Building information management system based on BIM technology

By using three-dimensional grid space and preset three-dimensional convolutional neural networks in the BIM system to identify the type of target entity and automatically supplement building information, the application problem of BIM technology in the absence of data is solved, and the efficiency and quality of building information management are improved. .

CN118941740AActive Publication Date: 2024-11-12SHENZHEN XIANGWUXIANG INFORMATION TECH CO LTD

Patent Information

Application Number
CN202410983606.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-22
Publication Date
2024-11-12
Estimated Expiration
2044-07-22

AI Technical Summary

Technical Problem

BIM technology cannot be used normally in the absence of data, resulting in the inability to guarantee the integrity of information entry, limiting its promotion speed and breadth in the construction industry.

Method used

A building information management system based on BIM technology was designed, including data analysis module, model building module, data encoding module, type identification module and information supplement module, and the target entity is identified through three-dimensional grid space and preset three-dimensional convolutional neural network. and automatically supplement building information based on this category.

Benefits of technology

The normal application of the BIM system in the absence of data is realized. By automatically filling the building information, the design and construction process are optimized, and the overall quality and efficiency are improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118941740A_ABST
    Figure CN118941740A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of building data digital management, in particular to a building information management system based on a BIM (Building Information Modeling) technology, which is characterized in that shape modeling data of a target entity is obtained through a data analysis module, and then a target entity model is established in a three-dimensional grid space through a model establishment module; a data coding module is used for coding to obtain a three-dimensional input matrix, then a type recognition module is used for recognizing and obtaining the type of a target entity by using a preset three-dimensional convolutional neural network, and finally building information of the target entity is supplemented through an information supplementing module. Compared with the prior art, coding is completed by using the coverage relationship between the three-dimensional grid space and the target entity model, so that identification of entity types can be completed through the three-dimensional convolutional neural network, and other information missing by the target entity is automatically supplemented based on the types of the target entity; the problem that the BIM technology cannot be normally applied under the condition of data missing in the prior art is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of digital management of building data, and in particular to a building information management system based on BIM technology. Background Art

[0002] Building Information Modeling (BIM) is a method of comprehensively managing construction projects through digital means. It is not only a design tool, but also an information integration platform for the entire building life cycle. Through BIM, construction professionals can share, manage and analyze building-related data at all stages of the project, thereby improving design quality, optimizing construction processes and reducing operating costs.

[0003] Although BIM contains rich data resources, in actual application, the technical levels of practitioners vary greatly, and the actual situation of construction projects is complex and changeable. This leads to the inability to ensure the integrity of information input, and data loss has become a common problem. The lack of complete data will seriously hinder the normal application of BIM technology and limit the speed and breadth of its promotion in the construction industry.

[0004] Therefore, people are in urgent need of a building information management system based on BIM technology that can work in the absence of data. Summary of the invention

[0005] Therefore, the present invention provides a building information management system based on BIM technology to solve the problem in the prior art that BIM technology cannot be normally applied when data is missing.

[0006] The present invention provides a building information management system based on BIM technology, comprising:

[0007] A data analysis module is used to obtain shape modeling data of a target entity based on a BIM model;

[0008] A model building module is used to build a three-dimensional grid space and to build a model in the three-dimensional grid space based on the shape modeling data to obtain a target entity model in the three-dimensional grid space;

[0009] A data encoding module is used to encode according to the covering relationship between the three-dimensional grid space and the target entity model to obtain a three-dimensional input matrix, wherein each element in the three-dimensional input matrix corresponds to a grid in the three-dimensional grid space, and the value of each element in the three-dimensional input matrix is ​​used to represent a part of the target entity model in the corresponding grid;

[0010] A category recognition module, used to input the three-dimensional input matrix into a preset three-dimensional convolutional neural network to obtain the target entity category output by the preset three-dimensional convolutional neural network;

[0011] The information supplement module is used to supplement the building information of the target entity in the BIM system based on the target entity type.

[0012] The present invention also provides a preferred solution: based on the BIM model, obtaining shape modeling data of the target entity, including:

[0013] Obtain input range selection data representing the target entity, and obtain BIM shape data of the target entity according to the BIM model based on the range selection data;

[0014] Based on the BIM shape data, obtain the general model data of the target entity in STEP format;

[0015] The entity number in the general model data, the position information and shape information corresponding to each entity number are obtained as the shape modeling data of the target entity.

[0016] The present invention also provides a preferred solution: establishing a three-dimensional grid space, and modeling in the three-dimensional grid space based on the shape modeling data to obtain a target entity model in the three-dimensional grid space, including:

[0017] Establishing a three-dimensional grid space and obtaining dimension data of the three-dimensional grid space;

[0018] Based on the size data of the three-dimensional grid space, the position information and the shape information in the shape modeling data, coordinate transformation is performed on the position information to obtain standard position data;

[0019] Based on the size data of the three-dimensional grid space, the position information and the shape information in the shape modeling data, the shape information is scaled to obtain standard shape data;

[0020] Based on the entity number, the standard position information and the standard shape information corresponding to each entity number, modeling is performed in the three-dimensional grid space to obtain a target entity model in the three-dimensional grid space.

