties, although the strength grade changes from C30 to C35, it still belongs to the category of concrete materials, and the changes in other material properties are relatively limited. According to the material property correlation calculation rules, the correlation value in the material property dimension is calculated to be 0.7.

[0146] In terms of connection nodes, the types and spacing of connection nodes have changed significantly, but considering the correlation of the overall structure, the partial value of the correlation calculated on the connection node dimension is 0.6.

[0147] Combining the partial values ​​of the correlation of these three dimensions (assuming the weighted average method is adopted, the weight of the geometry is 0.4, the weight of the material property is 0.3, and the weight of the connection node is 0.3), the estimated correlation between the first candidate model and the first simulated prefabricated component model is 0.71. This value is within the set correlation range (assuming the set correlation range is 0.6-0.9).

[0148] In the same way, calculate the estimated correlation of other candidate models:

[0149] For the second candidate model:

[0150] The geometry has changed significantly, such as the length, width and shape. The correlation value in the geometry dimension is 0.6. The material properties have also changed a lot, and the correlation value is 0.65. The connection nodes have changed significantly, and the correlation value is 0.55. The estimated correlation is 0.59.

[0151] For the third candidate model:

[0152] The geometry changes are relatively moderate, with a correlation value of 0.7. The material properties change a lot, with a correlation value of 0.6. The connection nodes have innovative changes, with a correlation value of 0.65. The estimated correlation is 0.65.

[0153] For the fourth candidate model:

[0154] The geometric shape changes greatly, and the correlation value is 0.55. The material properties have special changes, and the correlation value is 0.6. The innovative design of the connection node makes the correlation value 0.5. The estimated correlation is 0.55.

[0155] For the fifth candidate model:

[0156] The geometry has changed to some extent, and the correlation value is 0.75. The material properties have changed a lot, and the correlation value is 0.65. The optimization design of the connection node makes the correlation value 0.7. The estimated correlation is 0.7.

[0157] Then, the server starts to perform cyclic regularization conversion on each candidate simulation assembly component model. Take the first candidate model as an example.

[0158] In the first round of cyclic regularization conversion, in terms of geometry, the structural stability of the wall panels after the height increase was analyzed. According to the analysis results, it was found that due to the increase in height, the center of gravity of the wall panels moved upward, and some transverse reinforcement ribs needed to be added inside the wall panels to improve stability. After adjustment, the real correlation with the first simulated prefabricated component model in the geometry dimension was recalculated. Assuming that the original geometry correlation value caused by the increase in height was 0.8, after adjustment, due to the addition of reinforcement ribs and the change of internal structure, the geometry correlation value became 0.75.

[0159] In terms of material properties, the stress-strain performance of concrete with increased strength grade was tested. Based on the test results, it was found that the proportion of certain additives in concrete needed to be fine-tuned to optimize its long-term performance. After the adjustment, the true correlation with the first simulated prefabricated component model in the material property dimension was recalculated. Assuming that it was originally 0.7, it became 0.68 after adjustment.

[0160] In terms of connection nodes, the bearing capacity of the new mortise and tenon connection nodes was tested. It was found that the mortise and tenon connection had weak links under certain stress conditions, so the shape and size of the mortise and tenon were optimized. After optimization, the true correlation with the first simulated prefabricated component model in the connection node dimension was recalculated. Assuming that it was originally 0.6, it became 0.62 after adjustment.

[0161] Combining these three dimensions (assuming that the same weighted average method as that used to calculate the estimated correlation is used), the actual correlation after the first round of cyclic regularization conversion is 0.68.

[0162] Calculated correlation error: the true correlation is 0.68, the estimated correlation is 0.71, and the correlation difference is |0.68 -0.71| = 0.03.

[0163] In the second round of cyclic regularization transformation, in terms of geometry, the deformation of the wall panels with added reinforcement under different temperature and humidity conditions was further analyzed. It was found that the distribution of reinforcement needed to be fine-tuned to adapt to environmental changes. After the adjustment, the true correlation in the geometry dimension was recalculated. It was assumed to be changed to 0.73.

