Automatic identification method and system of complex structures in BIM based on artificial intelligence

By applying a complex structure automatic recognition method based on artificial intelligence in the BIM model, and using pre-trained models to automatically identify BIM model structures and elements, the problems of accuracy and inefficiency in the existing technology are solved, and more efficient BIM model management and application.

CN118779712BActive Publication Date: 2024-11-22壹仟零壹艺数智(北京)科技有限公司
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411259344.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-10
Publication Date
2024-11-22
Estimated Expiration
2044-09-10

AI Technical Summary

Technical Problem

Existing methods are difficult to effectively handle and manage complex BIM model structures, especially in terms of classification and feature extraction, resulting in poor identification accuracy and inefficiency.

Method used

Using an artificial intelligence-based method, we use a pre-trained complex BIM model to identify models and feature extraction models, and automatically identify and classify BIM model structures and target model elements to improve the recognition accuracy and efficiency.

Benefits of technology

It significantly improves the identification accuracy and processing efficiency of complex BIM model structures, reduces manual intervention, saves time and resources, and improves the quality and effectiveness of BIM model management and application.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118779712B_ABST
    Figure CN118779712B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and system for automatically identifying complex structures in BIM based on artificial intelligence, including: first responding to a user's query instruction for the current BIM model structure, and obtaining corresponding BIM model structure information. Then, calling a pre-trained complex BIM model recognition model, inputting the current BIM model structure into this recognition model, and obtaining a model classification result. If the classification result shows that the current BIM model structure is complex, then it is regarded as a pending complex BIM model structure. In response to a query instruction for a target model element in the pending complex BIM model structure, calling a pre-trained feature extraction model, and based on this, determining the classification result of the target model element corresponding to the pending complex BIM model structure. Such a design can effectively improve the recognition accuracy and processing efficiency of the complex BIM model structure, greatly reduce manual intervention, save a lot of time and resources, and at the same time improve the quality and effect of BIM model management and application.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based complex structure automatic recognition method and system in BIM. Background Art

[0002] BIM (Building Information Modeling) is an information processing method that covers the entire life cycle of buildings and facilities. Its core content is a multidimensional digital model that contains a large amount of information about the building's geometry, physics, functions, etc. However, due to the structural complexity of buildings, especially in large-scale construction projects, manual processing and management of BIM models is a very challenging task. Therefore, how to automatically identify and process complex BIM model structures is an urgent problem that needs to be solved. However, existing methods often cannot effectively process complex BIM model structures, especially in terms of classification and feature extraction. Summary of the invention

[0003] The purpose of the present invention is to provide a method and system for automatically identifying complex structures in BIM based on artificial intelligence.

[0004] In a first aspect, an embodiment of the present invention provides a method for automatically identifying complex structures in BIM based on artificial intelligence, comprising:

[0005] In response to a user's query instruction for a current BIM model structure, acquiring the current BIM model structure;

[0006] Calling a pre-trained complex BIM model recognition model, and inputting the current BIM model structure into the complex BIM model recognition model to obtain a model classification result of the current BIM model structure;

[0007] If the model classification result indicates that the current BIM model structure is a complex BIM model structure, the current BIM model structure is used as a to-be-determined complex BIM model structure;

[0008] In response to a query instruction for a target model element in the pending complex BIM model structure, calling a pre-trained feature extraction model;

[0009] Based on the pre-trained feature extraction model, a target model element classification result corresponding to the undetermined complex BIM model structure is determined.

[0010] In a second aspect, an embodiment of the present invention provides a server system, including a server, wherein the server is used to execute the method described in the first aspect.

[0011] Compared with the prior art, the beneficial effects provided by the present invention include: the method and system for automatically identifying complex structures in BIM based on artificial intelligence disclosed by the present invention responds to the user's query instructions for the current BIM model structure and obtains the corresponding BIM model structure information. Then, the pre-trained complex BIM model recognition model is called, and the current BIM model structure is input into this recognition model to obtain the model classification result. If the classification result shows that the current BIM model structure is complex, it is regarded as a pending complex BIM model structure. In response to the query instruction for the target model element in the pending complex BIM model structure, the pre-trained feature extraction model is called, and based on this, the classification result of the target model element corresponding to the pending complex BIM model structure is determined. Such a design can effectively improve the recognition accuracy and processing efficiency of the complex BIM model structure, greatly reduce manual intervention, save a lot of time and resources, and at the same time improve the quality and effect of BIM model management and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can also be obtained based on these drawings without creative work.

[0013] Figure 1 A schematic diagram of the steps of a method for automatically identifying complex structures in BIM based on artificial intelligence provided by an embodiment of the present invention;

[0014] Figure 2 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0016] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings.

[0017] In order to solve the technical problems in the aforementioned background technology, Figure 1A flow chart of a method for automatic identification of complex structures in BIM based on artificial intelligence provided in an embodiment of the present disclosure is provided below. The method for automatic identification of complex structures in BIM based on artificial intelligence is introduced in detail.

[0018] Step S201, in response to a user's query instruction for a current BIM model structure, obtaining the current BIM model structure;

[0019] Step S202, calling a pre-trained complex BIM model recognition model, and inputting the current BIM model structure into the complex BIM model recognition model to obtain a model classification result of the current BIM model structure;

[0020] Step 203: if the model classification result indicates that the current BIM model structure is a complex BIM model structure, the current BIM model structure is used as a to-be-determined complex BIM model structure;

[0021] Step 204, in response to a query instruction for a target model element in the undetermined complex BIM model structure, calling a pre-trained feature extraction model;

[0022] Step 205: Determine the target model element classification result corresponding to the undetermined complex BIM model structure based on the pre-trained feature extraction model.

[0023] In an embodiment of the present invention, for example, it is assumed that there is an architectural design company that is designing a large shopping mall. The project involves a large number of complex BIM model structures, such as escalators, elevators, ventilation systems, etc. on different floors. Designers need to query the information of these complex structures in order to better understand and modify them. In this case, this complex structure automatic recognition method in BIM based on artificial intelligence can be used to improve work efficiency. First, the designer obtains the current BIM model structure by inputting a query instruction, for example, he wants to view all elevator structures. The system will respond to the user's query instruction and obtain the current BIM model structure. Then, the system will call the pre-trained complex BIM model recognition model and input the current BIM model structure (i.e., the elevator structure) into this recognition model. Through this process, the system can obtain the model classification result, that is, determine whether these elevator structures are complex BIM model structures. If the model classification result shows that the current BIM model structure is complex, then the system will treat these elevator structures as pending complex BIM model structures. Then, the designer may want to further query specific target model elements in the pending complex BIM model structure, such as the elevator door. After receiving the query instruction, the system will call the previously trained feature extraction model. Based on this feature extraction model, the system can determine the classification results of the target model elements (i.e., elevator doors) corresponding to the complex BIM model structure to be determined (i.e., elevator structure). Designers can use this to understand the specific information of the elevator door, such as size, material, etc. In general, by using this method, designers can obtain and understand complex BIM model structure information more quickly, thereby improving work efficiency.

[0024] In another implementation of the embodiment of the present invention, illustratively, a user wants to obtain the BIM model structure information of a certain building project. The system obtains the BIM model data of the building project from the database, including the geometric information and attributes of elements such as walls, stairs, doors and windows. The system calls a pre-trained complex BIM model recognition model, which is trained through deep learning and can recognize various types of complex structures. The system passes the acquired BIM model structure as input to the complex BIM model recognition model. The complex BIM model recognition model analyzes the input BIM model structure and generates a model classification result. For example, the result may indicate that the building project has a complex three-dimensional surface structure. According to the model classification result, the system marks the building project as a pending complex BIM model structure for further analysis. The user issues a query instruction for a specific element (e.g., a column) in the pending complex BIM model structure, hoping to obtain detailed information about the element. The system calls a pre-trained feature extraction model, which can extract the features of the target model element in the pending complex BIM model structure. Based on the feature extraction model, the system can determine the classification result of the column in the pending complex BIM model structure. For example, the result may indicate that the column type is cylindrical and the height is 3 meters. Through the above practical examples, the previously described method of automatic identification of complex structures in BIM based on artificial intelligence can be understood in more detail. This method can automatically identify complex BIM model structures according to the user's query instructions by calling the pre-trained model and determine the classification results of the target model elements. In this way, users can obtain BIM model structure information more accurately and efficiently, providing support for the design and construction of construction projects.

[0025] In the embodiment of the present invention, the aforementioned step S205 may be implemented in the following manner.