[0021] The present invention also provides a preferred solution: encoding is performed according to the covering relationship between the three-dimensional grid space and the target entity model to obtain a three-dimensional input matrix, including:

[0022] Obtain a target grid, where the target grid is a grid in a three-dimensional grid space;

[0023] The spatial proportion of the target entity model in the target grid is calculated and normalized to obtain the element value of the three-dimensional input matrix corresponding to the target grid.

[0024] The present invention also provides a preferred solution: the preset three-dimensional convolutional neural network includes an input layer, at least one three-dimensional convolutional layer, at least one pooling layer, at least one fully connected layer and an output layer, wherein the input layer is located at the first layer of the preset three-dimensional convolutional neural network, a three-dimensional convolutional layer is connected after the input layer, the output layer is located at the last layer of the preset three-dimensional convolutional neural network, a fully connected layer is connected before the output layer, and other three-dimensional convolutional layers, pooling layers and fully connected layers are arranged and combined and connected in sequence between the three-dimensional convolutional layer after the input layer and the fully connected layer before the output layer.

[0025] The present invention also provides a preferred solution: the three-dimensional convolution layer after the input layer includes a shape three-dimensional convolution layer, a long edge convolution layer, a wide edge convolution layer, a high edge convolution layer and a fusion layer, wherein the input ends of the shape three-dimensional convolution layer, the long edge convolution layer, the wide edge convolution layer and the high edge convolution layer are all connected to the input layer, and the output ends of the shape three-dimensional convolution layer, the long edge convolution layer, the wide edge convolution layer and the high edge convolution layer are all connected to the input end of the fusion layer, wherein the shape three-dimensional convolution layer is used to detect the shape features of the target entity model, the long edge convolution layer is used to detect the shape edge features of the target entity model in the long dimension, the wide edge convolution layer is used to detect the shape edge features of the target entity model in the wide dimension, and the high edge convolution layer is used to detect the shape edge features of the target entity model in the high dimension, and the fusion layer is used to superimpose and fuse the output results of the long edge convolution layer, the wide edge convolution layer and the high edge convolution layer into the output results of the shape three-dimensional convolution layer in the three dimensions of length, width and height, and output a first convolution three-dimensional matrix.

[0026] The present invention also provides a preferred solution: the dimensions of the three-dimensional grid space in terms of length, width and height are all n; the dimensions of the convolution kernel of the shape three-dimensional convolution layer in terms of length, width and height are all a, wherein a and n are both positive integers and a is less than n, the convolution kernel of the shape three-dimensional convolution layer is used to move step by step in the length, width and height dimensions of the input three-dimensional matrix, and the shape three-dimensional convolution layer is used to output a shape feature three-dimensional matrix; the dimensions of the convolution kernel of the long edge convolution layer in terms of length, width and height are n, a and a respectively, and the convolution kernel of the long edge convolution layer is used to step by step in the width and height dimensions of the input three-dimensional matrix. The long edge convolution layer is used to output a two-dimensional matrix of long edge features; the sizes of the convolution kernel of the wide edge convolution layer in the length, width and height are a, n and a respectively, and the convolution kernel of the wide edge convolution layer is used to move in steps in the length and height of the input three-dimensional matrix, and the long edge convolution layer is used to output a two-dimensional matrix of wide edge features; the sizes of the convolution kernel of the high edge convolution layer in the length, width and height are a, a and n respectively, and the convolution kernel of the high edge convolution layer is used to move in steps in the length and width of the input three-dimensional matrix, and the high edge convolution layer is used to output a two-dimensional matrix of high edge features.

[0027] The present invention also provides a preferred solution: the fusion layer is used to superimpose and fuse the long edge feature two-dimensional matrix, the wide edge feature two-dimensional matrix and the high edge feature two-dimensional matrix into the shape feature three-dimensional matrix in the three dimensions of length, width and height, and output the first convolution three-dimensional matrix. The corresponding formula is:

[0028]

[0029] in, Represents the values ​​of the elements whose coordinates in length, width and height are i, j and k in the first convolution three-dimensional matrix, Represents the values ​​of the elements whose coordinates in the three dimensions of length, width and height in the three-dimensional matrix of shape features are i, j and k respectively. Represents the value of the element with coordinates j and k in the width and height dimensions in the two-dimensional matrix of the long edge feature, Represents the value of the element whose coordinates in the length and height dimensions are i and k in the two-dimensional matrix of the wide edge feature. Represents the value of the element with coordinates i and j in the length and width dimensions in the two-dimensional matrix of the high edge feature, and w0 represents the element in the fusion layer Corresponding nodes and shapes of elements in the 3D convolutional layer The weight values ​​between the corresponding nodes, w1 represents the elements in the fusion layer The corresponding nodes and elements in the long edge convolution layer The weight values ​​between the corresponding nodes, w2 represents the elements in the fusion layer The corresponding nodes and elements in the wide edge convolution layer The weight values ​​between the corresponding nodes, w3 represents the elements in the fusion layer The corresponding nodes and elements in the high edge convolution layer The corresponding weight value between nodes, b i,j,k Represents the elements in the fusion layer The corresponding node bias value, f() is the preset activation function, and m is the number of elements in the long dimension of the shape feature three-dimensional matrix.