[0164] In terms of material properties, the durability of concrete after fine-tuning the additive ratio is tested. According to the test results, the content of one additive needs to be adjusted again. After the adjustment, the true correlation in the material property dimension is recalculated. Assume that it becomes 0.65.

[0165] In terms of connection nodes, fatigue tests were conducted on the optimized mortise and tenon connection nodes. It was found that a small reinforcement component needed to be added at the mortise and tenon connection. After the addition, the true correlation degree in the connection node dimension was recalculated. It was assumed to be changed to 0.65.

[0166] The actual correlation after the second round of cyclic regularization conversion is 0.68.

[0167] Calculated correlation error: the true correlation is 0.68, the estimated correlation is 0.71, and the correlation difference is |0.68 -0.71| = 0.03.

[0168] In the same way, other candidate models are transformed into regularized forms in a loop, and the true correlation and correlation error are calculated for each round.

[0169] The server continuously performs cyclic regularization transformation on each candidate model and continuously calculates the correlation error.

[0170] As the number of cycles increases, the correlation error of each candidate model gradually decreases. For example, for the first candidate model, after multiple cycles, the correlation error gradually decreases from the initial 0.03 to 0.01.

[0171] When the correlation errors of all candidate models are minimized (for example, all are less than a pre-set minimum value, such as 0.01), the server terminates parameter learning.

[0172] At this time, the candidate simulated prefabricated component models after each round of regularization conversion when parameter learning is terminated are output as the first scale (i.e., 5) of the target simulated prefabricated component models of the same simulated component family. These target models not only retain a certain degree of correlation with the first simulated prefabricated component model, but also have been optimized and adjusted to better meet the needs of different scenarios in construction projects, such as structural requirements, appearance requirements, and functional requirements of different building units. For example, in architecture, some target models can be used for the facade of a building to meet aesthetic and thermal insulation requirements; some can be used for internal partition walls to meet space division and sound insulation requirements.

[0173] In a possible implementation, step S1222 includes:

[0174] Determine a correlation degree difference between each of the actual correlation degrees and the corresponding estimated correlation degree.

[0175] A correlation error is determined based on each of the correlation differences.

[0176] In this embodiment, the previously constructed candidate simulation assembly component models of prefabricated wall panels have been converted into circular regularization, and the real correlation and estimated correlation of each candidate model have been obtained. Now, the correlation difference between them needs to be determined.

[0177] For the first candidate model, the estimated correlation between the first candidate model and the first simulated prefabricated component model was calculated to be 0.71. After the first round of cyclic regularization transformation, the actual correlation was determined to be 0.68.

[0178] Calculate the correlation difference: the actual correlation is 0.68, the estimated correlation is 0.71, and the correlation difference is 0.71 -0.68 = 0.03.

[0179] After the second round of cyclic regularization transformation, it is assumed that the true correlation becomes 0.69. At this time, the correlation difference is 0.71- 0.69 = 0.02.

[0180] After the third round of regularization, if the true correlation becomes 0.70, the correlation difference is 0.71 -0.70 = 0.01.

[0181] For the second candidate model, the estimated correlation is 0.59. Assume that after the first round of regularization transformation, the actual correlation is 0.56.

[0182] Calculate the difference in correlation: 0.59 - 0.56 = 0.03.

[0183] After the second round of cyclic regularization, assume that the true correlation becomes 0.57. The correlation difference is 0.59-0.57 = 0.02.

[0184] After the third round of regularization, if the true correlation becomes 0.58, the correlation difference is 0.59 -0.58 = 0.01.

[0185] For the third candidate model, the estimated correlation is 0.65. Assume that after the first round of cyclic regularization transformation, the actual correlation is 0.62.

[0186] Calculate the difference in correlation: 0.65 - 0.62 = 0.03.

[0187] After the second round of cyclic regularization, assume that the true correlation becomes 0.63. At this time, the correlation difference is 0.65-0.63 = 0.02.

[0188] After the third round of regularization, if the true correlation becomes 0.64, the correlation difference is 0.65 -0.64 = 0.01.