[0026] (1) Perform spatial division on the complex BIM model structure to obtain at least two local BIM model structures;

[0027] (2) performing an image feature extraction operation on each local BIM model structure based on a pre-trained feature extraction model to obtain a local image feature vector; wherein the feature extraction model includes a basic feature extraction unit and an advanced feature extraction unit; the basic feature extraction unit is used to perform a basic image feature extraction operation on the local BIM model structure; the advanced feature extraction unit is used to perform an advanced image feature extraction operation on the local BIM model structure based on the basic image features obtained during the basic image feature extraction operation, and the accuracy of the advanced image feature extraction operation is greater than the accuracy of the basic image feature extraction operation;

[0028] (3) Obtaining the image feature vector of each reference BIM model element in the preset BIM model element library;

[0029] (4) classifying the model elements of each of the local BIM model structures according to the local image feature vector and the image feature vector of each reference BIM model element to obtain a local model element classification result;

[0030] (5) Determining a target model element classification result corresponding to the undetermined complex BIM model structure according to the local model element classification results of the at least two local BIM model structures.

[0031] In an embodiment of the present invention, for example, it is assumed that the designer now wants to gain a deeper understanding of the target model element of the door of the elevator structure. The system will first perform spatial division on the complex BIM model structure to be determined (i.e., the elevator structure) to obtain at least two local BIM model structures. For example, it can treat each elevator door as a separate local BIM model structure. Then, the system will perform an image feature extraction operation on each local BIM model structure (i.e., each elevator door) based on a pre-trained feature extraction model to obtain a local image feature vector. The feature extraction model includes a basic feature extraction unit and an advanced feature extraction unit. The basic feature extraction unit is used to perform a basic image feature extraction operation on the local BIM model structure, such as obtaining information such as the size and color of the elevator door. Then, the advanced feature extraction unit will perform an advanced image feature extraction operation on the local BIM model structure based on the basic image features, such as obtaining more complex information such as the texture and reflectivity of the elevator door. The accuracy of this advanced image feature extraction operation is greater than the accuracy of the basic image feature extraction operation. After completing the above steps, the system will obtain the image feature vector of each reference BIM model element in the preset BIM model element library. This model element library may include various common elevator door designs. Then, the system will classify the model elements of each local BIM model structure (i.e., each elevator door) according to the local image feature vector and the image feature vector of each reference BIM model element, and obtain the local model element classification results. For example, it can determine which type of elevator door in the model element library the design of a certain elevator door is most similar to. Finally, the system will determine the classification results of the target model elements (i.e., all elevator doors) corresponding to the undetermined complex BIM model structure (i.e., the entire elevator structure) based on the local model element classification results of at least two local BIM model structures (i.e., at least two elevator doors). Through this method, designers can not only understand the detailed information of each elevator door, but also understand the overall design style and characteristics of the doors of the entire elevator structure, so as to better carry out design work.

[0032] In the embodiment of the present invention, the aforementioned step of spatially dividing the complex BIM model structure to be determined to obtain at least two local BIM model structures can be implemented through the following examples.

[0033] (1) Using cubes with fixed intervals, the complex BIM model structure to be determined is spatially cut to obtain at least two local BIM model structures, where the at least two local BIM model structures have the same spatial scale.

[0034] In an embodiment of the present invention, for example, if the designer needs to study the door of the elevator structure in depth, the system will use cubes at fixed intervals to perform spatial cutting processing on the complex BIM model structure (i.e., the elevator structure). Specifically, assuming that the elevator door is a rectangular three-dimensional model, the system can use a series of cubes of the same size to "cover" or "cut" the elevator door model. For example, if the size of the elevator door is 2 meters wide and 3 meters high, the system can use 1 cubic meter cubes to cut the elevator door model. Each cube will contain a part of the elevator door model, and these cubes are local BIM model structures. Since all cubes are the same size, these local BIM model structures have the same spatial scale. In this way, the system can obtain and analyze various features of the elevator door in more detail and precision. This is crucial for understanding complex BIM model structures and accurately classifying model elements.

[0035] In another implementation of the embodiment of the present invention, illustratively, it is assumed that the complex BIM model structure to be determined has been successfully spatially divided, and the classification results of at least two local BIM model structures have been obtained. In order to further refine the analysis of the complex BIM model structure to be determined, the system needs to use a cube with a fixed interval to perform spatial cutting processing on the complex BIM model structure to be determined. This cube can be regarded as a virtual three-dimensional grid, and each cube unit has the same size. By comparing the complex BIM model structure to be determined with the boundary of the cube, the system can determine which cube units contain model elements, thereby obtaining at least two local BIM model structures. These local BIM model structures have the same spatial scale, that is, the size of the cube units is the same. For each local BIM model structure, the system can use the previously described method, that is, using feature extraction and model element classification technology, to perform image feature extraction operations and model element classification on each local structure. In this way, the classification results of the local model elements of each local BIM model structure can be obtained. According to the classification results of the local model elements of at least two local BIM model structures, the system can determine the classification results of the target model elements corresponding to the complex BIM model structure to be determined. For example, the final target model element classification result can be determined by aggregating multiple local results and performing voting, weighted averaging, etc.

[0036] In the embodiments of the present invention, the following implementation modes are also provided.

[0037] (1) Obtaining preset normative rules;

[0038] (2) performing data normalization on each of the local BIM model structures using the preset normalization rules, and obtaining at least two local BIM model structures after data normalization;

[0039] The aforementioned step of performing an image feature extraction operation on each local BIM model structure to obtain a local image feature vector can be implemented through the following example.

[0040] (1) Performing an image feature extraction operation on each local BIM model structure after data normalization to obtain the local image feature vector.

[0041] In the embodiment of the present invention, for example, preset specification rules are first obtained. These specification rules may include various data standardization or normalization methods, such as Max-Min normalization, Z-Score normalization, etc. These rules can ensure that different local BIM model structures (i.e., the cubes obtained by cutting in the previous step) can be compared under the same standard during feature extraction and classification, avoiding the influence caused by differences in scale, unit, etc.

[0042] Next, the system will use these preset normative rules to normalize the data of each local BIM model structure to obtain the local BIM model structure after data normalization. For example, it can normalize the three-dimensional coordinates, color values ​​and other information of the elevator door part in each cube.

[0043] Then, the system performs image feature extraction on each local BIM model structure after data normalization to obtain a local image feature vector. This may include using a basic feature extraction unit to obtain basic image features, and using an advanced feature extraction unit to obtain more complex image features.

[0044] Through this method, the system can ensure that when performing complex BIM model structure identification and classification, the processing and comparison of each local BIM model structure is fair and accurate, thereby improving the reliability of the entire identification and classification process.

[0045] In another implementation of the embodiment of the present invention, illustratively, it is assumed that the complex BIM model structure to be determined has been successfully spatially divided, and at least two local BIM model structures with the same spatial scale have been obtained. The system obtains specification rules related to architectural design, construction or other fields through a preset specification rule library. These specification rules describe information such as attributes, relationships and constraints of model elements. In order to make each local BIM model structure conform to the preset specification rules, the system uses the preset specification rules to perform data normalization processing on each local BIM model structure. By applying the specification rules, the attribute values, relationships and geometric shapes of the local model can be adjusted, and the corresponding local BIM model structure after data normalization is generated. For each local BIM model structure after data normalization, the system performs an image feature extraction operation to obtain a local image feature vector. These feature vectors represent the visual features of the normalized local model. By performing an image feature extraction operation on each local BIM model structure after data normalization, the local image feature vector can be obtained. These feature vectors can describe the shape, texture or other visual attributes of the local model.

[0046] In the embodiment of the present invention, the aforementioned step of performing an image feature extraction operation on each local BIM model structure based on a pre-trained feature extraction model to obtain a local image feature vector can be implemented through the following example.

[0047] (1) loading each of the local BIM model structures into the basic feature extraction unit, and extracting basic features of the local BIM model structure through the basic feature extraction unit to obtain an image feature vector with basic accuracy;

[0048] (2) The image feature vector with basic accuracy is loaded into the advanced feature extraction unit, and the advanced feature extraction unit is used to perform advanced feature extraction on the local BIM model structure to obtain an image feature vector with advanced accuracy; the basic accuracy is less than the advanced accuracy.

[0049] In an embodiment of the present invention, exemplarily, the system first loads each local BIM model structure (i.e., each part of the elevator door obtained by spatial cutting) into a basic feature extraction unit. In this unit, the system extracts basic features for each local BIM model structure to obtain an image feature vector with basic accuracy. These basic features may include the most intuitive image attributes such as color, shape, and size. Then, the system loads these image feature vectors with basic accuracy into an advanced feature extraction unit. In this unit, the system extracts advanced features for each local BIM model structure to obtain an image feature vector with advanced accuracy. These advanced features may include more complex image attributes such as texture and reflectivity. Since the algorithms and techniques used in the advanced feature extraction unit are more advanced than those in the basic feature extraction unit, the accuracy of the features extracted by the advanced feature extraction unit is greater than that of the basic feature extraction unit. In this way, the system can understand and describe each local BIM model structure at different levels, thereby more accurately identifying and classifying complex BIM model structures.