[0030] The present invention also provides a preferred solution: the output layer includes multiple output nodes, and the multiple output nodes are respectively connected to multiple nodes in the fully connected layer adjacent to the output layer. The number of output nodes is equal to the preset number of target entity types, each output node corresponds to a preset type, and the output node is used to output the probability value of the target entity type being the corresponding preset type.

[0031] The present invention also provides a preferred solution: the building information includes material information, attribute information, cost information, space information and structure information; based on the target entity type, the building information of the target entity is supplemented in the BIM system, including:

[0032] Retrieve entities of the same type as the target entity in the BIM system;

[0033] Obtain material information of entities of the same type as the target entity in the BIM system as the material information of the target entity;

[0034] Based on the material information of the target entity, the attribute information, cost information, space information and structure information of the target entity are supplemented.

[0035] The beneficial effects of adopting the above embodiment are:

[0036] The present invention provides a building information management system based on BIM technology, which obtains shape modeling data of a target entity based on a BIM model through a data parsing module, then establishes a three-dimensional grid space and models in the three-dimensional grid space through a model building module to obtain a target entity model, and then uses a data encoding module to encode according to the covering relationship between the three-dimensional grid space and the target entity model to obtain a three-dimensional input matrix, and then uses a type recognition module to use the three-dimensional input matrix and a preset three-dimensional convolutional neural network to identify the target entity type, and finally uses an information supplementation module to supplement the building information of the target entity in the BIM system based on the target entity type. Compared with the prior art, the present invention is based on the most important and basic shape information in the BIM system, that is, shape modeling data, modeling in the three-dimensional grid space, and uses the covering relationship between the three-dimensional grid space and the target entity model to complete the encoding of the entity shape, so that the entity type can be identified through the three-dimensional convolutional neural network, and then other missing information of the target entity can be automatically supplemented based on the type of the target entity, so that the entire BIM system can temporarily start working based on the automatically filled building information, and then change it when the subsequent information is entered, solving the problem that the BIM technology in the prior art cannot be normally applied in the case of data missing. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A system architecture diagram of an embodiment of a building information management system based on BIM technology provided by the present invention;

[0038] Figure 2 This is a schematic diagram of the structure of a simplified first three-dimensional convolutional layer in the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] Combination Figure 1 As shown, a specific embodiment of the present invention discloses a building information management system based on BIM technology, including:

[0041] The data analysis module 110 is used to obtain shape modeling data of the target entity based on the BIM model;

[0042] A model building module 120 is used to build a three-dimensional grid space and model in the three-dimensional grid space based on the shape modeling data to obtain a target entity model in the three-dimensional grid space;

[0043] The data encoding module 130 is used to encode according to the covering relationship between the three-dimensional grid space and the target entity model to obtain a three-dimensional input matrix, wherein each element in the three-dimensional input matrix corresponds to a grid in the three-dimensional grid space, and the value of each element in the three-dimensional input matrix is ​​used to represent a part of the target entity model in the corresponding grid;

[0044] A category recognition module 140 is used to input the three-dimensional input matrix into a preset three-dimensional convolutional neural network to obtain the target entity category output by the preset three-dimensional convolutional neural network;

[0045] The information supplement module 150 is used to supplement the building information of the target entity in the BIM system based on the target entity type.

[0046] In the above process, the three-dimensional grid space is a virtual space established in the computer system and is divided into multiple grids. By establishing a target entity model in the three-dimensional grid space, the shape of the target entity can be expressed in the form of a vector, that is, a three-dimensional input matrix.

[0047] A three-dimensional input matrix is ​​a combination of numbers in three dimensions, which can be represented by a two-dimensional matrix consisting of multiple one-dimensional vectors, for example:

[0048]

[0049] The above formula can be regarded as a 2×2×2 three-dimensional input matrix, corresponding to a 2×2×2 three-dimensional grid space. Each element in the above three-dimensional input matrix corresponds to a grid in a three-dimensional grid space, where 0 indicates that there is no entity in the grid, and 1 indicates that there is an entity in the grid. Therefore, the above formula can also be understood as that there is a 1×1×1 block model in the 2×2×2 three-dimensional grid space.

[0050] Through the above means, the shape information of a target entity can be represented in digital form, which can then be used as the input of the neural network. By utilizing the ability of the neural network to handle complex nonlinear problems, the type of target entity can be identified based on the shape information, such as pipes, walls or steel support structures.

[0051] Shape information is the basic carrier of other data information in the BIM model. In most cases, shape information will not be missing in the BIM model. After the type of the target entity is identified through shape information, other information can be supplemented based on the type through other existing relevant materials. For example, after knowing that the target entity is an exterior wall pipe, its material is most likely metal or plastic. Based on this inference, the cost, stress and other construction information of the pipe can be further automatically filled in to ensure the safety and reliability of the pipe during the design and construction stages. This type inference and selection process based on shape information can help optimize the design and construction process of construction projects and improve overall quality and efficiency.