[0189] For the fourth candidate model, the estimated correlation is 0.55. Assume that after the first round of cyclic regularization transformation, the actual correlation is 0.52.

[0190] Calculate the difference in correlation: 0.55 - 0.52 = 0.03.

[0191] After the second round of cyclic regularization, assume that the true correlation becomes 0.53. At this time, the correlation difference is 0.55-0.53 = 0.02.

[0192] After the third round of regularization, if the true correlation becomes 0.54, the correlation difference is 0.55 -0.54 = 0.01.

[0193] For the fifth candidate model, the estimated correlation is 0.7. Assume that after the first round of regularization transformation, the actual correlation is 0.68.

[0194] Calculate the correlation difference: 0.7 - 0.68 = 0.02.

[0195] After the second round of cyclic regularization transformation, it is assumed that the true correlation becomes 0.69. At this time, the correlation difference is 0.7 -0.69 = 0.01.

[0196] The server can use a simple summation method to determine the correlation error. Add the correlation differences of each candidate model. For the above example, the correlation differences after the first round of regularization conversion are 0.03, 0.03, 0.03, 0.03, and 0.02. The correlation error (simple summation) is 0.03+0.03+0.03+0.03+0.02 = 0.14.

[0197] After the second round of cyclic regularization, the correlation differences are 0.02, 0.02, 0.02, 0.02, and 0.01. At this time, the correlation error (simple summation) is 0.02+0.02+0.02+0.02+0.01 = 0.09.

[0198] After the third round of regularized transformation, the correlation differences are 0.01, 0.01, 0.01, 0.01, and 0.01, respectively. The correlation error (simple summation) is 0.01+0.01+0.01+0.01+0.01 = 0.05.

[0199] As the cyclic regularization transformation proceeds, the correlation error (simple sum) gradually decreases, indicating that the true correlation of each candidate model is getting closer and closer to the estimated correlation.

[0200] The server may also use a root mean square (RMS) method to determine the correlation error.

[0201] For the first round of cyclic regularization transformation, the correlation differences are 0.03, 0.03, 0.03, 0.03, and 0.02.

[0202] First, calculate the square of each difference: 0.03² = 0.0009, 0.03² = 0.0009, 0.03² = 0.0009, 0.03² = 0.0009, 0.02² = 0.0004. Then find the average of these squared values: (0.0009 + 0.0009 +0.0009 + 0.0009 + 0.0004) / 5 = 0.0008. Finally, take the square root of the average to get the correlation error (RMS): √0.0008 ≈ 0.0283.

[0203] For the second round of regularized transformation, the correlation difference is 0.02, 0.02, 0.02, 0.02, 0.01. Calculate the square value: 0.02² = 0.0004, 0.02² = 0.0004, 0.02² = 0.0004, 0.02² = 0.0004, 0.01² = 0.0001. Calculate the average value: (0.0004 + 0.0004 + 0.0004 + 0.0004 + 0.0001) / 5 = 0.00036. Take the square root to get the correlation error (RMS): √0.00036 ≈ 0.0190.

[0204] For the third round of regularized transformation, the correlation difference is 0.01, 0.01, 0.01, 0.01, 0.01. Calculate the square value: 0.01² = 0.0001, 0.01² = 0.0001, 0.01² = 0.0001, 0.01² = 0.0001, 0.01² = 0.0001. Calculate the average value: (0.0001 + 0.0001 + 0.0001 + 0.0001 + 0.0001) / 5 = 0.0001. Take the square root to get the correlation error (RMS): √0.0001 = 0.01.

[0205] Similarly, as the cyclic regularization transformation proceeds, the correlation error (root mean square) also gradually decreases, which indicates that the overall difference between the true correlation and the estimated correlation of the candidate model is shrinking.

[0206] By determining the correlation error, the optimization effect of each candidate model in the cyclic regularization conversion process can be evaluated. When the correlation error is minimized, it can be determined that the target simulation prefabricated component model that meets the requirements is obtained.

[0207] In a possible implementation, for each of the target simulated prefabricated component models, step S130 includes:

[0208] Step S131, determining at least one random disturbance model.