[0050] In another implementation of the embodiment of the present invention, illustratively, it is assumed that a feature extraction model has been obtained through prior training. This model can extract image feature vectors with different accuracy levels based on the input local BIM model structure. For each local BIM model structure, it is first loaded into the basic feature extraction unit. By using the basic feature extraction unit, the system performs basic feature extraction on the local BIM model structure, thereby obtaining image feature vectors with basic accuracy. These image feature vectors reflect the accuracy of the basic feature extraction model in extracting low-level visual features of the local model. The image feature vectors with basic accuracy are loaded into the advanced feature extraction unit. By using the advanced feature extraction unit, the system performs advanced feature extraction on the local BIM model structure to obtain image feature vectors with advanced accuracy. These image feature vectors reflect the accuracy of the advanced feature extraction model in extracting high-level visual features of the local model. By comparing the basic accuracy and the advanced accuracy, it can be found that the accuracy of the image feature vector obtained by the basic feature extraction is lower than the image feature vector obtained by the advanced feature extraction.

[0051] In an embodiment of the present invention, the aforementioned step of classifying the model elements of each of the local BIM model structures according to the local image feature vector and the image feature vector of each reference BIM model element to obtain the local model element classification result can be implemented through the following example.

[0052] (1) determining a vector distance between the local image feature vector and the image feature vector of each reference BIM model element;

[0053] (2) When the vector distance between the local image feature vector and the image feature vector of any reference BIM model element is less than a preset vector distance threshold, the local model element classification result of the local BIM model structure is determined to be an expected BIM element recognition result; the expected BIM element recognition result is used to indicate that there is a model element in the local BIM model corresponding to the local BIM model structure that has the same type as the reference BIM model element in the preset BIM model element library.

[0054] In an embodiment of the present invention, exemplarily, continuing the above-mentioned shopping mall design scenario, the designer is conducting an in-depth study on the door of the elevator structure. In the above-mentioned method, the system first determines the vector distance between the local image feature vector (i.e., the image feature of each elevator door) and the image feature vector of each reference BIM model element (which may be various elevator door designs in the preset BIM model element library). This vector distance can be used to measure the similarity between the local BIM model structure (i.e., the elevator door) and the reference BIM model element. If the vector distance between the local image feature vector and the image feature vector of any reference BIM model element is less than the preset vector distance threshold, the system will determine that the local model element classification result of the local BIM model structure is the expected BIM element recognition result. For example, if the image feature vector of a certain elevator door is very close to the image feature vector of a reference elevator door in the preset BIM model element library, the system will determine that the design of this elevator door is similar or consistent with the design of the reference elevator door. This expected BIM element recognition result can be used to indicate that there is a model element in the local BIM model corresponding to the local BIM model structure (i.e., the elevator door) that has the same type as the reference BIM model element in the preset BIM model element library. Designers can use this to understand the design style and characteristics of elevator doors, as well as their similarities and differences with other common designs, so as to make better design decisions.

[0055] In another implementation of an embodiment of the present invention, illustratively, it is assumed that a reference BIM model element library has been established, which contains various types of standard BIM model elements and their corresponding image feature vectors. For each local BIM model structure, the system classifies the model elements according to the local image feature vector and the image feature vector of the reference BIM model element. This process is used to identify the specific model element type existing in the local model. The system calculates the vector distance between the local image feature vector and the image feature vector of each reference BIM model element. This vector distance can measure the similarity or difference between the two vectors. The system sets a preset vector distance threshold as a criterion for judging whether the local image feature vector is similar to the image feature vector of the reference BIM model element. When the vector distance between the local image feature vector and the image feature vector of any reference BIM model element is less than the preset vector distance threshold, the system will determine that the local model element classification result of the local BIM model structure is the expected BIM element recognition result. This means that there are model elements in the local model that are consistent with the reference BIM model elements in the reference BIM model library. The expected BIM element identification result obtained may be used to indicate that a model element having the same type as a reference BIM model element in a preset BIM model element library exists in the local model corresponding to the local BIM model structure.

[0056] In an embodiment of the present invention, the aforementioned step of determining the classification result of the target model element corresponding to the undetermined complex BIM model structure according to the classification results of the local model elements of the at least two local BIM model structures can be implemented through the following example.

[0057] (1) When the local model element classification results of the at least two local BIM model structures are both corresponding expected BIM element identification results, determining that the target model element classification result corresponding to the undetermined complex BIM model structure is a model element classification result determined by the expected BIM element identification results corresponding to the local model element classification results of the at least two local BIM model structures.

[0058] In an embodiment of the present invention, exemplarily, if the local model element classification results of at least two local BIM model structures (i.e., the parts of the elevator door obtained by spatial cutting) are both corresponding expected BIM element recognition results, the system will determine the classification results of the target model elements (i.e., all elevator doors) corresponding to the pending complex BIM model structure (i.e., the entire elevator structure).

[0059] For example, if the system classifies each elevator door as a specific type of elevator door in the preset BIM model element library when processing it, the system will determine that the design of the doors of the entire elevator structure is similar or consistent with this specific type of elevator door.

[0060] In this way, the designer can understand the overall design style and characteristics of the door of the entire elevator structure, as well as its similarities and differences with other common designs. This is of great significance for designers to make higher-level design decisions, such as whether the design of the elevator door needs to be modified and what design style should be adopted after the modification.

[0061] In another implementation of the embodiment of the present invention, illustratively, for each local BIM model structure, the system first uses the aforementioned method to classify and identify the local model elements thereof. The system will compare each local model element with the expected BIM element, and obtain the classification result of the local model element, that is, the expected BIM element identification result. Taking into account the complexity of the overall building structure, it may be necessary to conduct a comprehensive analysis of the classification results of multiple local BIM model structures. If there are several floors of the same type in this large-scale building project, for example, two floors are correctly classified as "office areas", then it can be considered that the classification results of these local model elements are consistent. Based on the consistency of the classification results of multiple local BIM model structures, the system can determine the classification results of the target model elements corresponding to the pending complex BIM model structure. In the above example, it can be determined that the classification results of the target model elements of these floors of the same type in the entire building project are "office areas".

[0062] In an embodiment of the present invention, the preset BIM model element library includes a reference BIM model structure of each reference BIM model element; the aforementioned step of obtaining an image feature vector of each reference BIM model element in the preset BIM model element library can be implemented through the following example.

[0063] (1) Performing an image feature extraction operation on the reference BIM model structure of each reference BIM model element through the pre-trained feature extraction model to obtain an image feature vector of each reference BIM model element.

[0064] In an embodiment of the present invention, exemplarily, the system first obtains the image feature vector of each reference BIM model element in the preset BIM model element library. The preset BIM model element library is a database containing various common or standard BIM model elements, and each reference BIM model element has a corresponding reference BIM model structure, for example, elevator doors of different types, styles, and sizes. In order to obtain the image feature vectors of these reference BIM model elements, the system performs an image feature extraction operation on the reference BIM model structure of each reference BIM model element based on a pre-trained feature extraction model. This operation is similar to the operation previously performed on the pending complex BIM model structure (i.e., the elevator structure), and both include two steps: basic feature extraction and advanced feature extraction. In this way, the system can obtain the image feature vector of each reference BIM model element, and these feature vectors will be used in the subsequent model element classification process to help the system determine which reference BIM model element the pending complex BIM model structure (i.e., the elevator structure) is most similar to, thereby providing designers with more accurate information.

[0065] In another implementation of the embodiment of the present invention, illustratively, in order to support the automated recognition process, a preset BIM model element library has been established. The library contains various types of reference BIM model elements and their corresponding reference BIM model structures. For example, there may be an "office area" element in the reference BIM model element library, which includes the reference BIM model structure of the element in different floors and buildings. The system performs an image feature extraction operation on the reference BIM model structure of each reference BIM model element by using a pre-trained feature extraction model. This operation converts the reference BIM model structure into an image feature vector, which contains key description information of the structure, such as geometric shape, texture features, etc. The system obtains the image feature vector of each reference BIM model element in the preset BIM model element library. For example, for the "office area" element, the image feature vector of the element in each floor and building can be obtained. When automatically identifying a complex structure, when a pending model element is processed, the system will perform a similar image feature extraction operation on it and obtain the image feature vector of the element. The system finds the best matching reference BIM model element type by comparing the image feature vector of the pending model element with the image feature vector in the preset BIM model element library. For example, if the image feature vector of the pending model element is highly similar to the image feature vector of the "office area" element, the pending element is classified as "office area". This design enables the system to compare the pending model element with the reference elements in the preset BIM model element library and find the best matching model element type. This method can improve the accuracy and reliability of the automated recognition system and can more accurately classify and identify model elements when processing complex BIM model structures.