[0052] Further, in a preferred embodiment, the step performed by the data parsing module 110 is to obtain shape modeling data of the target entity based on the BIM model, specifically including:

[0053] Obtain input range selection data representing the target entity, and obtain BIM shape data of the target entity according to the BIM model based on the range selection data;

[0054] Based on the BIM shape data, obtain the general model data of the target entity in STEP format;

[0055] The entity number in the general model data, the position information and shape information corresponding to each entity number are obtained as the shape modeling data of the target entity.

[0056] The scope selection data in the above process is the scope of the entity that is manually specified in the BIM model and hopes to supplement the data.

[0057] The STEP (Standard for the Exchange of Product model data) format in the above process is part of the ISO 10303 standard and is used to represent product data shapes. It is a universal data format that is widely used in the fields of CAD (Computer Aided Design) and BIM (Building Information Modeling). STEP files mainly use text format to facilitate the description of complex three-dimensional models and related information.

[0058] STEP files usually include the following main parts:

[0059] 1. Header Section: Contains the file description information, such as file name, creation time, author, etc. It also indicates the model standard used by the file (such as IFC4).

[0060] 2. Data Section: Contains the actual model data, expressed in the form of STEP entities. Each entity has a unique number (#1, #2, etc., i.e., the entity number).

[0061] 3. Entities and attributes: Specific information of each entity, such as type, attribute value, relationship, etc. Entities can contain information such as geometric representation, text attributes, relationship links, etc.

[0062] To extract specific information (such as entity number, location information, and shape information) from a STEP file, you can follow these steps:

[0063] 1. Identify the entity number:

[0064] Entities are identified in STEP files by unique numbers (e.g. #1, #2). These numbers are defined in the Data Section and are used throughout the file to reference and identify specific building elements or attributes.

[0065] 2. Find geometry and position information:

[0066] Geometric information and position information are usually included in the specific entity description. For example, for an entity representing a wall, its attributes may include geometric shape (such as size, coordinates), position information (such as relationship with other components), etc.

[0067] 3. Analyze geometric shape information:

[0068] The geometric shape information may involve coordinate data, dimensions, shape description, etc. This information is defined according to the IFC standard and can be found in the entity properties of the STEP file.

[0069] In this way, the entity number, the position information corresponding to each entity number and the shape information are extracted as the shape modeling data of the target entity.

[0070] STEP file is currently the most common model data representation format. The STEP file format can even be used directly in the BIM system. In this embodiment, the STEP file is used as a bridge. On the one hand, the versatility of STEP is utilized to make this embodiment easier to implement. At the same time, the clear structure of the STEP file is utilized to realize the extraction of shape modeling data.

[0071] Further, in a preferred embodiment, the steps performed by the model building module 120: establishing a three-dimensional grid space, and modeling in the three-dimensional grid space based on the shape modeling data to obtain a target entity model in the three-dimensional grid space, specifically include:

[0072] Establishing a three-dimensional grid space and obtaining dimension data of the three-dimensional grid space;

[0073] Based on the size data of the three-dimensional grid space, the position information and the shape information in the shape modeling data, coordinate transformation is performed on the position information to obtain standard position data;

[0074] Based on the size data of the three-dimensional grid space, the position information and the shape information in the shape modeling data, the shape information is scaled to obtain standard shape data;

[0075] Based on the entity number, the standard position information and the standard shape information corresponding to each entity number, modeling is performed in the three-dimensional grid space to obtain a target entity model in the three-dimensional grid space.

[0076] The above process corrects the shape modeling data based on the three-dimensional grid space by means of coordinate transformation and size scaling to obtain standard position information and standard shape information with uniform specifications. In this way, entities of any specifications can be encoded through the three-dimensional grid space, and thus recognized through a preset three-dimensional convolutional neural network model, and only one preset three-dimensional convolutional neural network model needs to be trained.

[0077] After the target entity model is established, it can be encoded according to the coverage relationship between the three-dimensional grid space and the target entity model. For example, if the target entity model exists in a three-dimensional grid, then the element value corresponding to the three-dimensional grid in the three-dimensional input matrix can be 1, otherwise it can be 0.

[0078] Obviously, the above method will lose the information of the target entity model surface. For example, if a part on the edge of the target entity model is not filled with the three-dimensional grid, then setting the element value corresponding to the three-dimensional grid to 1 will cause shape error.

[0079] Therefore, further, in a preferred embodiment, the step performed by the data encoding module 130 is to encode according to the covering relationship between the three-dimensional grid space and the target entity model to obtain a three-dimensional input matrix, specifically including:

[0080] Obtain a target grid, where the target grid is a grid in a three-dimensional grid space;

[0081] The spatial proportion of the target entity model in the target grid is calculated and normalized to obtain the element value of the three-dimensional input matrix corresponding to the target grid.

[0082] The above design further describes the edge features of the target entity model by means of the spatial proportion of the target entity model, thereby retaining more shape information in the three-dimensional input matrix, making the features described more accurate, thereby improving the accuracy of the preset three-dimensional convolutional neural network output.