[0209] Step S132, for each of the random perturbation models, perform a predefined number of loop optimizations on the target simulated prefabricated component model based on the random perturbation model, and generate a simulated prefabricated building model based on the target component model sequence generated by the last round of loop optimization.

[0210] Each round of cyclic optimization process includes: randomly perturbing and adjusting the target component model sequence generated in the previous round of cyclic process based on the random perturbation model to generate a random perturbation adjustment model, and generating a target adjustment model based on the random perturbation adjustment model generated in this round of cyclic process.

[0211] Among them, the target adjustment model of the first round of cycle process is generated based on the random perturbation adjustment model generated after the random perturbation model performs random perturbation adjustment on the target simulation prefabricated component model.

[0212] In a possible implementation, the random perturbation adjustment model generated based on the current cycle process generates a target adjustment model, including: adjusting the target adjustment model generated by the previous cycle process for structural adaptability, and generating a structural feature model that carries the characteristics of the simulation component group corresponding to the first simulation assembly component model. The random perturbation adjustment model generated by the current cycle process and the structural feature model are merged to generate the target adjustment model. The structural feature model of the first cycle process is generated after the structural adaptability adjustment is performed on the corresponding target simulation assembly component model.

[0213] In this embodiment, a target simulation assembly component model has been obtained, and the prefabricated wall panel target simulation assembly component model mentioned above is taken as an example. The server starts to determine at least one random disturbance model.

[0214] For example, for the wind disturbance model, considering that the construction site may be affected by wind, the server builds a wind disturbance model. This model includes the factors affecting the installation of prefabricated wall panels at different wind levels (such as breeze, moderate wind, and strong wind). For example, a breeze (wind speed of 1-5 m / s) may cause a slight horizontal displacement of the wall panel during installation, a moderate wind (wind speed of 6-11 m / s) will cause a larger displacement and may affect the verticality of the wall panel, and a strong wind (wind speed greater than 12 m / s) may have a serious impact on the installation stability of the wall panel, and may even require additional temporary support measures.

[0215] The wind disturbance model also takes into account the wind direction. Different wind directions (such as north wind, south wind, east wind, west wind, etc.) will cause the wall panels to be subjected to forces in different directions, thus affecting their installation position and the firmness of the connection.

[0216] For another example, in the perturbation model of worker operation errors, in actual construction, worker operation errors are inevitable. The server builds this perturbation model to simulate this situation. For example, when workers install prefabricated wall panels, there may be errors in the tightening force of the connection nodes. If the tightening force is insufficient, the connection may not be strong enough; if the tightening force is too large, the connection node or the wall panel itself may be damaged.

[0217] When workers lift wall panels, the lifting position may be inaccurate, which may cause deviations when the wall panels are placed in the predetermined position, affecting the connection accuracy with other components.

[0218] For another example, for the material property fluctuation perturbation model, although some fluctuations in material properties have been considered when constructing the prefabricated wall panel target simulation assembly component model, there may be other material property fluctuations during the actual construction process. For example, the actual setting time of concrete may be different from expectations due to environmental factors such as temperature and humidity at the construction site. This perturbation model takes into account the impact of these factors on the material properties of prefabricated wall panels.

[0219] The thermal insulation properties of insulation materials may change due to poor storage conditions at the construction site. For example, if the insulation material is damp, its thermal conductivity will increase, which will affect the insulation effect of the wall panel.

[0220] In the cyclic optimization process under the wind disturbance model, it is assumed that the predefined number of cyclic optimizations is 3, and the prefabricated wall panel target simulation prefabricated component model is taken as an example.

[0221] Based on the wind disturbance model, the target simulated prefabricated component model (prefabricated wall panel) is randomly disturbed and adjusted. For example, the current simulation is a moderate wind situation with a southerly wind. According to the calculation of the wind disturbance model, the southerly wind will cause the prefabricated wall panel to move 5 cm northward in the horizontal direction and sink 1 cm in the vertical direction. The server adjusts the position of the prefabricated wall panel based on these calculation results and generates a random disturbance adjustment model.

