A BIM-based three-dimensional construction method for geotechnical space

Abstract

The present invention belongs to the field of three-dimensional modeling technology, and in particular relates to a BIM-based method for constructing three-dimensional geotechnical space. This method comprises: collecting three-dimensional point cloud data and geological drilling data to extract geotechnical parameter characteristics; preprocessing the point cloud data to extract surface feature points and contour lines to generate a point cloud feature dataset; registering and fusing the two types of data to generate a geotechnical spatial fusion dataset; constructing a three-dimensional model based on the dataset using a BIM platform; and performing accuracy verification and optimization on the model. This method improves the accuracy of multi-source data fusion and model reliability through multimodal feature extraction, targeted point cloud preprocessing, hierarchical feature point registration, and comprehensive accuracy verification, meeting the requirements for fine modeling in geotechnical engineering.

Description

Technical Field

[0001] The present invention belongs to the technical field of three-dimensional modeling, and in particular relates to a method for constructing three-dimensional geotechnical space based on BIM. Background Art

[0002] In geotechnical engineering, three-dimensional modeling of geotechnical spaces is crucial for engineering design, construction, and safety assessment. With the development of Building Information Modeling (BIM), it has demonstrated significant advantages in integrating multi-source data and enhancing model visualization and collaboration.

[0003] However, in the existing three-dimensional construction of geotechnical space, the fusion of multi-source data is difficult. Due to the significant differences in sources, formats and features between three-dimensional point cloud data and geological borehole data, problems such as low registration accuracy and poor data integration often occur. Geological borehole data contains multimodal information such as text, images, and numerical values. Traditional feature extraction methods are difficult to fully explore the characteristics of geotechnical geological parameters and are prone to losing key information. In the preprocessing of three-dimensional point cloud data, traditional denoising and filtering methods are not adaptable enough to different curvature areas, resulting in inaccurate extraction of surface feature points and contour lines, affecting the quality of the point cloud feature data set. When aligning feature points, there is a lack of effective optimization objectives and algorithms for matching feature points on geotechnical sub-surface surfaces, and the calculation of the transformation matrix is ​​unreasonable, further reducing the accuracy of data fusion. At the same time, the model accuracy verification method is imperfect, making it difficult to comprehensively evaluate the errors in spatial coordinates and geotechnical parameters, resulting in low accuracy and poor reliability of the constructed three-dimensional model, which cannot meet the engineering needs for fine modeling of geotechnical space. Summary of the Invention

[0004] In response to the technical problems existing in the above background technologies, the present invention proposes a three-dimensional construction method of geotechnical space based on BIM.

[0005] In order to achieve the above object, the technical solution adopted by the present invention includes the following steps:

[0006] S1. Collecting three-dimensional point cloud data and geological drilling data of the rock and soil space, wherein the three-dimensional point cloud data is obtained by a laser scanning device, and performing feature extraction on the geological drilling data to extract rock and soil geological parameter characteristics;

[0007] S2. Preprocessing the three-dimensional point cloud data, including denoising, filtering, and point cloud simplification, extracting surface feature points and contour lines of the rock and soil mass, and generating a point cloud feature data set;

[0008] S3, registering the point cloud feature dataset with the rock and soil geological parameter features, achieving data fusion through rock and soil sub-surface feature point matching and coordinate conversion, and generating a rock and soil spatial fusion dataset;

[0009] S4. Based on the geotechnical spatial fusion dataset, construct a geotechnical spatial three-dimensional model using a BIM platform;

[0010] S5. Performing accuracy verification on the geotechnical spatial three-dimensional model. If the accuracy verification fails, optimizing and adjusting the model parameters until the accuracy verification passes.

[0011] Preferably, the geological drilling data in step S1 includes text data, image data and numerical data.

[0012] Preferably, feature extraction is performed on the geological drilling data to extract rock and soil geological parameter features, specifically comprising:

[0013] First, preprocessing is performed to segment and encode text data in geological drilling data, grayscale and size normalize image data, and normalize and map numerical data.

[0014] A multimodal feature extraction network integrating an attention mechanism was constructed. The multimodal feature extraction network included a text feature branch, an image feature branch, and a numerical feature branch. The text feature branch used a bidirectional LSTM to process word vector sequences, connected a self-attention mechanism to the output layer, and outputted a text feature vector. The image feature branch used an improved ResNet18 network to delete the last three fully connected layers, added a 1×1 convolutional layer to compress the channel dimension, and outputted an image feature vector through a spatial attention module. The numerical feature branch used a three-layer fully connected network to process the numerical sequence and output a numerical feature vector in combination with position encoding.