[0083] Furthermore, in a preferred embodiment, in the above-mentioned category recognition module 140, a specific structure of a three-dimensional convolutional neural network is preset, including: an input layer, at least one three-dimensional convolutional layer, at least one pooling layer, at least one fully connected layer and an output layer, wherein the input layer is located at the first layer of the preset three-dimensional convolutional neural network, a three-dimensional convolutional layer is connected after the input layer, the output layer is located at the last layer of the preset three-dimensional convolutional neural network, a fully connected layer is connected before the output layer, and other three-dimensional convolutional layers, pooling layers and fully connected layers are arranged and combined and connected in sequence between the three-dimensional convolutional layer after the input layer and the fully connected layer before the output layer.

[0084] Among them, the input layer and output layer are used to input and output data respectively, the three-dimensional convolution layer is used to extract features from the three-dimensional input matrix, the pooling layer is used to reduce the magnitude of the data to reduce the pressure of the preset three-dimensional convolutional neural network operation and training, while improving the running speed, and the fully connected layer can be understood as a complex nonlinear analysis of the features extracted by the three-dimensional convolution layer to obtain the results. It can be understood that in the above structure, the other three-dimensional convolution layers, pooling layers and fully connected layers in the middle part can be flexibly combined and arranged according to specific circumstances.

[0085] The three-dimensional convolution layer after the input layer is the first three-dimensional convolution layer for feature extraction, and its accuracy affects the accuracy of subsequent processing. It can use a preset three-dimensional convolution kernel for convolution to achieve feature extraction. The present invention also provides a preferred embodiment for improving the first three-dimensional convolution layer to obtain a more accurate feature extraction effect.

[0086] Specifically, in a preferred embodiment, the three-dimensional convolution layer after the input layer includes a shape three-dimensional convolution layer, a long edge convolution layer, a wide edge convolution layer, a high edge convolution layer and a fusion layer, wherein the input ends of the shape three-dimensional convolution layer, the long edge convolution layer, the wide edge convolution layer and the high edge convolution layer are all connected to the input layer, and the output ends of the shape three-dimensional convolution layer, the long edge convolution layer, the wide edge convolution layer and the high edge convolution layer are all connected to the input end of the fusion layer, wherein the shape three-dimensional convolution layer is used to detect the shape features of the target entity model, the long edge convolution layer is used to detect the shape edge features of the target entity model in the long dimension, the wide edge convolution layer is used to detect the shape edge features of the target entity model in the wide dimension, and the high edge convolution layer is used to detect the shape edge features of the target entity model in the high dimension, and the fusion layer is used to superimpose and fuse the output results of the long edge convolution layer, the wide edge convolution layer and the high edge convolution layer into the output results of the shape three-dimensional convolution layer in the three dimensions of length, width and height, and output a first convolution three-dimensional matrix.

[0087] It can be understood that the three dimensions of length, width and height in the above process can be understood as three orthogonal directions in the three-dimensional grid space, and can also be understood as three dimensions of digital arrangement in the three-dimensional input matrix. Since the data of the edge of the model is processed more accurately by the spatial proportion of the target entity model when encoding the three-dimensional input matrix in the previous text, in order to amplify this accuracy, this embodiment further divides the first three-dimensional convolution layer into a shape three-dimensional convolution layer mainly used to extract the shape and a long edge convolution layer, a wide edge convolution layer and a high edge convolution layer respectively used to extract edge features of three dimensions. This design can be understood as further correcting the extracted shape features through the edge information of the three views of the target entity model. This design can greatly improve the effective information amount of feature extraction and improve the accuracy of the entire preset three-dimensional convolutional neural network.

[0088] The above-mentioned shaped three-dimensional convolution layer, long edge convolution layer, wide edge convolution layer and high edge convolution layer can all use the convolution kernel of the same specification (but the specific values ​​of the convolution kernel are different) to convolve the three-dimensional input matrix, thereby ensuring that the shaped three-dimensional convolution layer, long edge convolution layer, wide edge convolution layer and high edge convolution layer all output data of the same specification (that is, the three-dimensional matrix of the same specification), thereby ensuring that the subsequent fusion layer can be fused and superimposed.

[0089] But obviously, the above method will greatly increase the amount of data in the first three-dimensional convolutional layer, so this embodiment once again improves the above method.

[0090] Specifically, in a preferred embodiment, the dimensions of the three-dimensional grid space in length, width and height are all n; the dimensions of the convolution kernel of the shape three-dimensional convolution layer in length, width and height are all a, where a and n are both positive integers and a is less than n, and the convolution kernel of the shape three-dimensional convolution layer is used to move step by step in the length, width and height of the input three-dimensional matrix, and the shape three-dimensional convolution layer is used to output a shape feature three-dimensional matrix; the dimensions of the convolution kernel of the long edge convolution layer in length, width and height are n, a and a respectively, and the convolution kernel of the long edge convolution layer is used to move step by step in the width and height of the input three-dimensional matrix. The long edge convolution layer is used to output a two-dimensional matrix of long edge features; the sizes of the convolution kernel of the wide edge convolution layer in the length, width and height are a, n and a respectively, and the convolution kernel of the wide edge convolution layer is used to move in steps in the length and height dimensions of the input three-dimensional matrix, and the long edge convolution layer is used to output a two-dimensional matrix of wide edge features; the sizes of the convolution kernel of the high edge convolution layer in the length, width and height are a, a and n respectively, and the convolution kernel of the high edge convolution layer is used to move in steps in the length and width dimensions of the input three-dimensional matrix, and the high edge convolution layer is used to output a two-dimensional matrix of high edge features.