[0222] Then, the target simulated prefabricated component model (the original prefabricated wall panel) is structurally adapted to accommodate this possible disturbance. For example, due to the displacement and sinking of the wall panel, the connection method with adjacent components (such as prefabricated floor slabs and prefabricated beams and columns) needs to be adjusted. It may be necessary to add some adjustable gaskets at the connection nodes to ensure the tightness of the connection. In this way, a structural feature model carrying the simulated component group characteristics of the simulated component group (prefabricated wall panel group) corresponding to the first simulated prefabricated component model is generated.

[0223] Finally, the random perturbation adjustment model and the structural feature model generated in this cycle are integrated. For example, the new position information of the wall panels in the random perturbation adjustment model and the connection mode adjustment information in the structural feature model are combined to generate the target adjustment model.

[0224] Continue to perform random disturbance adjustment based on the wind disturbance model. Assume that in the second round of simulation, the wind force becomes strong and the wind direction is easterly. According to the model calculation, the wall panel will move 8 cm to the west in the horizontal direction and sink 2 cm in the vertical direction. Adjust the wall panel position again to generate a new random disturbance adjustment model.

[0225] Adjust the structural adaptability of the target adjustment model generated in the first cycle. Since the wall panel has moved and sunk again, it may be necessary to adjust the type of connection nodes or add additional support structures. For example, change some connection nodes from bolt connections to welded connections, and add temporary supports at the bottom of the wall panel. Generate a new structural feature model.

[0226] The random disturbance adjustment model of this round and the new structural feature model are integrated to generate a new target adjustment model.

[0227] In this round, it is assumed that the wind weakens to a breeze and the wind direction is northerly. According to the wind disturbance model, the wall panel will be displaced 2 cm to the south in the horizontal direction, and there will be no obvious change in the vertical direction. The wall panel position is adjusted according to the calculation results to generate a random disturbance adjustment model.

[0228] The target adjustment model generated in the second cycle is adjusted for structural adaptability. Since the displacement is small, it may only be necessary to fine-tune the tightness of the connection nodes. Generate a structural feature model.

[0229] The random disturbance adjustment model and structural feature model of this round are integrated to generate the final target component model sequence.

[0230] Based on the target component model sequence generated by this last round of optimization (the final state of the prefabricated wall panels after three rounds of optimization), the server begins to build a simulated prefabricated building model. In addition to prefabricated wall panels, the server adds other prefabricated components, such as prefabricated floor slabs, prefabricated beams and columns, according to the building design plan. According to the layout requirements of the building, the prefabricated wall panels are placed in appropriate positions and connected to the prefabricated floor slabs and prefabricated beams and columns. For example, the prefabricated wall panels are connected to the prefabricated beams and columns through pre-designed connection nodes to ensure the stability of the structure. At the same time, according to the functional requirements of the building, partition walls, doors and windows and other components are set inside the building to finally generate a complete simulated prefabricated building model.

[0231] For the loop optimization under the worker operation error disturbance model, the number of loop optimizations is also predefined as 3 times.

[0232] Based on the worker operation error perturbation model, the prefabricated wall panel target simulation assembly component model is randomly perturbed and adjusted. Assuming that the simulated worker does not tighten the connection node with sufficient force, according to the model calculation, the tightness of the connection node is only 80% of the standard value. The server adjusts the mechanical performance parameters of the wall panel connection node based on this result and generates a random perturbation adjustment model. The original target simulation assembly component model is adjusted for structural adaptability. Due to the insufficient tightness of the connection node, some auxiliary fixing devices may need to be added, such as additional buckles or binding straps. Generate a structural feature model. The random perturbation adjustment model and the structural feature model are merged to generate a target adjustment model.

[0233] The random disturbance adjustment is performed again based on the worker operation error disturbance model. This time, the simulation is about the inaccurate lifting position. It is assumed that the wall panel deviates from the predetermined position by 10 cm in the horizontal direction. The server adjusts the position coordinates of the wall panel and generates a new random disturbance adjustment model.