[0015] The feature fusion layer generates a comprehensive geotechnical parameter feature vector by assigning learnable weights to the text, image, and numerical feature vectors respectively; the learnable weights are obtained by reversely adjusting the accuracy of the validation set during the training process;

[0016] The output layer maps the comprehensive rock and soil geological parameter feature vector through a two-layer fully connected network and outputs the initial rock and soil geological parameter characteristics;

[0017] Finally, a credibility evaluation is performed. Based on the probability distribution of the output layer, the credibility score of each initial geotechnical parameter feature is calculated, and the initial geotechnical parameter features with a credibility score exceeding the set threshold are obtained as the final selected geotechnical parameter features.

[0018] Preferably, the three-dimensional point cloud data is preprocessed in step S2, including denoising, filtering and point cloud simplification, extracting surface feature points and contour lines of the rock and soil body, and generating a point cloud feature data set, which includes:

[0019] First, perform point cloud data denoising. Calculate the local density of each point, set a density threshold, and mark points with density below the threshold as suspected noise points. Then perform secondary screening based on the angle between the point and the normal vector in the neighborhood to remove the noise points.

[0020] Then, filtering is performed. The area with curvature less than or equal to the set threshold is recorded as a flat area, and Gaussian filtering is performed on the flat area. The area with curvature greater than the set threshold is recorded as a feature area, and bilateral filtering is performed on the feature area.

[0021] Then, the curvature value and neighborhood entropy of each point are calculated, and retention weights are assigned to points in different regions based on the curvature value and neighborhood entropy. An adaptive voxel grid is used, and the grid size is dynamically adjusted according to the retention weight, retaining the point with the highest weight in each grid.

[0022] Finally, a point cloud graph structure is constructed, graph node connections are built using neighboring points, and local geometric features are learned through a graph convolutional network. When extracting feature points, the feature probability is predicted through the activation function of the output layer of the graph convolutional network, and points whose feature probabilities meet the preset conditions are selected as surface feature points. The feature points are clustered, and the normal vector angle is used as the clustering constraint. The clustered point sequence is fitted with a curve to generate a continuous rock and soil contour line. The extracted feature points and contour lines are divided into spatial blocks to generate a point cloud feature dataset.

[0023] Preferably, the step S3 registers the point cloud feature dataset with the geotechnical geological parameter features, realizes data fusion through geotechnical surface feature point matching and coordinate transformation, and generates a geotechnical spatial fusion dataset as follows:

[0024] First, the characteristic points of the rock and soil layers are grouped, and the geological parameter characteristics of the rock and soil are extracted to extract the landmark characteristic points of each layer. ,in is the layer sequence number, Index the feature points in the layer; for the point cloud feature data set, identify the lithologic interface through the curvature mutation point of the contour line and extract the surface feature points of each layer ;

[0025] Taking the grouped feature points as input, the objective function whose optimization goal is to minimize the sum of weighted Euclidean distances is defined as: ,in is the total number of layers, is the number of feature point pairs in the kth layer, is the weight parameter, is the transformation matrix;

[0026] Particle swarm optimization algorithm is used to search for the optimal , iterate until the maximum number of iterations stops;

[0027] The overall transformation matrix obtained based on particle swarm optimization , perform layered optimization processing, for the Layered transformation matrix , update the transformation matrix by minimizing the weighted sum of squared distances , the calculation method of the updated transformation matrix is: ,in For the The number of valid point pairs after stratified screening;

[0028] When the average residual within a layer is less than the set threshold, the layer iteration is stopped and the optimal transformation matrices of all layers are collected;

[0029] Based on the optimal transformation matrix of each layer, the point cloud feature dataset and rock and soil geological parameter features are uniformly transformed into the global coordinate system;

[0030] The attribute information of rock and soil geological parameter characteristics is mapped to the point cloud feature dataset and integrated into the rock and soil spatial fusion dataset based on the coordinates.

[0031] Preferably, the step S4 is implemented by constructing a geotechnical spatial three-dimensional model using a BIM platform based on the geotechnical spatial fusion dataset, specifically comprising:

[0032] First, the geotechnical spatial fusion dataset is converted into an exchangeable data format compatible with the BIM platform;

[0033] Generate BIM structures corresponding to each layer based on geotechnical spatial fusion data set modeling;

[0034] Combine each layered BIM structure into a BIM structure model according to the actual spatial position relationship, clarify the upper and lower interfaces and adjacency relationships between layers, and generate a complete geotechnical spatial three-dimensional model.