[0091] The above design makes the convolution kernels in the long edge convolution layer, wide edge convolution layer and high edge convolution layer occupy the dimension of the three-dimensional input matrix in one dimension respectively. In this way, the long edge convolution layer, wide edge convolution layer and high edge convolution layer actually perform two-dimensional convolution, which greatly reduces the amount of data in the preset three-dimensional convolutional neural network. And because the long edge convolution layer, wide edge convolution layer and high edge convolution layer themselves are also used to extract edge features, this design does not actually affect the accuracy of the overall feature extraction. While simplifying the network structure, it can also ensure accuracy. Figure 2 A simplified schematic diagram of the structure of the first three-dimensional convolutional layer.

[0092] Furthermore, in a preferred embodiment, the fusion layer is used to superimpose and fuse the long edge feature two-dimensional matrix, the wide edge feature two-dimensional matrix and the high edge feature two-dimensional matrix into the shape feature three-dimensional matrix in the three dimensions of length, width and height, and output the first convolution three-dimensional matrix. The corresponding formula is:

[0093]

[0094] in, Represents the values ​​of the elements whose coordinates in length, width and height are i, j and k in the first convolution three-dimensional matrix, Represents the values ​​of the elements whose coordinates in the three dimensions of length, width and height in the three-dimensional matrix of shape features are i, j and k respectively. Represents the value of the element with coordinates j and k in the width and height dimensions in the two-dimensional matrix of the long edge feature, Represents the value of the element whose coordinates in the length and height dimensions are i and k in the two-dimensional matrix of the wide edge feature. Represents the value of the element with coordinates i and j in the length and width dimensions in the two-dimensional matrix of the high edge feature, and w0 represents the element in the fusion layer Corresponding nodes and shapes of elements in the 3D convolutional layer The weight values ​​between the corresponding nodes, w1 represents the elements in the fusion layer The corresponding nodes and elements in the long edge convolution layer The weight values ​​between the corresponding nodes, w2 represents the elements in the fusion layer The corresponding nodes and elements in the wide edge convolution layer The weight values ​​between the corresponding nodes, w3 represents the elements in the fusion layer The corresponding nodes and elements in the high edge convolution layer The corresponding weight value between nodes, b i,j,k Represents the elements in the fusion layer The corresponding node bias value, f() is the preset activation function, and m is the number of elements in the long dimension of the shape feature three-dimensional matrix.

[0095] In practice, the long edge feature two-dimensional matrix, the wide edge feature two-dimensional matrix and the high edge feature two-dimensional matrix can be directly fused into the shape feature three-dimensional matrix by simple linear superposition. In this embodiment, the fusion layer is directly connected with the shape three-dimensional convolution layer, the long edge convolution layer, the wide edge convolution layer and the high edge convolution layer by full connection, so that the fusion rule of the fusion layer can be trained (i.e., w0, w1, w2, w3 and b in the above formula i,j,k ), further improving the accuracy of feature extraction.

[0096] Furthermore, in a preferred embodiment, in the preset three-dimensional convolutional neural network, the output layer includes multiple output nodes, and the multiple output nodes are respectively connected to multiple nodes in the fully connected layer adjacent to the output layer, the number of output nodes is equal to the preset number of target entity types, each output node corresponds to a preset type, and the output node is used to output the probability value of the target entity type being the corresponding preset type.

[0097] This structure is highly understandable, allowing the system to classify in an efficient and accurate manner by analyzing the shape information of the target entity. Through the training and optimization of the 3D convolutional neural network, the automatic recognition capability of different entity types can be improved, thereby speeding up modeling and data entry, reducing the need for manual intervention, and improving overall work efficiency and accuracy.

[0098] Further, in a preferred embodiment, the building information includes material information, attribute information, cost information, space information and structure information, and the steps performed by the information supplementation module 150 are: based on the target entity type, supplementing the building information of the target entity in the BIM system, specifically including:

[0099] Retrieve entities of the same type as the target entity in the BIM system;

[0100] Obtain material information of entities of the same type as the target entity in the BIM system as the material information of the target entity;

[0101] Based on the material information of the target entity, the attribute information, cost information, space information and structure information of the target entity are supplemented.

[0102] The above process uses the known information of entities of the same type in the BIM system to supplement data according to the target entity type, thereby improving the accuracy of data filling, and using material information as a basis to supplement most of the building data. For example, after knowing the shape information and material information of the target entity in the building, it is possible to calculate the attribute information of the target entity (such as the type, strength, durability, environmental performance and other material properties of the materials used in the building components, such as thermal insulation performance, fire resistance performance, acoustic performance and other technical properties), cost information (such as the material budget and material expenditure of each component, system or the entire project, etc.), space information (functional zoning and use planning of the internal space of the building, actual use and occupancy of each space, etc.) and structural information (such as structural calculation and analysis results, including load distribution, stress strain, etc., or safety performance such as seismic performance and fire resistance performance of the structure). The above information can be inferred using any existing technology, so this article will not explain it in detail.