[0066] In an embodiment of the present invention, the feature extraction model is obtained in the following manner.

[0067] (1) loading a first sample BIM model structure in a basic BIM model structure set into the basic feature extraction unit, and training the basic feature extraction unit by a comparative learning method to obtain a trained basic feature extraction unit;

[0068] (2) loading a second sample BIM model structure in the independent BIM model structure set into the trained basic feature extraction unit, and performing basic feature extraction on the second sample BIM model structure through the trained basic feature extraction unit to obtain a sample image feature vector with a third accuracy;

[0069] (3) loading the sample image feature vector with the third accuracy into the advanced feature extraction unit, and performing advanced feature extraction on the second sample BIM model structure by the advanced feature extraction unit to obtain a sample image feature vector with a fourth accuracy; the third accuracy is less than the fourth accuracy;

[0070] (4) classifying the second sample BIM model structure by a preset category classification unit based on the sample image feature vector with the fourth accuracy to obtain a sample BIM element recognition result;

[0071] (5) loading the sample BIM element recognition result and the target category information of the second sample BIM model structure into a preset cost unit, and outputting the cost parameter through the preset cost unit;

[0072] (6) Adjusting the network architecture in the advanced feature extraction unit based on the cost parameter to obtain a trained feature extraction model.

[0073] In an embodiment of the present invention, exemplarily, first, the system loads the first sample BIM model structure (for example, a standard elevator door model) in the basic BIM model structure set into a basic feature extraction unit, and trains the unit by comparative learning to obtain a trained basic feature extraction unit.

[0074] Next, the system will load the second sample BIM model structure in the independent BIM model structure set (for example, an elevator door model with a specific design style) into the trained basic feature extraction unit, and use the unit to extract basic features of the second sample BIM model structure to obtain a sample image feature vector with a third accuracy.

[0075] Then, these sample image feature vectors with the third accuracy are loaded into an advanced feature extraction unit, which performs advanced feature extraction on the second sample BIM model structure to obtain sample image feature vectors with a fourth accuracy, wherein the fourth accuracy is greater than the third accuracy.

[0076] Next, the preset category classification unit will classify the model elements of the second sample BIM model structure based on the sample image feature vector with the fourth accuracy to obtain a sample BIM element recognition result.

[0077] The system loads the sample BIM element recognition results and the target category information of the second sample BIM model structure into the preset cost unit, and outputs the cost parameter through the unit. Then, the system adjusts the network architecture in the advanced feature extraction unit based on this cost parameter, and finally obtains the trained feature extraction model.

[0078] This trained feature extraction model can more accurately extract the features of complex BIM model structures such as elevator doors, and can effectively classify different types and styles of elevator door designs, thereby helping designers make more accurate design decisions.

[0079] In another implementation of the present invention, illustratively, a first sample BIM model structure, such as an office floor, is selected from a basic BIM model structure set of a commercial complex project, and the sample is loaded into a basic feature extraction unit.

[0080] By contrastive learning, the feature extraction unit trains the basic sample to extract important features. These features may include the geometric shape of the floor, the texture features of the wall, etc. After training, a trained basic feature extraction unit is obtained.

[0081] Next, a second sample BIM model structure is selected from the independent BIM model structure set, such as the internal structure of a store, and the sample is loaded into the trained basic feature extraction unit.

[0082] Using the trained basic feature extraction unit, basic feature extraction is performed on the internal structure of the second sample store. This will generate a sample image feature vector with higher accuracy, which contains important feature information of the internal structure of the store.

[0083] The sample image feature vector with high accuracy is loaded into the advanced feature extraction unit.

[0084] By using the advanced feature extraction unit, further feature extraction is performed on the internal structure of the store to obtain sample image feature vectors with higher accuracy. These vectors can more accurately describe the characteristics of the internal structure of the store.

[0085] Using the sample image feature vectors with higher accuracy, the model elements of the store interior structure can be classified using a preset category division unit.

[0086] The sample BIM element recognition results and the actual target category information of the store’s internal structure are loaded into the preset cost unit.

[0087] By presetting the cost unit output cost parameters, the accuracy of the model element classification can be evaluated and the cost of applying the classification scheme to the internal structure of the store can be obtained.

[0088] Based on the cost parameter, the network architecture in the advanced feature extraction unit is adjusted to optimize the performance and accuracy of the model.

[0089] After multiple adjustments and trainings, we finally get a trained feature extraction model that can more accurately extract and describe the features of the store’s internal structure.

[0090] In an embodiment of the present invention, the basic feature extraction unit includes a geometric feature extraction unit and an attribute feature extraction unit; the aforementioned step of loading the first sample BIM model structure in the basic BIM model structure set into the basic feature extraction unit, training the basic feature extraction unit by comparative learning, and obtaining the trained basic feature extraction unit can be implemented through the following example.

[0091] (1) Loading a first sample BIM model structure in a basic BIM model structure set into the basic feature extraction unit;

[0092] (2) extracting geometric features from the first sample BIM model structure by the geometric feature extraction unit to obtain a basic feature vector;

[0093] (3) extracting attribute features from the basic feature vector by the attribute feature extraction unit to obtain an image feature vector having a preset number of attributes;

[0094] (4) loading the image feature vector having the preset number of attributes into a first cost unit, and determining a first cost parameter corresponding to the image feature vector having the preset number of attributes through a first cost function in the first cost unit;

[0095] (5) Adjusting the model architecture in the geometric feature extraction unit and the attribute feature extraction unit based on the first cost parameter to obtain the trained basic feature extraction unit.

[0096] In the embodiment of the present invention, the system loads the first sample BIM model structure (e.g., standard elevator door model) in the basic BIM model structure set into the basic feature extraction unit. Then, the geometric feature extraction unit extracts geometric features from the first sample BIM model structure to obtain a basic feature vector. These geometric features may include information such as the shape and size of the elevator door.

[0097] Next, the system will extract attribute features from the basic feature vector through the attribute feature extraction unit to obtain an image feature vector with a preset number of attributes. These attribute features may include information such as the color and material of the elevator door.

[0098] Then, the system loads the image feature vector with the preset number of attributes into the first cost unit, and determines the first cost parameter corresponding to the image feature vector with the preset number of attributes through the first cost function in the unit. This cost parameter can be used to measure the performance of the current model.

[0099] Finally, the system adjusts the model architecture in the geometric feature extraction unit and the attribute feature extraction unit based on the first cost parameter to obtain a trained basic feature extraction unit. This trained unit can more accurately extract the geometric features and attribute features of complex BIM model structures such as elevator doors, laying a solid foundation for subsequent feature extraction and model element classification processes.

[0100] In another implementation of the embodiment of the present invention, for example, it is assumed that a BIM model of a large residential project is being processed, and the model contains multi-story buildings, various types of rooms, and various components. It is hoped that a feature extraction model can be trained to automatically identify and classify the elements in these models. Geometric feature extraction unit: This unit is responsible for extracting geometric features from the BIM model, such as the shape of the floor, the length of the wall, the height of the column, etc.

[0101] Attribute feature extraction unit: This unit is responsible for extracting attribute features from the BIM model, such as the purpose of the room, the type of material, the specifications of the equipment, etc. First, select the first sample BIM model structure from the basic BIM model structure set of the residential project, such as a floor of a public area. Then load the sample into the basic feature extraction unit.

[0102] By contrastive learning, the basic feature extraction unit trains the sample to extract important geometric and attribute features. After training, a trained basic feature extraction unit is obtained. Another sample BIM model structure is loaded into the basic feature extraction unit, such as the internal structure of a bedroom.

[0103] The geometric feature extraction unit extracts geometric features from the sample and obtains basic feature vectors therefrom, such as the shape of the room, the length of the wall, etc.

[0104] The attribute feature extraction unit extracts attribute features from the basic feature vector to generate an image feature vector with a preset number of attributes, such as the purpose of the room, the type of material, and the like.

[0105] An image feature vector having a preset number of attributes is loaded into the first cost unit.

[0106] The first cost function in the first cost unit is used to determine the first cost parameter corresponding to the image feature vector, which can measure the importance and cost of the feature vector in the automatic recognition task.

[0107] Based on the first cost parameter, the model architecture in the geometric feature extraction unit and the attribute feature extraction unit is adjusted.