[0035] Preferably, the accuracy check of the geotechnical spatial three-dimensional model in step S5 is performed by selecting on-site sampling points, obtaining the measured spatial coordinates and corresponding geotechnical parameter values ​​of the sampling points, extracting the predicted coordinates and predicted parameter values ​​of the same spatial positions in the model, and calculating the errors between the two; wherein, the spatial coordinate error is calculated using the Euclidean distance, and the geotechnical parameter value error is calculated using the ratio of the absolute difference to the measured value, and the spatial coordinate error and the geotechnical parameter value error are normalized and weighted summed, and compared with the set error threshold; the proportion of the number of sampling points whose errors exceed the threshold is counted, and when the proportion is lower than the set proportion threshold, it is determined that the accuracy check has passed.

[0036] Compared with the prior art, the advantages and positive effects of the present invention are:

[0037] 1. Efficiently extract multimodal geological features, process multimodal data by integrating attention mechanism networks, and combine credibility assessment to reduce the loss of key information.

[0038] 2. Optimize point cloud preprocessing, targeted denoising, regional filtering, and adaptive simplification, combined with graph convolution to extract features and improve the accuracy of surface feature points and contour lines.

[0039] 3. Accurately fusion multi-source data, through hierarchical feature point matching, particle swarm optimization and hierarchical transformation matrix adjustment, improve fusion accuracy and achieve coordinate unification.

[0040] 4. Improve accuracy verification, integrate spatial coordinate and geotechnical parameter errors, and perform normalized weighted evaluation to ensure that the model accuracy and reliability meet engineering requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0042] Figure 1 This is a structural flow diagram of a BIM-based geotechnical space three-dimensional construction method. DETAILED DESCRIPTION

[0043] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0044] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0045] Example 1: In modern geotechnical engineering construction, as the volume and complexity of projects continue to increase, traditional two-dimensional drawings and manual experience are difficult to meet the design and construction management needs of complex underground structures. Especially in engineering scenarios with complex geological conditions and varied topography, how to accurately express geotechnical spatial information and improve the safety and controllability of projects has become a major technical problem in the field of geotechnical engineering. Although existing technologies have attempted to combine building information models (BIM) with geological data, there are still many difficulties in actual operation, which often lead to problems such as poor model accuracy. Therefore, a three-dimensional construction method of geotechnical space based on BIM is proposed. The specific implementation process is as follows: Figure 1 shown.

[0046] First, the three-dimensional point cloud data and geological drilling data of the geotechnical space are collected. The three-dimensional point cloud data is obtained by laser scanning equipment, and the geological drilling data is subjected to feature extraction to extract the geotechnical parameter characteristics. Specifically, the feature extraction of the geological drilling data and the extraction of geotechnical parameter characteristics specifically include: first, preprocessing, segmentation encoding of the text data in the geological drilling data, grayscale and size unification of the image data, and normalization mapping of the numerical data to the [0, 1] interval; constructing a multimodal feature extraction network that integrates the attention mechanism, the multimodal feature extraction network includes a text feature branch, an image feature branch, and a numerical feature branch; the text feature branch uses a bidirectional LSTM to process the word vector sequence, connects the self-attention mechanism to the output layer, and outputs a text feature vector; the image feature branch uses an improved ResNet18 network to delete the last three fully connected layers, and adds a 1×1 convolutional layer to compress the channel dimension. The image feature vector is output through the spatial attention module; the numerical feature branch: a three-layer fully connected network is used to process the numerical sequence, and the numerical feature vector is output in combination with the position encoding; the feature fusion layer performs feature fusion by assigning learnable weights to the text, image, and numerical feature vectors respectively, and generates a comprehensive geotechnical parameter feature vector; the learnable weight is obtained by reversely adjusting the accuracy of the validation set during the training process; the output layer maps the comprehensive geotechnical parameter feature vector through a two-layer fully connected network, and outputs the initial geotechnical parameter feature; finally, a credibility assessment is performed, and the credibility score of each initial geotechnical parameter feature is calculated based on the probability distribution of the output layer, and the initial geotechnical parameter feature with a credibility score exceeding the set threshold is obtained as the final selected geotechnical parameter feature.

[0047] Then, in order to reduce redundant noise points and improve the quality of point cloud modeling, the three-dimensional point cloud data is preprocessed, including denoising, filtering and point cloud simplification, and the surface feature points and contour lines of the rock and soil body are extracted to generate a point cloud feature data set. Specifically, the point cloud data denoising operation is first performed, the local density of each point is calculated, and a density threshold is set. Points with a density lower than the threshold are recorded as suspected noise points, and the noise points are removed by performing a secondary screening based on the angle between them and the normal vector in the neighborhood; then filtering is performed, and areas with curvature less than or equal to the set threshold are recorded as flat areas, and Gaussian filtering is performed on the flat areas. Areas with curvature greater than the set threshold are recorded as feature areas, and bilateral filtering is performed on the feature areas; then the curvature value and neighborhood entropy of each point are calculated, and retention weights are assigned to points in different areas based on the curvature value and neighborhood entropy; an adaptive voxel grid is used to select the points with the same density. The grid size is dynamically adjusted according to the retention weight, retaining the point with the highest weight within each grid. Finally, a point cloud graph structure is constructed, and graph nodes are connected using neighboring points. Local geometric features are learned through a graph convolutional network. When extracting feature points, the activation function of the graph convolutional network output layer is used to predict feature probabilities, and points whose feature probabilities meet preset conditions are selected as surface feature points. Feature points are clustered, using the normal vector angle as the clustering constraint. The clustered point sequence is then fitted with a curve to generate a continuous geotechnical contour line. The extracted feature points and contour lines are divided into spatial blocks to generate a point cloud feature dataset. This complete process not only achieves data noise reduction and simplification, but also significantly improves the boundary clarity and local geometric expression capabilities of geotechnical 3D model construction.