[0103] The present invention provides a building information management system based on BIM technology, which obtains shape modeling data of a target entity based on a BIM model through a data parsing module, then establishes a three-dimensional grid space and models in the three-dimensional grid space through a model building module to obtain a target entity model, and then uses a data encoding module to encode according to the covering relationship between the three-dimensional grid space and the target entity model to obtain a three-dimensional input matrix, and then uses a type recognition module to use the three-dimensional input matrix and a preset three-dimensional convolutional neural network to identify the target entity type, and finally uses an information supplementation module to supplement the building information of the target entity in the BIM system based on the target entity type. Compared with the prior art, the present invention is based on the most important and basic shape information in the BIM system, that is, shape modeling data, modeling in the three-dimensional grid space, and uses the covering relationship between the three-dimensional grid space and the target entity model to complete the encoding of the entity shape, so that the entity type can be identified through the three-dimensional convolutional neural network, and then other missing information of the target entity can be automatically supplemented based on the type of the target entity, so that the entire BIM system can temporarily start working based on the automatically filled building information, and then change it when the subsequent information is entered, solving the problem that the BIM technology in the prior art cannot be normally applied in the case of data missing.

[0104] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0105] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A building information management system based on BIM technology, characterized in that: include: A data analysis module is used to obtain shape modeling data of a target entity based on a BIM model; A model building module is used to build a three-dimensional grid space and to build a model in the three-dimensional grid space based on the shape modeling data to obtain a target entity model in the three-dimensional grid space; A data encoding module is used to encode according to the covering relationship between the three-dimensional grid space and the target entity model to obtain a three-dimensional input matrix, wherein each element in the three-dimensional input matrix corresponds to a grid in the three-dimensional grid space, and the value of each element in the three-dimensional input matrix is ​​used to represent a part of the target entity model in the corresponding grid; A category recognition module, used to input the three-dimensional input matrix into a preset three-dimensional convolutional neural network to obtain the target entity category output by the preset three-dimensional convolutional neural network; The information supplement module is used to supplement the building information of the target entity in the BIM system based on the target entity type.

2. The building information management system based on BIM technology according to claim 1 is characterized in that: Based on the BIM model, the shape modeling data of the target entity is obtained, including: Obtain input range selection data representing the target entity, and obtain BIM shape data of the target entity according to the BIM model based on the range selection data; Based on the BIM shape data, obtain the general model data of the target entity in STEP format; The entity number in the general model data, the position information and shape information corresponding to each entity number are obtained as the shape modeling data of the target entity.

3. The building information management system based on BIM technology according to claim 2 is characterized in that: A three-dimensional grid space is established, and modeling is performed in the three-dimensional grid space based on the shape modeling data to obtain a target entity model in the three-dimensional grid space, including: Establishing a three-dimensional grid space and obtaining dimension data of the three-dimensional grid space; Based on the size data of the three-dimensional grid space, the position information and the shape information in the shape modeling data, coordinate transformation is performed on the position information to obtain standard position data; Based on the size data of the three-dimensional grid space, the position information and the shape information in the shape modeling data, the shape information is scaled to obtain standard shape data; Based on the entity number, the standard position information and the standard shape information corresponding to each entity number, modeling is performed in the three-dimensional grid space to obtain a target entity model in the three-dimensional grid space.

4. The building information management system based on BIM technology according to claim 3 is characterized in that: According to the covering relationship between the three-dimensional grid space and the target entity model, the three-dimensional input matrix is ​​obtained, including: Obtain a target grid, where the target grid is a grid in a three-dimensional grid space; The spatial proportion of the target entity model in the target grid is calculated and normalized to obtain the element value of the three-dimensional input matrix corresponding to the target grid.

5. The building information management system based on BIM technology according to claim 1, characterized in that: The preset three-dimensional convolutional neural network includes an input layer, at least one three-dimensional convolutional layer, at least one pooling layer, at least one fully connected layer and an output layer, wherein the input layer is located at the first layer of the preset three-dimensional convolutional neural network, a three-dimensional convolutional layer is connected after the input layer, the output layer is located at the last layer of the preset three-dimensional convolutional neural network, a fully connected layer is connected before the output layer, and other three-dimensional convolutional layers, pooling layers and fully connected layers are arranged and combined and connected in sequence between the three-dimensional convolutional layer after the input layer and the fully connected layer before the output layer.