[0108] By optimizing the structure and parameters of these models, a trained basic feature extraction unit is obtained, which can better extract and describe the geometric and attribute features of the BIM model structure.

[0109] In an embodiment of the present invention, the advanced feature extraction unit includes a model structure information extraction unit and a weight allocation unit; the aforementioned step of loading the sample image feature vector with the third accuracy into the advanced feature extraction unit, performing advanced feature extraction on the second sample BIM model structure through the advanced feature extraction unit, and obtaining the sample image feature vector with the fourth accuracy can be implemented through the following example.

[0110] (1) loading the sample image feature vector having the third accuracy into the advanced feature extraction unit;

[0111] (2) extracting key model features of the sample image feature vector under different model components through the model structure information extraction unit;

[0112] (2) The weight allocation unit performs weight calculations on the key model features under the different model components in the model component sequence in sequence to obtain the sample image feature vector with the fourth accuracy.

[0113] In the embodiment of the present invention, illustratively, first, through the model structure information extraction unit, the system extracts key model features of the sample image feature vector under different model components. For example, if an elevator door is composed of a door frame, a door panel, and a door handle, the system will extract the key model features of each of these parts.

[0114] Then, through the weight allocation unit, the system will calculate the weights of the key model features under these different model components in the model component sequence. This process may consider the importance of each component in the entire elevator door design. For example, the design of the door panel may be more critical than the design of the door handle, so a larger weight may be assigned to the features of the door panel.

[0115] Finally, after weight calculation, the system obtains a sample image feature vector with fourth-order accuracy. This feature vector not only contains the basic features and attribute features of the elevator door, but also contains the key model features and corresponding weights of each component, so it can more accurately describe the design of the elevator door.

[0116] In another implementation of the embodiment of the present invention, exemplarily, a model structure information extraction unit: This unit is responsible for extracting key model structure information from the BIM model, such as the layout of floors, the connection method of walls, the location of columns, etc. A weight allocation unit: This unit is responsible for calculating the weights of model components to determine their importance in feature extraction.

[0117] First, a sample image feature vector with a third accuracy is loaded into an advanced feature extraction unit as a reference feature vector.

[0118] The advanced feature extraction unit performs advanced feature extraction on the second sample BIM model structure to obtain a sample image feature vector with a fourth accuracy.

[0119] The steps of feature extraction include:

[0120] The sample image feature vector having the third accuracy is loaded into the advanced feature extraction unit.

[0121] The key model features of the sample image feature vector under different model components are extracted through the model structure information extraction unit. For example, the connection mode between floors and the opening position on the wall are identified.

[0122] Using the weight allocation unit, weight calculations are performed on the model component sequence in turn to determine the importance of key model features under different model components.

[0123] A sample image feature vector with fourth accuracy is obtained, which contains more accurate and detailed structural information after advanced feature extraction.

[0124] In an embodiment of the present invention, the advanced feature extraction unit includes a cost evaluation unit, and the cost evaluation unit includes a second cost function; the embodiment of the present invention also provides the following implementation manner.

[0125] (1) loading the sample image feature vector with the fourth accuracy and the feature annotation information of the second sample BIM model structure into the cost evaluation unit;

[0126] (2) determining a second cost parameter corresponding to the sample image feature vector having a fourth accuracy by using a second cost function of the cost evaluation unit;

[0127] (3) Adjusting the model architecture in the model structure information extraction unit and the weight allocation unit based on the second cost parameter to obtain a trained advanced feature extraction unit.

[0128] In an embodiment of the present invention, exemplarily, first, the system loads the feature annotation information of the sample image feature vectors with the fourth accuracy (these feature vectors are obtained by the model structure information extraction unit and the weight allocation unit) and the second sample BIM model structure (for example, an elevator door model with a specific design style) into the cost evaluation unit. Then, through the second cost function of the cost evaluation unit, the system determines the second cost parameter corresponding to the sample image feature vector with the fourth accuracy. This cost parameter can be used to measure the performance of the current model. Finally, the system adjusts the model architecture in the model structure information extraction unit and the weight allocation unit based on the second cost parameter to obtain a trained advanced feature extraction unit.

[0129] In an embodiment of the present invention, the complex BIM model recognition model is obtained in the following manner.

[0130] (1) obtaining a sample BIM model file, wherein the sample BIM model file includes at least one complex BIM model structure and at least one pseudo-complex BIM model structure, wherein the pseudo-complex BIM model structure is generated based on the complex BIM model structure;

[0131] (2) performing feature recognition on the complex BIM model structure and the pseudo-complex BIM model structure respectively through the complex BIM model recognition model to obtain a first classification feature of the complex BIM model structure and a second classification feature of the pseudo-complex BIM model structure, wherein the complex BIM model recognition model is used to infer whether the input BIM model structure is a complex BIM model structure;

[0132] (3) determining a type error according to a first deviation between a first classification feature of the complex BIM model structure and a first target value, and a second deviation between a second classification feature of the pseudo-complex BIM model structure and a second target value;

[0133] (4) The complex BIM model recognition model is trained by the type error and the homogeneity restriction, wherein the homogeneity restriction is used to constrain the homogeneity of a first feature recognition result and a second feature recognition result, wherein the first feature recognition result is obtained by the complex BIM model recognition model performing feature recognition on the complex BIM model structure, and the second feature recognition result is obtained by the complex BIM model recognition model performing feature recognition on the pseudo-complex BIM model structure.

[0134] In the embodiment of the present invention, for example, it is assumed that an architectural design company needs to design a high-rise office building with a complex structure. In order to quickly and accurately identify the various parts of the complex structure, an automatic identification method of complex structures in BIM based on artificial intelligence is used.

[0135] After collecting a large number of sample BIM model files, these samples include a complex BIM model structure and a pseudo-complex BIM model structure. The pseudo-complex BIM model structure is generated by modifying the complex BIM model structure to simulate different types of structures that may appear in reality.

[0136] Using these sample BIM models, the researchers developed a complex BIM model recognition model. The model identified the features of complex BIM model structures and pseudo-complex BIM model structures, and obtained the first classification features of complex BIM model structures and the second classification features of pseudo-complex BIM model structures.

[0137] When the designer inputs the current BIM model into the system, the complex BIM model recognition model will make an inference based on the first feature recognition result and the second feature recognition result to determine whether the input BIM model structure is a complex structure.

[0138] During the training process, the type error is determined based on the deviation between the first classification feature of the complex BIM model structure and the first target value, and the deviation between the second classification feature of the pseudo-complex BIM model structure and the second target value.

[0139] The complex BIM model recognition model is trained by combining type error with homogeneity restriction. Homogeneity restriction is used to constrain the similarity between the first feature recognition result and the second feature recognition result, ensuring that the model can more accurately distinguish between complex structures and pseudo-complex structures.

[0140] This design can help design teams improve efficiency and accuracy when dealing with building projects with complex structures. For example, in the design of a high-rise office building, this method can help quickly identify complex cantilever structures, staggered floors, or complex facade elements so that corresponding design and construction plans can be made.

[0141] In an embodiment of the present invention, the complex BIM model recognition model includes a plurality of stacked feature recognition units; the aforementioned step of training the complex BIM model recognition model through the type error and homogeneity restriction can be implemented through the following example.

[0142] (1) determining a first homogeneity deviation according to the homogeneity of the first feature recognition result and the second feature recognition result of each feature recognition unit in all or part of the feature recognition units;

[0143] (2) Training the complex BIM model recognition model by using the type error and the first homogeneity deviation.

[0144] In an embodiment of the present invention, illustratively, it is assumed that a design company is responsible for a construction project of a large gymnasium. The gymnasium has a complex roof structure and support system, and the details of each part need to be accurately identified and analyzed. An automatic identification method for complex structures in BIM based on artificial intelligence is adopted, and a complex BIM model recognition model is used to process this project. The complex BIM model recognition model includes multiple stacked feature recognition units. These feature recognition units can be independent units specifically used to identify roof beams, support columns or other complex structural elements. Each feature recognition unit is trained to identify different structural features. First, by providing a sample BIM model and training, a first feature recognition result and a second feature recognition result of each feature recognition unit are obtained. These results reflect the classification of complex structures and pseudo-complex structures by each feature recognition unit. Next, according to the homogeneity between the first feature recognition result and the second feature recognition result in each feature recognition unit, a first homogeneity deviation is determined. This homogeneity deviation is used to measure the consistency between each feature recognition unit to ensure that their recognition results for the same structural element are basically consistent. When training the complex BIM model recognition model, the type error and the first homogeneity deviation are combined. The classification ability of the model is adjusted through type error, and the first homogeneity deviation is used to enhance the consistency between the various feature recognition units. With this design, different feature recognition units can be used to analyze and identify complex structural elements such as roof beams and support columns, and ensure that their recognition results remain consistent to a certain extent. This helps the design team better understand and deal with the complex structure of the gymnasium, optimize the design plan and improve construction efficiency. For example, for the identification of the roof beams of the gymnasium, the feature recognition unit can extract and classify features based on the shape, size and position of the structure to achieve accurate identification and analysis.