[0048] The point cloud feature dataset is aligned with the rock and soil geological parameter features, and data fusion is achieved through rock and soil layer feature point matching and coordinate transformation to generate a rock and soil spatial fusion dataset. Specifically, the rock and soil layer feature points are first grouped, and the rock and soil geological parameter features are extracted to extract the landmark feature points of each layer. ,in is the layer sequence number, Index the feature points in the layer; for the point cloud feature data set, identify the lithologic interface through the curvature mutation point of the contour line and extract the surface feature points of each layer ; Taking the grouped feature points as input, the objective function is defined as the minimum sum of weighted Euclidean distances. ,in is the total number of layers, is the number of feature point pairs in the kth layer, is the weight parameter, The objective function is to find the transformation matrix that can make the two types of feature point sets overlap in space by minimizing the sum of the weighted distances. The particle swarm optimization algorithm is used to search for the optimal , iterate until the maximum number of iterations is reached; the overall transformation matrix obtained based on particle swarm optimization , perform layered optimization processing, for the Layered transformation matrix , Directly inherit the overall transformation matrix obtained by global particle swarm optimization , update the transformation matrix by minimizing the weighted sum of squared distances , the calculation method of the updated transformation matrix is: ,in For the The number of valid point pairs after hierarchical screening is obtained by K-nearest neighbor In-field screening and Matched point pairs are obtained by removing abnormal point pairs whose differences are greater than twice the average difference within the layer , The function is to find a way to make each layer smallest When the average residual within a layer falls below a set threshold, the layer-by-layer iteration is terminated and the optimal transformation matrices for all layers are pooled. Specifically, during the registration process between the point cloud and geotechnical parameters, the optimal transformation matrix is ​​independently calculated for each layer to ensure registration accuracy. During the layer-by-layer optimization iteration, the weighted Euclidean distance between all feature point pairs within the layer is used as the evaluation metric, and the transformation matrix is ​​continuously updated through iterative optimization. When the average residual between all valid point pairs within a layer falls below a preset accuracy threshold, the registration optimization for that layer is considered complete, and the iteration process for the current layer is terminated. Finally, the optimal registration results for each layer are aggregated to form a set of layer-by-layer optimal transformation matrices covering the entire geotechnical region, laying the foundation for subsequent global coordinate unification. Based on the optimal transformation matrices for each layer, the point cloud feature dataset and geotechnical parameters are uniformly transformed into a global coordinate system. Specifically, after pooling the optimal transformation matrices for each layer, these local registration results are applied to the coordinate system transformation of the entire point cloud feature dataset and geotechnical parameters. Specifically, a transformation matrix is ​​used to coordinate map each layer of point cloud data and its corresponding geological parameter points, unifying them all into the same global coordinate system to ensure the consistency and continuity of the spatial structure. The unified coordinate system not only eliminates coordinate deviations from the original data source but also provides a unified foundation for BIM platform modeling. Finally, the attribute information of the geotechnical geological parameter features is mapped to the point cloud feature dataset and integrated with the coordinates to form a geotechnical spatial fusion dataset. That is, the attribute information in the geotechnical geological parameter features is mapped point by point to the corresponding point cloud feature data point according to the coordinate position matching rules. Each point cloud point thus not only has spatial coordinates and surface geometric features, but also carries the corresponding geological attribute information. Through the dual integration of coordinates and attributes, a complete and semantically rich geotechnical spatial fusion dataset is ultimately constructed. This fusion dataset, serving as the basic input for BIM three-dimensional modeling, greatly improves the model's expression accuracy and application value.