6. The building information management system based on BIM technology according to claim 5 is characterized in that: The three-dimensional convolution layer after the input layer includes a shape three-dimensional convolution layer, a long edge convolution layer, a wide edge convolution layer, a high edge convolution layer and a fusion layer, wherein the input ends of the shape three-dimensional convolution layer, the long edge convolution layer, the wide edge convolution layer and the high edge convolution layer are all connected to the input layer, and the output ends of the shape three-dimensional convolution layer, the long edge convolution layer, the wide edge convolution layer and the high edge convolution layer are all connected to the input end of the fusion layer, wherein the shape three-dimensional convolution layer is used to detect the shape features of the target entity model, the long edge convolution layer is used to detect the shape edge features of the target entity model in the long dimension, the wide edge convolution layer is used to detect the shape edge features of the target entity model in the wide dimension, and the high edge convolution layer is used to detect the shape edge features of the target entity model in the high dimension, and the fusion layer is used to superimpose and fuse the output results of the long edge convolution layer, the wide edge convolution layer and the high edge convolution layer into the output results of the shape three-dimensional convolution layer in the three dimensions of length, width and height, and output the first convolution three-dimensional matrix.

7. The building information management system based on BIM technology according to claim 6 is characterized in that: The size of the three-dimensional grid space in length, width and height is n; the size of the convolution kernel in the three dimensions of length, width and height of the shape three-dimensional convolution layer is a, where a and n are both positive integers and a is less than n. The convolution kernel of the shape three-dimensional convolution layer is used to move step by step in the length, width and height of the input three-dimensional matrix respectively, and the shape three-dimensional convolution layer is used to output a three-dimensional matrix of shape features; The convolution kernel of the long edge convolution layer has the sizes of n, a and a in length, width and height respectively. The convolution kernel of the long edge convolution layer is used to move step by step in the width and height dimensions of the input three-dimensional matrix respectively. The long edge convolution layer is used to output a two-dimensional matrix of long edge features. The convolution kernel of the wide edge convolution layer has the sizes of a, n and a in length, width and height respectively. The convolution kernel of the wide edge convolution layer is used to move step by step in the length and height of the input three-dimensional matrix respectively. The long edge convolution layer is used to output a two-dimensional matrix of wide edge features. The sizes of the convolution kernel of the high edge convolution layer in length, width and height are a, a and n respectively. The convolution kernel of the high edge convolution layer is used to move step by step in the length and width of the input three-dimensional matrix respectively. The high edge convolution layer is used to output a high edge feature two-dimensional matrix.

8. The building information management system based on BIM technology according to claim 7 is characterized in that: The fusion layer is used to superimpose and fuse the long edge feature two-dimensional matrix, the wide edge feature two-dimensional matrix and the high edge feature two-dimensional matrix into the shape feature three-dimensional matrix in the three dimensions of length, width and height, and output the first convolution three-dimensional matrix. The corresponding formula is: in, Represents the values ​​of the elements whose coordinates in length, width and height are i, j and k in the first convolution three-dimensional matrix, Represents the values ​​of the elements whose coordinates in the three dimensions of length, width and height in the three-dimensional matrix of shape features are i, j and k respectively. Represents the value of the element with coordinates j and k in the width and height dimensions in the two-dimensional matrix of the long edge feature, Represents the value of the element whose coordinates in the length and height dimensions are i and k in the two-dimensional matrix of the wide edge feature. Represents the value of the element with coordinates i and j in the length and width dimensions in the two-dimensional matrix of the high edge feature, and w0 represents the element in the fusion layer Corresponding nodes and shapes of elements in the 3D convolutional layer The weight values ​​between the corresponding nodes, w1 represents the elements in the fusion layer The corresponding nodes and elements in the long edge convolution layer The weight values ​​between the corresponding nodes, w2 represents the elements in the fusion layer The corresponding nodes and elements in the wide edge convolution layer The weight values ​​between the corresponding nodes, w3 represents the elements in the fusion layer The corresponding nodes and elements in the high edge convolution layer The corresponding weight value between nodes, b i,j,k Represents the elements in the fusion layer The corresponding node bias value, f() is the preset activation function, and m is the number of elements in the long dimension of the shape feature three-dimensional matrix.

9. The building information management system based on BIM technology according to claim 5, characterized in that: The output layer includes multiple output nodes, and the multiple output nodes are respectively connected to multiple nodes in the fully connected layer adjacent to the output layer. The number of output nodes is equal to the preset number of target entity types. Each output node corresponds to a preset type, and the output node is used to output the probability value of the target entity type being the corresponding preset type.

10. The building information management system based on BIM technology according to claim 1, characterized in that: Building information includes material information, attribute information, cost information, space information and structural information; Based on the target entity type, the building information of the target entity is supplemented in the BIM system, including: Retrieve entities of the same type as the target entity in the BIM system; Obtain material information of entities of the same type as the target entity in the BIM system as the material information of the target entity; Based on the material information of the target entity, the attribute information, cost information, space information and structure information of the target entity are supplemented.

Citation Information

Patent Citations

  • Three-dimensional reconstruction method based on convolution neural network

    CN109410321A

  • Building information model data storage and conversion method based on national standard

    CN116126809A

  • Water conservancy project information management system and method based on BIM

    CN116561879A

  • Method and device for constructing power transformation engineering knowledge graph based on graph convolutional network

    CN117669724A

  • Classifying elements in an infrastructure model using convolutional graph neural networks

    US20240127046A1

Cited By

  • Building performance prediction method and evaluation system based on hybrid neural network

    CN119940138A