[0234] Adjust the structural adaptability of the target adjustment model generated in the first cycle. Due to the offset of the wall panel position, the connection position with the adjacent components needs to be adjusted. It may be necessary to reposition the connection nodes or add some transition connection components. Generate a new structural feature model. Fuse the random perturbation adjustment model of this round with the new structural feature model to generate a new target adjustment model.

[0235] In the third round of optimization, the worker was simulated to install the wall panel at an angle that was off. According to the model calculation, the wall panel was 5 degrees off the vertical. The server adjusted the angle parameters of the wall panel and generated a random perturbation adjustment model.

[0236] Adjust the structural adaptability of the target adjustment model generated by the second cycle. Due to angle deviation, it may be necessary to adjust the force distribution of the connection node. For example, add reinforcement ribs on one side of the connection node to ensure the stability of the structure. Generate a structural feature model.

[0237] The random disturbance adjustment model and structural feature model of this round are integrated to generate the final target component model sequence.

[0238] Based on the target component model sequence generated by this last round of optimization, the server constructs a simulated prefabricated building model. According to the building design plan, the adjusted prefabricated wall panels are combined with other prefabricated components (such as prefabricated floor slabs, prefabricated beams and columns, etc.). Inside the building, electrical lines, water supply and drainage pipes and other facilities are set according to functional requirements, and finally a complete simulated prefabricated building model is generated, which takes into account the impact of workers' operating errors on the building structure and function.

[0239] For the cyclic optimization process under the material property fluctuation perturbation model, the number of predefined cyclic optimizations is 3. The first round of cyclic optimization: random perturbation adjustment is performed on the target simulation prefabricated wall panel assembly component model based on the material property fluctuation perturbation model. Assume that the setting time of concrete is extended by 20% due to the low temperature at the construction site. The server adjusts the relevant parameters of the construction schedule of the wall panel according to this situation, such as postponing the start time of subsequent processes. At the same time, since the extension of the solidification time may affect the early strength development of the wall panel, the server adjusts the mechanical property parameters of the wall panel and generates a random perturbation adjustment model.

[0240] Adjust the structural adaptability of the original target simulation prefabricated component model. Since the early strength development of concrete is affected, some temporary support structures may need to be added to ensure the stability of the wall panels when they have not fully reached the design strength. Generate a structural characteristic model. Fusion the random perturbation adjustment model and the structural characteristic model to generate a target adjustment model.

[0241] Second round of optimization: random perturbation adjustment is performed again based on the material performance fluctuation perturbation model. Assume that the insulation material is damp and the thermal conductivity increases by 30%. The server adjusts the insulation performance parameters of the wall panel and generates a new random perturbation adjustment model.

[0242] The target adjustment model generated in the first cycle is adjusted for structural adaptability. Due to the decline in thermal insulation performance, it may be necessary to adjust the thermal insulation design of the building. For example, other thermal insulation measures can be added, such as adding a layer of thermal insulation film inside the wall panel. A new structural feature model is generated.

[0243] The random disturbance adjustment model of this round and the new structural feature model are integrated to generate a new target adjustment model.

[0244] Third round of optimization: In this round, assume that the shrinkage of concrete increases by 15% due to high humidity at the construction site. The server adjusts the size parameters of the wall panels according to this situation and generates a random perturbation adjustment model. Perform structural adaptability adjustments on the target adjustment model generated in the second round of cycles. Due to size changes, it may be necessary to adjust the connection method with adjacent components. For example, add some expansion joint devices at the connection nodes to accommodate the shrinkage of the wall panels. Generate a structural feature model. The random perturbation adjustment model and the structural feature model of this round are merged to generate the final target component model sequence.

[0245] Based on the target component model sequence generated by this last round of optimization, the server constructs a simulated prefabricated building model. In architectural design, the influence of material performance fluctuations on the overall performance of the building is taken into account, and the selection and layout of other prefabricated components are reasonably arranged. For example, the energy-saving design of the building is adjusted according to the changes in thermal insulation performance, and the structural safety assessment of the building is adjusted according to the changes in concrete performance, etc., and finally a complete simulated prefabricated building model is generated.