[0049] Based on the geotechnical spatial fusion dataset, a geotechnical spatial 3D model is constructed using the BIM platform. First, the geotechnical spatial fusion dataset is converted into an exchangeable data format compatible with the BIM platform. Based on the geotechnical spatial fusion dataset model, BIM structures corresponding to each layer are generated. The BIM structures of each layer are combined into BIM structure models based on their actual spatial positions, with the upper and lower interfaces and adjacency relationships between layers clearly defined, to generate a complete geotechnical spatial 3D model. Specifically, to efficiently construct and visualize the geotechnical spatial 3D model on the BIM platform, the registered and fused geotechnical spatial dataset must first be converted into an exchangeable data format compatible with the BIM system. Common exchangeable formats include IFC (Industry Foundation Classes) or custom intermediate formats (such as XML and JSON). During the conversion process, the data structure specifications of the BIM platform must be strictly adhered to to avoid format loss or information misalignment. After the format conversion is completed, corresponding BIM structures are constructed based on the spatial extent and geological characteristics of each layer in the fused dataset. Each structure represents a specific geotechnical layer entity. During modeling, a regular body or curved surface model is generated using contour lines, and the model surface is finely divided to ensure that it conforms to the geometric characteristics of the original point cloud. This process not only preserves the three-dimensional spatial relationship of the strata, but also achieves a high degree of restoration from data to model. Subsequently, each layered BIM structure is combined into a complete geotechnical structure model based on its actual spatial position. The upper and lower interfaces between adjacent layers are automatically identified, and a complete hierarchical topological structure is established through boundary connections between surfaces. At the same time, the adjacency relationship between each structure is analyzed to ensure the continuity and logical rationality of the model structure, and a complete geotechnical spatial three-dimensional model is generated, which contains the geotechnical geological attribute information in the fused dataset.

[0050] Finally, the geotechnical spatial three-dimensional model is subjected to an accuracy check. If it fails the accuracy check, the model parameters are optimized and adjusted until it passes the accuracy check. The accuracy check is to select on-site sampling points, obtain the measured spatial coordinates of the sampling points and the corresponding geotechnical parameter values, extract the predicted coordinates and predicted parameter values ​​of the same spatial position in the model, and calculate the error between the two; wherein, the spatial coordinate error is calculated using the Euclidean distance, and the geotechnical parameter value error is calculated using the ratio of the absolute difference to the measured value. The spatial coordinate error and the geotechnical parameter value error are normalized and weighted summed, and compared with the set error threshold; the percentage of sampling points with errors exceeding the threshold is statistically calculated. When the percentage is lower than the set percentage threshold, the accuracy check is determined to have passed. For those that fail the accuracy check, a neural network algorithm is used to optimize and adjust the model parameters until the accuracy check step is passed.

[0051] Example 2. For the step of collecting and preprocessing data, the geological borehole data we collected includes text-type, image-type, and numerical-type data. For each type of data, the following preprocessing operations are carried out: Text-type data: For example, the description information of the rock and soil layers such as "Layer 3: Sandy soil, good density". The text is segmented and encoded, and the description of each rock and soil layer is decomposed into words or phrases. For example, "Layer 3" is transformed into ["Layer", "3"], and the subsequent rock and soil description "Sandy soil, good density" is transformed into ["Sandy soil", "Density", "Good"]. Image-type data: The microscopic structure image of the core. The image is grayscale processed and the size is unified to ensure that each image input into the model has the same resolution (such as 256×256 pixels). Numerical-type data: The physical parameters of the rock and soil layers (such as shear strength, etc.). The data is normalized and mapped to the interval [0,1]. Assuming that the original data shear strength range is [10, 50], then for the value of 25 kPa, its normalized value is (25 - 10) / (50 - 10) = 0.375.

[0052] After extracting various feature vectors, they are processed through a bidirectional LSTM to obtain a text feature vector representing the semantic vector of each word. Using a self-attention mechanism, a comprehensive 32-dimensional text feature vector is obtained. For example, the text "sand soil, good density" is processed as [0.34, 0.56, 0.12, ...], representing the semantic information of the textual description of the soil layer. Using the spatial attention module, the output image feature vector may be a 32-length vector, such as [0.25, 0.12, 0.80, ...], representing the microstructural characteristics of the rock core. Numerical data (such as shear strength) is processed through a fully connected layer, potentially outputting a 32-length numerical feature vector, such as [0.375, 0.42, 0.61, ...], representing the physical properties of the soil layer. Learnable weights are assigned to the text, image, and numerical feature vectors, and weighted averaged in the fusion layer. Assuming we use learning weights of 0.3, 0.4, and 0.3, the final comprehensive geotechnical parameter feature vector is the weighted sum of the three vectors. The text, image, and numerical feature vectors are: [0.34, 0.56, 0.12, ...] for text features, [0.25, 0.12, 0.80, ...] for image features, and [0.375, 0.42, 0.61, ...] for numerical features. After weighting, the resulting comprehensive feature vector is [0.34 × 0.3 + 0.25 × 0.4 + 0.375 × 0.3, 0.56 × 0.3 + 0.12 × 0.4 + 0.42 × 0.3, ...]. After weighting, the resulting comprehensive feature vector is [0.31, 0.33, 0.45, ...]. The comprehensive feature vector is mapped using the Softmax function at the output layer to generate a credibility distribution for each feature. Suppose the probability distribution of the shear strength feature vector of a certain geotechnical layer after Softmax processing is: [0.65, 0.35], which means that there is a 65% probability that this feature is the dominant feature of shear strength and a 35% probability that it is a secondary feature. Set a confidence threshold (for example, 0.6). If the confidence exceeds the threshold (such as 0.65), then the feature is selected as the final geotechnical parameter. After processing text, image, and numerical features, if text data is used alone as a feature for modeling, the prediction error is ±15%. However, after combining multimodal data, the error is reduced to ±5%. That is, by combining multimodal data, the error of the model in geotechnical feature extraction is reduced by about 2 times, reflecting the advantage of the fusion method in improving accuracy.