[0145] In the embodiment of the present invention, the aforementioned step of determining the first homogeneity deviation according to the homogeneity of the first feature recognition result and the second feature recognition result of each feature recognition unit in all or part of the feature recognition units can be implemented through the following example.

[0146] (1) determining a second homogeneity deviation of each feature recognition unit in all the feature recognition units according to the homogeneity of the first feature recognition result and the second feature recognition result of each feature recognition unit in all the feature recognition units;

[0147] (2) Determine a mean deviation of the second homogeneity deviation of each feature recognition unit in all the feature recognition units to obtain the first homogeneity deviation.

[0148] In the embodiment of the present invention, for example, it is assumed that an architectural design company is responsible for a project of a high-rise office building, which has a complex facade and multi-layer floor structure, and the details of each part need to be accurately identified and analyzed.

[0149] An artificial intelligence-based method for automatic identification of complex structures in BIM was adopted, and a complex BIM model recognition model was used to handle this project.

[0150] The complex BIM model recognition model consists of multiple stacked feature recognition units, which are used to recognize different structural elements, such as floors, facades, etc. Each feature recognition unit is trained to recognize different structural features.

[0151] Next, a first homogeneity deviation is determined based on the homogeneity between the first feature recognition result and the second feature recognition result of each feature recognition unit in all feature recognition units. This homogeneity deviation is used to evaluate the consistency between different feature recognition units.

[0152] For each feature recognition unit, a second homogeneity deviation is further determined according to the homogeneity of the first feature recognition result and the second feature recognition result of each feature recognition unit in all the feature recognition units.

[0153] Finally, the deviation mean of the second homogeneity deviation of each feature recognition unit in all feature recognition units is calculated to obtain the first homogeneity deviation. The deviation mean is used to quantify the homogeneity differences between the feature recognition units. Designed in this way, in high-rise office building projects, this method can use different feature recognition units to analyze and identify different parts of complex structures such as floors and facades, and ensure that their recognition results remain consistent to a certain extent. For example, based on the first feature recognition results and the second feature recognition results, the feature recognition unit can identify and analyze features such as the floor plan and facade materials of different floors, and evaluate the consistency between the feature recognition units by calculating the deviation mean of the second homogeneity deviation. In this way, the design team can better understand and handle complex building structures, optimize design solutions and improve construction efficiency.

[0154] In the embodiment of the present invention, the step of determining the deviation mean of the second homogeneity deviation of each feature recognition unit in all the feature recognition units to obtain the first homogeneity deviation may be implemented through the following example.

[0155] (1) obtaining a weight corresponding to each feature recognition unit in all the feature recognition units, wherein the weight is related to a stacking sequence of the feature recognition unit in all the feature recognition units;

[0156] (2) Determine a weighted deviation mean of the second homogeneity deviation of each feature identification unit in all the feature identification units according to the weight corresponding to each feature identification unit in all the feature identification units and the second homogeneity deviation, and obtain the first homogeneity deviation.

[0157] In an embodiment of the present invention, illustratively, it is assumed that an architectural design company is responsible for a project of a high-rise office building. The office building has a complex facade and a multi-layer floor structure, and the details of each part need to be accurately identified and analyzed. An automatic identification method for complex structures in BIM based on artificial intelligence is adopted, and a complex BIM model recognition model is used to process this project. The complex BIM model recognition model consists of a plurality of stacked feature recognition units for identifying different structural elements, such as floors, facades, etc. Each feature recognition unit is trained to recognize different structural features. In order to calculate the first homogeneity deviation, the architectural design company obtains the weight corresponding to each feature recognition unit in all feature recognition units. In this example, the weight of the floor feature recognition unit is 0.6, and the weight of the facade feature recognition unit is 0.4. Next, according to the weight corresponding to each feature recognition unit in all feature recognition units and the second homogeneity deviation, the weighted deviation mean of the second homogeneity deviation of each feature recognition unit in all feature recognition units is calculated. Assume that the second homogeneity deviation of the floor feature recognition unit is 0.2, and the second homogeneity deviation of the facade feature recognition unit is 0.3. According to the calculation of weights and weighted deviation means, the weighted deviation means of the floor feature recognition unit are 0.12 (0.6*0.2) and the weighted deviation means of the facade feature recognition unit are 0.12 (0.4*0.3). Finally, the sum of the weighted deviation means of the second homogeneity deviation of each feature recognition unit in all feature recognition units is calculated, and the first homogeneity deviation is 0.24 (0.12+0.12). With this design, different feature recognition units can be used to analyze and identify different parts of complex structures such as floors and facades, and the homogeneity differences between each feature recognition unit can be quantified by considering the weights and weighted deviation means. In this way, the design team can better understand and deal with complex building structures, optimize design solutions and improve construction efficiency.

[0158] In the embodiment of the present invention, the aforementioned step of obtaining a sample BIM model file can be implemented through the following example.

[0159] (1) Obtaining the at least one complex BIM model structure;

[0160] (2) generating a plurality of pseudo-complex BIM model structures corresponding to each complex BIM model structure through a plurality of pseudo-complex BIM model data creation strategies, wherein the plurality of pseudo-complex BIM model data creation strategies are used to remove different core complex elements in the complex BIM model structure;

[0161] (3) The at least one complex BIM model structure and the multiple pseudo-complex BIM model structures corresponding to each complex BIM model structure are used as the sample BIM model files.

[0162] In an embodiment of the present invention, illustratively, it is assumed that an architectural design company is developing an automatic identification system for complex structures in BIM based on artificial intelligence. It is hoped that a complex commercial complex project can be processed through this system. In this scenario, in order to train the automatic identification model, a complex BIM model structure containing multiple core complex elements is first obtained. This BIM model includes multiple structural elements such as floors, facades, pipelines, and electrical equipment. In order to increase the diversity and coverage of sample data, a variety of pseudo-complex BIM model data creation strategies are used. For example, for each core complex element, a strategy can be created to remove the element from the complex BIM model structure and generate a pseudo-complex BIM model structure. In this way, multiple pseudo-complex BIM model structures can be obtained, each of which removes different core complex elements to some extent. Through the above strategies, multiple pseudo-complex BIM model structures are generated to form a sample set with the original complex BIM model structure. These pseudo-complex BIM models have different changes and configurations, covering various possible complex structure situations. Finally, the obtained at least one complex BIM model structure and the multiple pseudo-complex BIM model structures corresponding to each complex BIM model structure are used as sample BIM model files for training the automatic identification model. Through the above methods, the artificial intelligence-based complex structure automatic identification method in BIM uses a variety of pseudo-complex BIM model data creation strategies to generate pseudo-complex BIM models containing various changes and configurations, and trains them together with real complex BIM models as samples. Such training can improve the automatic recognition model's ability to recognize different types of complex structures, enabling it to accurately analyze, identify and understand different parts of various complex building projects. Design teams can use this system to quickly and accurately process complex BIM models, optimize design solutions and improve construction efficiency. For example, in a commercial complex project, the system can help automatically identify the location and attributes of different floors, facades and equipment, so as to better understand and process the overall structure and improve the planning and management level of the project.

[0163] In an embodiment of the present invention, the aforementioned step of generating multiple pseudo-complex BIM model structures corresponding to each complex BIM model structure through multiple pseudo-complex BIM model data creation strategies can be implemented through the following implementation methods.

[0164] (1) generating at least one first pseudo-complex BIM model structure corresponding to each complex BIM model structure by at least one first pseudo-complex BIM model data creation strategy; and generating at least one second pseudo-complex BIM model structure corresponding to each complex BIM model structure by at least one second pseudo-complex BIM model data creation strategy;

[0165] Among them, the first pseudo-complex BIM model data creation strategy is used to remove surface complex elements in the complex BIM model structure, and the second pseudo-complex BIM model data creation strategy is used to remove structural complex elements in the complex BIM model structure.