[0246] Further, Figure 2 FIG. 1 is a schematic diagram showing a hardware structure of a BIM service system 100 for implementing the method provided in an embodiment of the present application. Figure 2As shown, the BIM service system 100 may include at least one processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 and a controller 108 for communication functions. It can be understood by those skilled in the art that Figure 2 The structure shown is only for illustration and does not limit the structure of the BIM service system 100. For example, the BIM service system 100 may also include Figure 2 More or fewer components as shown, or with Figure 2 Different configurations shown.

[0247] The memory 104 can be used to store software programs and modules of application software, such as program instructions corresponding to the above-mentioned method embodiments in the embodiments of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, to implement the above-mentioned BIM-based prefabricated component management method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the BIM service system 100 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0248] The transmission device 106 is used to obtain or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the BIM service system 100. In one example, the transmission device 106 includes a network adapter, which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency module, which is used to communicate with the Internet wirelessly.

[0249] It should be noted that the order of the above embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. The above describes specific embodiments of the embodiments of the present application. Other embodiments are within the scope of the attached claims. In some cases, the exceptions or steps recorded in the claims can be performed according to an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or can be advantageous.

[0250] Each embodiment in the embodiments of the present application is described in a progressive manner, and the parts that are consistent and similar between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the above different embodiments, since they are basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0251] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program. The above program may be stored in a computer-readable storage medium, and the above storage medium may be a read-only memory, a disk or an optical disk, etc.

Claims

1. A method for managing assembled components based on BIM, characterized in that: Applied to the BIM service system, the method includes: Acquire a first simulated prefabricated component model from a prefabricated component model library, wherein the prefabricated component model library includes at least one simulated prefabricated component model; Based on the first simulated prefabricated component model, generating at least one target simulated prefabricated component model of the same simulated component group; For each of the target simulated prefabricated component models, at least one simulated prefabricated building model is generated based on the target simulated prefabricated component model, and the simulated prefabricated building models generated based on each of the target simulated prefabricated component models correspond to the same simulated component group; wherein the step of determining each simulated prefabricated component model in the prefabricated component model library includes: Determine a first component characteristic parameter and a second component characteristic parameter, wherein the first component characteristic parameter and the second component characteristic parameter respectively reflect a focused node and a dispersed derivative parameter of a model characteristic distribution corresponding to a group of assembled component models; A simulated assembled component model is generated based on the first component characteristic parameter and the second component characteristic parameter.

2. The BIM-based prefabricated component management method according to claim 1, characterized in that: The generating a simulated assembled component model based on the first component characteristic parameter and the second component characteristic parameter comprises: Loading the first component characteristic parameter and the second component characteristic parameter into a target generation network to generate a simulated assembled component model; The training step of the target generation network includes: Determine a sample learning data sequence, wherein the sample learning data sequence includes a plurality of real prefabricated building models; Loading each of the real prefabricated building models into a basic component parsing unit of a basic generation network to generate a real prefabricated component model; For each of the real prefabricated component models, the real prefabricated component model is loaded into the basic component feature encoding unit of the basic generation network to generate a first sample component feature parameter and a second sample component feature parameter, and the first sample component feature parameter and the second sample component feature parameter are loaded into the initial generation unit of the basic generation network to generate a simulated prefabricated component model corresponding to the real prefabricated component model; the first sample component feature parameter and the second sample component feature parameter respectively reflect the focused node and the dispersed derivative parameter after the real prefabricated component model is mapped to the model feature distribution; A global training error is determined based on each of the real prefabricated component models and the corresponding simulated prefabricated component models, and cyclic parameter learning is performed on the basic generation network based on the sample learning data sequence until the parameter learning is terminated when the global training error is minimized, and the initial generation unit when the parameter learning is terminated is output as the target generation network.