[0053] During point cloud data preprocessing, denoising is crucial for improving model quality. The local density of each point is calculated, a density threshold is set, and suspected noise points with low density are removed. Noise points are then removed through secondary filtering using the angle between the points and the normal vectors within the neighborhood. Initially, a Trimble X7 laser scanner was used to scan the slope's entire length of 200 meters and width of 50 meters, with a point cloud density of 30 points / ㎡, resulting in 12 million raw point cloud points. Point cloud denoising was performed by calculating the local density with a threshold of 8 points / ㎡, marking points with a density <8 as suspected noise points. Secondary filtering using the angle between the normal vectors within the neighborhood (>25°) removed 600,000 noise points (5%). Filtering: The curvature threshold was set to 0.3, and Gaussian filtering (σ=0.8) was applied to flat areas (such as silty clay surfaces). Bilateral filtering (spatial σ=1.5, grayscale σ=0.05) was used to filter characteristic areas (such as the interface between gravel soil and shale). The resulting point cloud smoothness improved by 35%. Point cloud reduction: Retention weights (0.2-0.9) were assigned based on curvature (0-0.8) and neighborhood entropy (0-1.5). The adaptive voxel grid size was dynamically adjusted (0.3m for weights > 0.7, 0.6m for weights < 0.4). The final point cloud retained 5 million points (a 58% reduction, with a 96% retention rate of key feature points). Feature extraction was then performed. A graph convolutional network learned local geometric features, and an activation function predicted feature probabilities. Points with probabilities > 0.65 were selected as surface feature points (e.g., slope crest inflection points and rock layer interface vertices, totaling 30,000 points). Feature points were clustered using a constraint of normal vector angles < 15°, and contour lines were generated using curve fitting (e.g., the boundary curve between gravel soil and shale, with a fitting error < 0.08m). The points were divided into 10m×10m×3m blocks, generating a point cloud feature dataset (including feature point coordinates, normal vectors, and contour line parameters). Traditional denoising methods are typically based on static density thresholds or simple filtering methods. These methods use a global threshold to remove low-density points, but fail to account for the spatially localized nature of point cloud density variations, making it easy to miss important feature points. Traditional denoising methods remove 30% of redundant points, leaving 70% of the point cloud data remaining after denoising. However, this solution not only considers the global point cloud density but also dynamically removes noise points based on local features, thereby more accurately retaining key information. This results in a more streamlined point cloud, a 45% reduction in data volume, and the preservation of the most critical information.

[0054] Next, data registration is performed to align the point cloud feature dataset with the geotechnical parameter features. Geotechnical parameter features: extract the landmark feature points of three layers (such as the top elevation points of each layer); for the point cloud feature dataset, identify the interface through the contour curvature mutation point and extract the surface feature points of each layer. Based on the objective function, particle swarm optimization is used, and after 40 iterations, the error of the global transformation matrix T is < 0.08m. For the obtained global transformation matrix, each layer is optimized separately, and the kth layer is updated by minimizing the weighted sum of squared distances. The transformation matrix of each layer is generated. When the average residual within a layer falls below a set threshold, the layer iteration is terminated and the optimal transformation matrices for all layers are pooled. Based on the optimal transformation matrix for each layer, the point cloud feature dataset and geotechnical parameter features are uniformly transformed into a global coordinate system. The attribute information of the geotechnical parameter features is mapped to the point cloud feature dataset and, combined with the coordinates, integrated into a geotechnical spatial fusion dataset. The final geotechnical spatial fusion dataset includes both spatial coordinate information and geotechnical attribute information. Traditionally, point cloud data only contains geometric coordinates, while geological parameters are often stored as separate documents. Querying the geotechnical attributes of a point requires manual comparison, which is inefficient and prone to errors. This method achieves coordinate unification through the layered optimal transformation matrix. The average residual within each layer can be controlled to within 0.05m (for example, the registration error of the weathered rock layer in one project was reduced from 0.28m to 0.04m). This ensures clear layer boundaries and accurate spatial topological relationships, providing a reliable foundation for subsequent layered modeling of BIM models.