[0166] In an embodiment of the present invention, for example, it is assumed that an architectural design company is using an automatic recognition system for complex structures in BIM based on artificial intelligence to handle a large commercial center project. In this scenario, in order to train the automatic recognition model, a complex BIM model structure containing multiple core complex elements is first obtained. This BIM model includes multiple structural elements such as floors, facades, pipelines, electrical equipment, etc. In order to increase the diversity and coverage of sample data, a variety of pseudo-complex BIM model data creation strategies are adopted. Among them, a first pseudo-complex BIM model data creation strategy and a second pseudo-complex BIM model data creation strategy are used. The first pseudo-complex BIM model data creation strategy is used to remove surface complex elements, such as decorations, billboards, etc., in a complex BIM model structure. Through this strategy, at least one first pseudo-complex BIM model structure is generated, in which the surface complex elements are removed or simplified. The second pseudo-complex BIM model data creation strategy is used to remove structural complex elements, such as certain structural components, complex support systems, etc., in a complex BIM model structure. Through this strategy, at least one second pseudo-complex BIM model structure is generated, in which the structural complex elements are removed or simplified. Through the above strategy, multiple pseudo-complex BIM model structures corresponding to each complex BIM model structure are generated, including multiple first pseudo-complex BIM models and multiple second pseudo-complex BIM models. Finally, the obtained at least one complex BIM model structure and multiple first pseudo-complex BIM models corresponding to each complex BIM model structure and multiple second pseudo-complex BIM models are used as sample BIM model files for training the automatic recognition model. Through the above method, the complex structure automatic recognition method in BIM based on artificial intelligence uses different pseudo-complex BIM model data creation strategies to remove different types of complex elements. The trained automatic recognition model can accurately analyze, identify and understand different parts in various complex building projects, including surface complex elements and structural complex elements. The design team can use this system to quickly and accurately process complex BIM models and improve design efficiency and accuracy. For example, in a commercial center project, the system can help automatically identify and analyze different floors, facades, and core complex elements such as pipelines and electrical equipment, and generate multiple pseudo-complex BIM models for further analysis and decision-making. By removing or simplifying complex surface elements and structural complex elements, the system can help the design team better understand the overall structure and spatial layout, optimize the design plan and improve the feasibility and efficiency of construction.

[0167] In an embodiment of the present invention, the aforementioned step of performing feature recognition on the complex BIM model structure and the pseudo-complex BIM model structure respectively through the complex BIM model recognition model to obtain the first classification feature of the complex BIM model structure and the second classification feature of the pseudo-complex BIM model structure can be implemented through the following example.

[0168] (1) performing feature recognition on the complex BIM model structure and the pseudo-complex BIM model structure respectively by using the complex BIM model recognition model to obtain the first classification feature and the second classification feature;

[0169] The first classification feature and the second classification feature are used to infer that the category of the input BIM model structure is a complex BIM model category, or to infer a pseudo-complex BIM model data creation strategy corresponding to the input BIM model structure.

[0170] In an embodiment of the present invention, for example, it is assumed that an architectural design company is using an automatic identification system for complex structures in BIM based on artificial intelligence to handle a large hospital project. The BIM model of the project contains multiple complex structural elements, such as floors, facades, pipelines, electrical equipment, etc. The design team first inputs the BIM model into the complex BIM model identification model. The complex BIM model identification model analyzes and classifies the BIM model through feature recognition. It will perform feature recognition on the complex BIM model structure to obtain the first classification feature, which describes the complexity, diversity and different types of elements contained in the complex BIM model structure. At the same time, during the training phase, the design team also generated multiple pseudo-complex BIM model structures, some of which removed or simplified surface complex elements (such as decorations, billboards), and others removed or simplified structural complex elements (such as certain structural components, complex support systems). These pseudo-complex BIM model structures are used as training samples and input into the complex BIM model recognition model for feature recognition to obtain second classification features. These features describe the degree of simplification of the pseudo-complex BIM model structure, the complex elements removed, etc. In actual operation, when the design team needs to process a new BIM model, they input the BIM model into the complex BIM model recognition model. The complex BIM model recognition model analyzes the input BIM model through feature recognition and makes inferences based on the obtained first classification features and second classification features. If the features of the input BIM model match the first classification features of the complex BIM model, it can be inferred that the input BIM model belongs to the complex BIM model category. If the features of the input BIM model match the second classification features of the pseudo-complex BIM model, it can be inferred that the input BIM model corresponds to the pseudo-complex BIM model data creation strategy. Through the above method, the complex structure automatic recognition method in BIM based on artificial intelligence uses the complex BIM model recognition model to perform feature recognition on the input BIM model structure and infer its category. The design team can quickly analyze, understand and process different types of BIM model structures based on the results of automatic recognition, thereby optimizing the design plan, improving construction efficiency, and ensuring the successful implementation of the project. In the hospital project, by automatically identifying the complex BIM model category, the design team can focus on processing and optimizing key elements such as the complex floor layout, piping system, and electrical equipment inside the hospital. For structures that adopt a pseudo-complex BIM model data creation strategy, corresponding adjustments can be made based on the obtained classification characteristics to restore the complexity of the original design.

[0171] In an embodiment of the present invention, the aforementioned step of determining the type error based on the first deviation between the first classification feature of the complex BIM model structure and the first target value, and the second deviation between the second classification feature of the pseudo-complex BIM model structure and the second target value can be implemented through the following example.

[0172] (1) Determining the type error according to a first deviation between a first classification feature of the complex BIM model structure and a first target value, and a second deviation between a second classification feature of the pseudo-complex BIM model structure and a real complex BIM model data creation strategy corresponding to the pseudo-complex BIM model structure.

[0173] In an embodiment of the present invention, it is assumed that an architectural design company is using an artificial intelligence-based complex structure automatic recognition system in BIM to handle a high-rise office building project. The design team first inputs the BIM model of the project into the complex BIM model recognition model. The complex BIM model recognition model analyzes and classifies the BIM model through feature recognition. It will perform feature recognition on the complex BIM model structure to obtain a first classification feature, which describes the complexity, diversity and different types of elements contained in the complex BIM model structure. At the same time, during the training phase, the design team generated a plurality of pseudo-complex BIM model structures, some of which removed or simplified surface complex elements (such as decorations, billboards), and others removed or simplified structural complex elements (such as certain structural components, complex support systems). These pseudo-complex BIM model structures are used as training samples and input into the complex BIM model recognition model for feature recognition to obtain second classification features. These features describe the degree of simplification of the pseudo-complex BIM model structure, the complex elements removed, etc. In actual operation, when the design team needs to process a new BIM model, they input the BIM model into the complex BIM model recognition model. The complex BIM model recognition model analyzes the input BIM model through feature recognition, and makes inferences based on the obtained first classification feature and second classification feature. If the feature of the input BIM model matches the first classification feature of the complex BIM model, it can be inferred that the input BIM model belongs to the complex BIM model category. If the feature of the input BIM model matches the second classification feature of the pseudo-complex BIM model, it can be inferred that the input BIM model corresponds to the use of a pseudo-complex BIM model data creation strategy. Next, the type error is determined based on the first deviation of the first classification feature of the complex BIM model structure and the first target value, and the second deviation of the second classification feature of the pseudo-complex BIM model structure and the real complex BIM model data creation strategy corresponding to the pseudo-complex BIM model structure. Through the above method, the complex structure automatic recognition method in BIM based on artificial intelligence uses the complex BIM model recognition model to perform feature recognition on the input BIM model structure and infer its category. The design team can quickly analyze, understand and process different types of BIM model structures based on the results of automatic recognition, thereby optimizing the design plan, improving construction efficiency, and ensuring the successful implementation of the project. In a high-rise office building project, by automatically identifying complex BIM model categories, the design team can focus on processing and optimizing key elements such as floor layout, plumbing systems, and electrical equipment.

[0174] The embodiment of the present invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the aforementioned method for automatically identifying complex structures in BIM based on artificial intelligence. Figure 2 As shown, Figure 2 The computer device 100 is a block diagram of a structure of a computer device 100 provided in an embodiment of the present invention. The computer device 100 includes a memory 111, a processor 112 and a communication unit 113. To achieve data transmission or interaction, the memory 111, the processor 112 and the communication unit 113 are directly or indirectly electrically connected to each other.

[0175] For illustrative purposes, the foregoing description is made with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the present disclosure to the precise form disclosed. Numerous modifications and variations are possible in accordance with the above teachings. These embodiments are selected and described in order to best illustrate the principles of the present disclosure and its practical application, so that those skilled in the art can best utilize the present disclosure and utilize various embodiments with different modifications to suit the intended specific application.