3. The BIM-based prefabricated component management method according to claim 2, characterized in that: The determining of the global training error based on each of the real prefabricated component models and the corresponding simulated prefabricated component models includes: Determining a first network learning cost based on each of the real prefabricated component models and the corresponding simulated prefabricated component models, wherein the first network learning cost reflects a characteristic distance between each of the real prefabricated component models and the corresponding simulated prefabricated component models; Determine a second network learning cost based on the first sample component feature parameter, the second sample component feature parameter, and the focus node and the dispersed derivative parameter of the prior model feature distribution, wherein the second network learning cost reflects the feature distance between the model feature distribution generated by mapping the real prefabricated component model and the prior model feature distribution; A global training error is determined based on the first network learning cost and the second network learning cost.

4. The BIM-based prefabricated component management method according to claim 1, characterized in that: The correlation degree between any two simulated prefabricated component models in the prefabricated component model library is less than a set threshold value.

5. The BIM-based prefabricated component management method according to claim 1, characterized in that: The step of generating at least one target simulated prefabricated component model of the same simulated component group based on the first simulated prefabricated component model comprises: Determine a first scale of a target simulated prefabricated component model to be generated, and construct a candidate simulated prefabricated component model of the first scale; Based on the first simulated prefabricated component model and each of the candidate simulated prefabricated component models, a first-scale target simulated prefabricated component model of the same simulated component group is generated.

6. The BIM-based prefabricated component management method according to claim 5, characterized in that: The step of generating a first-scale target simulated prefabricated component model of the same simulated component group based on the first simulated prefabricated component model and each of the candidate simulated prefabricated component models comprises: Determining an estimated correlation between each of the candidate simulated prefabricated component models and the first simulated prefabricated component model, each of the estimated correlations being within a set correlation range, and each of the candidate simulated prefabricated component models being used to generate a target simulated prefabricated component model; Performing a cyclic regularization conversion on each of the candidate simulated prefabricated component models, determining after each round of regularization conversion the true correlation between the candidate simulated prefabricated component model after each round of regularization conversion and the first simulated prefabricated component model, and determining a correlation error based on each of the true correlations and each of the estimated correlations; In the cyclic regularization conversion process, parameter learning is terminated when the correlation error is minimized, and the candidate simulated prefabricated component models after each round of regularization conversion when the parameter learning is terminated are output as the first-scale target simulated prefabricated component models of the same simulated component group.

7. The BIM-based prefabricated component management method according to claim 6, characterized in that: The determining of the correlation error based on each of the real correlations and each of the estimated correlations comprises: Determining a correlation difference between each of the actual correlations and the corresponding estimated correlation; A correlation error is determined based on each of the correlation differences.

8. The BIM-based prefabricated component management method according to claim 1, characterized in that: For each of the target simulated prefabricated component models, generating at least one simulated prefabricated building model based on the target simulated prefabricated component model includes: determining at least one random perturbation model; For each of the random perturbation models, a predefined number of cyclic optimizations are performed on the target simulated prefabricated component model based on the random perturbation model, and a simulated prefabricated building model is generated based on a target component model sequence generated by the last round of cyclic optimization; Each round of optimization process includes: Based on the random disturbance model, a random disturbance adjustment is performed on the target component model sequence generated in the previous cycle process to generate a random disturbance adjustment model, and a target adjustment model is generated based on the random disturbance adjustment model generated in the current cycle process; Among them, the target adjustment model of the first round of cycle process is generated based on the random perturbation adjustment model generated after the random perturbation model performs random perturbation adjustment on the target simulation prefabricated component model.

9. The BIM-based prefabricated component management method according to claim 8, characterized in that: The generating of the target adjustment model based on the random disturbance adjustment model generated in the current cycle process includes: Performing structural adaptability adjustment on the target adjustment model generated in the previous cycle process to generate a structural feature model carrying the simulated component group features of the simulated component group corresponding to the first simulated prefabricated component model; The random disturbance adjustment model generated in this round of cycle process and the structural characteristic model are integrated to generate a target adjustment model; wherein the structural characteristic model of the first round of cycle process is generated after the structural adaptability adjustment of the corresponding target simulation prefabricated component model.

10. A BIM service system, characterized in that: The BIM service system includes a processor and a readable storage medium, wherein the readable storage medium stores a program, and when the program is executed by the processor, the BIM-based prefabricated component management method described in any one of claims 1 to 9 is implemented.

Images