[0055] Then, the model is constructed, and a three-dimensional geotechnical spatial model is constructed using the BIM platform based on the geotechnical spatial fusion dataset;

[0056] Finally, the three-dimensional geotechnical spatial model is precision-checked. If it fails the precision check, the model parameters are optimized and adjusted until it passes the precision check. In traditional methods, the precision check of the model is often performed by comparing only the spatial coordinate error. If the existing spatial coordinate error is 2.4 meters, the set standard value is 2.5 meters, and the error threshold is 0.95, the traditional method only considers the coordinate error and calculates it as 2.4 / 2.5=0.96. The method of the present invention takes into account the geotechnical parameter error while considering the spatial coordinate error. For the same data, if the existing spatial coordinate error is 2.4, the set standard value is 2.5, the geotechnical parameter error is 0.13, and the set standard is 0.15, then according to the comprehensive error calculation method, the thresholds are determined to be 0.7 and 0.3 to calculate the comprehensive error. The comprehensive error = 0.7×0.96+0.3×0.87=0.672+0.261=0.933, which is less than 0.96. The percentage of sampling points with errors exceeding the threshold is calculated. When the percentage is lower than the set percentage threshold, the precision check is determined. This situation demonstrates the advantages of the method of the present invention over traditional methods in terms of comprehensiveness and rigor of precision verification, especially in considering geotechnical parameter errors and comprehensive errors, which can more accurately evaluate and optimize the model, thereby improving the overall reliability and accuracy of the geotechnical spatial model.

[0057] The above description is merely a preferred embodiment of the present invention and does not constitute any other form of limitation to the present invention. Any person skilled in the art may utilize the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes for application in other fields. However, any simple modification, equivalent change, and modification of the above embodiments made in accordance with the technical essence of the present invention without departing from the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A BIM-based three-dimensional geotechnical space construction method, characterized in that: The following steps are involved: S1. Collecting three-dimensional point cloud data and geological drilling data of the rock and soil space, wherein the three-dimensional point cloud data is obtained by a laser scanning device, and performing feature extraction on the geological drilling data to extract rock and soil geological parameter characteristics; S2. Preprocessing the three-dimensional point cloud data, including denoising, filtering, and point cloud simplification, extracting surface feature points and contour lines of the rock and soil mass, and generating a point cloud feature data set; S3, registering the point cloud feature dataset with the rock and soil geological parameter features, achieving data fusion through rock and soil sub-surface feature point matching and coordinate conversion, and generating a rock and soil spatial fusion dataset; S4. Based on the geotechnical spatial fusion dataset, construct a geotechnical spatial three-dimensional model using a BIM platform; S5. Performing an accuracy check on the geotechnical spatial three-dimensional model. If the accuracy check fails, optimizing and adjusting the model parameters until the accuracy check passes. The implementation of preprocessing the three-dimensional point cloud data in step S2, including denoising, filtering and point cloud simplification, extracting surface feature points and contour lines of the rock and soil body, and generating a point cloud feature data set includes: First, perform point cloud data denoising. Calculate the local density of each point, set a density threshold, and mark points with density below the threshold as suspected noise points. Then perform secondary screening based on the angle between the point and the normal vector in the neighborhood to remove the noise points. Then, filtering is performed. The area with curvature less than or equal to the set threshold is recorded as a flat area, and Gaussian filtering is performed on the flat area. The area with curvature greater than the set threshold is recorded as a feature area, and bilateral filtering is performed on the feature area. Then, the curvature value and neighborhood entropy of each point are calculated, and retention weights are assigned to points in different regions based on the curvature value and neighborhood entropy. An adaptive voxel grid is used, and the grid size is dynamically adjusted according to the retention weight, retaining the point with the highest weight in each grid. Finally, a point cloud graph structure is constructed, graph node connections are built using neighboring points, and local geometric features are learned through a graph convolutional network. When extracting feature points, the feature probability is predicted through the activation function of the output layer of the graph convolutional network, and points whose feature probabilities meet the preset conditions are selected as surface feature points. The feature points are clustered, and the normal vector angle is used as the clustering constraint. The clustered point sequence is fitted with a curve to generate a continuous rock and soil contour line. The extracted feature points and contour lines are divided into spatial blocks to generate a point cloud feature dataset.

2. The method for constructing a three-dimensional geotechnical space based on BIM according to claim 1, characterized in that: The geological drilling data in step S1 includes text data, image data and numerical data.