Claims

1. A method for automatic identification of complex structures in BIM based on artificial intelligence, characterized in that: include: In response to a user's query instruction for a current BIM model structure, acquiring the current BIM model structure; Calling a pre-trained complex BIM model recognition model, and inputting the current BIM model structure into the complex BIM model recognition model to obtain a model classification result of the current BIM model structure; If the model classification result indicates that the current BIM model structure is a complex BIM model structure, the current BIM model structure is used as a to-be-determined complex BIM model structure; In response to a query instruction for a target model element in the pending complex BIM model structure, calling a pre-trained feature extraction model; Based on the previously trained feature extraction model, determining the target model element classification result corresponding to the undetermined complex BIM model structure; The complex BIM model recognition model is obtained by the following methods, including: Obtain at least one complex BIM model structure; Generate at least one first pseudo-complex BIM model structure corresponding to each complex BIM model structure by at least one first pseudo-complex BIM model data creation strategy; and generate at least one second pseudo-complex BIM model structure corresponding to each complex BIM model structure by at least one second pseudo-complex BIM model data creation strategy; The first pseudo-complex BIM model data creation strategy is used to remove surface complex elements in the complex BIM model structure, and the second pseudo-complex BIM model data creation strategy is used to remove structural complex elements in the complex BIM model structure; The at least one complex BIM model structure and the multiple pseudo-complex BIM model structures corresponding to each complex BIM model structure are used as a sample BIM model file, wherein the sample BIM model file includes at least one complex BIM model structure and at least one pseudo-complex BIM model structure, and the pseudo-complex BIM model structure is generated based on the complex BIM model structure; Performing feature recognition on the complex BIM model structure and the pseudo-complex BIM model structure respectively through the complex BIM model recognition model to obtain a first classification feature and a second classification feature; The first classification feature and the second classification feature are used to infer that the category of the input BIM model structure is a complex BIM model category, or to infer a pseudo-complex BIM model data creation strategy corresponding to the input BIM model structure, and the complex BIM model recognition model is used to infer whether the input BIM model structure is a complex BIM model structure; Determine a type error according to a first deviation between a first classification feature of the complex BIM model structure and a first target value, and a second deviation between a second classification feature of the pseudo complex BIM model structure and a real complex BIM model data creation strategy corresponding to the pseudo complex BIM model structure; the complex BIM model recognition model includes a plurality of stacked feature recognition units; Determining a second homogeneity deviation of each feature recognition unit in all feature recognition units according to the homogeneity of the first feature recognition result and the second feature recognition result of each feature recognition unit in all feature recognition units; Obtaining a weight corresponding to each feature recognition unit in all the feature recognition units, wherein the weight is related to a stacking sequence of the feature recognition unit in all the feature recognition units; Determine a weighted deviation mean of the second homogeneity deviation of each feature identification unit in all the feature identification units according to the weight corresponding to each feature identification unit in all the feature identification units and the second homogeneity deviation, to obtain a first homogeneity deviation; The complex BIM model recognition model is trained by the type error and the first homogeneity deviation, the first feature recognition result is obtained by the complex BIM model recognition model performing feature recognition on the complex BIM model structure, and the second feature recognition result is obtained by the complex BIM model recognition model performing feature recognition on the pseudo-complex BIM model structure.

2. The method according to claim 1, characterized in that The feature extraction model based on the pre-trained feature extraction model determines the target model element classification result corresponding to the undetermined complex BIM model structure, including: Perform spatial division on the complex BIM model structure to obtain at least two local BIM model structures; By performing an image feature extraction operation on each local BIM model structure based on a feature extraction model trained in advance, a local image feature vector is obtained; wherein the feature extraction model includes a basic feature extraction unit and an advanced feature extraction unit; the basic feature extraction unit is used to perform a basic image feature extraction operation on the local BIM model structure; the advanced feature extraction unit is used to perform an advanced image feature extraction operation on the local BIM model structure based on the basic image features obtained during the basic image feature extraction operation, and the accuracy of the advanced image feature extraction operation is greater than the accuracy of the basic image feature extraction operation; Obtaining an image feature vector of each reference BIM model element in a preset BIM model element library; Determining a vector distance between the local image feature vector and the image feature vector of each reference BIM model element; When the vector distance between the local image feature vector and the image feature vector of any reference BIM model element is less than a preset vector distance threshold, determining that the local model element classification result of the local BIM model structure is an expected BIM element recognition result; The expected BIM element identification result is used to indicate that there is a model element having the same type as the reference BIM model element in the preset BIM model element library in the local BIM model corresponding to the local BIM model structure; When the local model element classification results of the at least two local BIM model structures are both corresponding expected BIM element recognition results, it is determined that the target model element classification result corresponding to the undetermined complex BIM model structure is the model element classification result determined by the expected BIM element recognition result corresponding to the local model element classification results of the at least two local BIM model structures.

3. The method according to claim 2, characterized in that The spatial division of the complex BIM model structure to be determined to obtain at least two local BIM model structures includes: The undetermined complex BIM model structure is spatially cut using cubes with fixed intervals to obtain at least two local BIM model structures, and the at least two local BIM model structures have the same spatial scale.

4. The method according to claim 3, characterized in that The method further comprises: Get preset specification rules; The preset standardization rule is used to perform data normalization on each of the local BIM model structures, and correspondingly obtain at least two local BIM model structures after data normalization; The performing of an image feature extraction operation on each local BIM model structure to obtain a local image feature vector includes: An image feature extraction operation is performed on each local BIM model structure after data normalization to obtain the local image feature vector.

5. The method according to claim 2, characterized in that: The method performs an image feature extraction operation on each local BIM model structure based on a feature extraction model trained in advance to obtain a local image feature vector, including: Loading each of the local BIM model structures into the basic feature extraction unit, and performing basic feature extraction on the local BIM model structure through the basic feature extraction unit to obtain an image feature vector with basic accuracy; The image feature vector with basic accuracy is loaded into the advanced feature extraction unit, and the advanced feature extraction unit is used to perform advanced feature extraction on the local BIM model structure to obtain an image feature vector with advanced accuracy; the basic accuracy is less than the advanced accuracy.

6. The method according to claim 2, characterized in that The preset BIM model element library includes a reference BIM model structure of each reference BIM model element; and obtaining an image feature vector of each reference BIM model element in the preset BIM model element library includes: By using the feature extraction model based on the pre-training, an image feature extraction operation is performed on the reference BIM model structure of each reference BIM model element to obtain an image feature vector of each reference BIM model element.

7. The method according to claim 2, characterized in that: The basic feature extraction unit includes a geometric feature extraction unit and an attribute feature extraction unit, the advanced feature extraction unit includes a model structure information extraction unit and a weight allocation unit, and the feature extraction model is obtained by the following method, including: Loading a first sample BIM model structure in a basic BIM model structure set into the basic feature extraction unit; Extracting geometric features from the first sample BIM model structure by the geometric feature extraction unit to obtain a basic feature vector; Extracting attribute features from the basic feature vector by the attribute feature extraction unit to obtain an image feature vector having a preset number of attributes; Loading the image feature vector with the preset number of attributes into a first cost unit, and determining a first cost parameter corresponding to the image feature vector with the preset number of attributes through a first cost function in the first cost unit; Adjusting the model architecture in the geometric feature extraction unit and the attribute feature extraction unit based on the first cost parameter to obtain a trained basic feature extraction unit; Loading a second sample BIM model structure in the independent BIM model structure set into the trained basic feature extraction unit, performing basic feature extraction on the second sample BIM model structure through the trained basic feature extraction unit, and obtaining a sample image feature vector with a third accuracy; Loading the sample image feature vector with the third accuracy into the advanced feature extraction unit; Extracting key model features of the sample image feature vector under different model components through the model structure information extraction unit; The weight distribution unit sequentially performs weight calculations on the key model features under the different model components on the model component sequence to obtain a sample image feature vector with a fourth accuracy; the third accuracy is less than the fourth accuracy; By means of a preset category classification unit, based on the sample image feature vector with the fourth accuracy, the second sample BIM model structure is classified into model elements to obtain a sample BIM element recognition result; Loading the sample BIM element recognition result and the target category information of the second sample BIM model structure into a preset cost unit, and outputting a cost parameter through the preset cost unit; The network architecture in the advanced feature extraction unit is adjusted based on the cost parameter to obtain a trained feature extraction model.

8. The method according to claim 7, characterized in that The advanced feature extraction unit includes a cost evaluation unit, and the cost evaluation unit includes a second cost function; the method further includes: Loading the sample image feature vector with fourth accuracy and the feature annotation information of the second sample BIM model structure into the cost evaluation unit; Determining a second cost parameter corresponding to the sample image feature vector having a fourth accuracy by using a second cost function of the cost evaluation unit; Based on the second cost parameter, the model architecture in the model structure information extraction unit and the weight allocation unit is adjusted to obtain a trained advanced feature extraction unit.

Citation Information

Patent Citations

  • BIM model lightweight exporting method and system

    CN111695183A

  • Rail transit component classification method, system and equipment based on BIM (Building Information Modeling)

    CN115661553A

  • Method for automated construction progress resource optimization employing building information model

    WO2021031336A1