3. The method for constructing a three-dimensional geotechnical space based on BIM according to claim 2, characterized in that: The feature extraction of the geological drilling data to extract the rock and soil geological parameter features specifically includes: First, preprocessing is performed to segment and encode text data in geological drilling data, grayscale and size normalize image data, and normalize and map numerical data. A multimodal feature extraction network integrating an attention mechanism was constructed. The multimodal feature extraction network included a text feature branch, an image feature branch, and a numerical feature branch. The text feature branch used a bidirectional LSTM to process word vector sequences, connected a self-attention mechanism to the output layer, and outputted a text feature vector. The image feature branch used an improved ResNet18 network to delete the last three fully connected layers, added a 1×1 convolutional layer to compress the channel dimension, and outputted an image feature vector through a spatial attention module. The numerical feature branch used a three-layer fully connected network to process the numerical sequence and output a numerical feature vector in combination with position encoding. The feature fusion layer generates a comprehensive geotechnical parameter feature vector by assigning learnable weights to the text, image, and numerical feature vectors respectively; the learnable weights are obtained by reversely adjusting the accuracy of the validation set during the training process; The output layer maps the comprehensive rock and soil geological parameter feature vector through a two-layer fully connected network and outputs the initial rock and soil geological parameter characteristics; Finally, a credibility evaluation is performed. Based on the probability distribution of the output layer, the credibility score of each initial geotechnical parameter feature is calculated, and the initial geotechnical parameter features with a credibility score exceeding the set threshold are obtained as the final selected geotechnical parameter features.

4. The method for constructing a three-dimensional geotechnical space based on BIM according to claim 1, characterized in that: The step S3 aligns the point cloud feature dataset with the geotechnical geological parameter features, and realizes data fusion through geotechnical surface feature point matching and coordinate transformation to generate the geotechnical spatial fusion dataset. The specific implementation is as follows: First, the characteristic points of the rock and soil layers are grouped, and the characteristic points of each layer are extracted based on the characteristics of the rock and soil geological parameters. ,in is the layer sequence number, Index the feature points in the layer; for the point cloud feature data set, identify the lithologic interface through the curvature mutation point of the contour line and extract the surface feature points of each layer ; Taking the grouped feature points as input, the objective function whose optimization goal is to minimize the sum of weighted Euclidean distances is defined as: ,in is the total number of layers, is the number of feature point pairs in the kth layer, is the weight parameter, is the transformation matrix; Particle swarm optimization algorithm is used to search for the optimal , iterate until the maximum number of iterations stops; The overall transformation matrix obtained based on particle swarm optimization , perform layered optimization processing, for the Layered transformation matrix , update the transformation matrix by minimizing the weighted sum of squared distances , the calculation method of the updated transformation matrix is: ,in For the The number of valid point pairs after stratified screening; When the average residual within a layer is less than the set threshold, the layer iteration is stopped and the optimal transformation matrices of all layers are collected; Based on the optimal transformation matrix of each layer, the point cloud feature dataset and rock and soil geological parameter features are uniformly transformed into the global coordinate system; The attribute information of rock and soil geological parameter characteristics is mapped to the point cloud feature dataset and integrated into the rock and soil spatial fusion dataset based on the coordinates.

5. The method for constructing three-dimensional geotechnical space based on BIM according to claim 1, characterized in that: The implementation of step S4 of constructing a geotechnical spatial three-dimensional model using the BIM platform based on the geotechnical spatial fusion dataset specifically includes: First, the geotechnical spatial fusion dataset is converted into an exchangeable data format compatible with the BIM platform; Generate BIM structures corresponding to each layer based on geotechnical spatial fusion data set modeling; Combine each layered BIM structure into a BIM structure model according to the actual spatial position relationship, clarify the upper and lower interfaces and adjacency relationships between layers, and generate a complete geotechnical spatial three-dimensional model.

6. The method for constructing three-dimensional geotechnical space based on BIM according to claim 1, characterized in that: In step S5, the accuracy check of the geotechnical spatial three-dimensional model is performed by selecting on-site sampling points, obtaining the measured spatial coordinates and corresponding geotechnical parameter values ​​of the sampling points, extracting the predicted coordinates and predicted parameter values ​​of the same spatial positions in the model, and calculating the errors between the two; wherein, the spatial coordinate error is calculated using the Euclidean distance, and the geotechnical parameter value error is calculated using the ratio of the absolute difference to the measured value, and the spatial coordinate error and the geotechnical parameter value error are normalized and weighted summed, and compared with the set error threshold; the proportion of the number of sampling points whose errors exceed the threshold is counted, and when the proportion is lower than the set proportion threshold, it is determined that the accuracy check has passed.

Images