Foundation pit inner support scheme generation method, device and medium based on single-mode two-stage model

Abstract

The application discloses a kind of based on single mode two-stage model's foundation pit inner support scheme generation method, equipment and medium, the method includes the following steps: based on the drawing of real building foundation pit inner support scheme, obtains the information data of building foundation pit inner support scheme;The information data of obtaining building foundation pit inner support scheme is preprocessed, and based on the image data set of preprocessed information data is constructed, and is divided into training set and test set;Two-stage model is constructed;Two-stage model takes building foundation pit picture as the input of model, and takes building foundation pit inner support scheme graph as generated image;Model training is carried out using training set;Model verification is carried out using test set;After using the model of verification, foundation pit inner support scheme is generated.The application introduces two-stage model of single mode, uses two-stage generation strategy, first by foundation pit contour to generate structure point image, then by structure point image to generate complete foundation pit support scheme, can effectively improve design efficiency, reduce project cost.

Description

Technical Field

[0001] This invention relates to civil engineering technology, and more particularly to a method, equipment and medium for generating in-situ support schemes for foundation pits based on a single-modal two-stage model. Background Technology

[0002] In the field of civil engineering, foundation pit engineering is a crucial link in infrastructure construction, and the design of the internal support scheme for the foundation pit plays a vital role in ensuring its stability, safety, and construction feasibility. With the acceleration of urbanization, the scale of foundation pit projects is increasing, and the geological conditions and surrounding environment are becoming more complex and diverse.

[0003] Traditional design methods for internal support systems in foundation pits primarily rely on engineers' experience and expertise, involving manual calculations, reference to standards, and past engineering cases. This approach faces numerous challenges. Firstly, designers must expend significant time and effort processing massive amounts of data, including geological survey reports, detailed information on surrounding buildings, and precise measurements of the pit's corner bracing. Furthermore, the difficulty of manual design and calculation increases exponentially when dealing with complex pit shapes, such as irregular polygons or those with internal perforations, increasing the risk of human error. Secondly, traditional methods are relatively inefficient, failing to generate multiple feasible design solutions for comparison and optimization within a short timeframe, thus failing to meet the demands of rapid construction and the need for highly refined design solutions in modern engineering projects. Therefore, a more efficient design method capable of quickly processing complex data is urgently needed. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method, equipment and medium for generating a foundation pit support scheme based on a single-modal two-stage model, which addresses the deficiencies in the prior art.

[0005] The technical solution adopted by this invention to solve its technical problem is: a method for generating an in-situ support scheme for a foundation pit based on a single-modal two-stage model, comprising the following steps: 1) Based on the drawings of the actual building foundation pit support scheme, obtain the information data of the building foundation pit support scheme, including building foundation pit pictures and foundation pit structural point information; the building foundation pit pictures include building foundation pit corner bracing pictures and building foundation pit flat tiling pictures; 2) Preprocess the information data on the support scheme in the building foundation pit, and construct an image dataset based on the preprocessed information data, which is divided into a training set and a test set; 3) Construct a two-stage model; The two-stage model uses images of the building foundation pit as input and a diagram of the support scheme within the foundation pit as the generated image. The specific construction of the two-stage model is as follows: 3.1) First stage: Using the images of the corner bracing and the flat surface of the foundation pit as input, generate structural point information; 3.2) Second stage: Using the structural points generated in the first stage as constraints, adjust the model generation logic and output images of the foundation pit support scheme that meet the structural point constraints; 4) Use the training set for model training; 5) Use the test set to validate the model; 6) Use the validated model to generate the support scheme for the foundation pit.

[0006] According to the above scheme, in step 2), the preprocessing is to convert unstructured drawings into single-modal images and standardized structural point data.

[0007] According to the above scheme, in step 3), the model includes a diffusion module and a denoising module; The diffusion module is used to gradually add noise to the input building foundation pit image until it becomes random noise, simulating the data degradation process; The denoising module is used to train the model so that it can recover the original foundation pit structure information from the foundation pit image data that is disturbed by noise, predict the noise distribution and gradually restore the clear image, and output the structure point information.

[0008] According to the above scheme, in step 3), the basic model used in the second-stage model is the stable diffusion model.

[0009] According to the above scheme, in step 3), a control network (controlnet) is used to fine-tune the stable diffusion model; specifically as follows: Lock the original model structure and parameters of the stable diffusion model, and copy the model structure and parameters to a trainable copy of ControlNet; A connection is established between the trainable copy of ControlNet and the stable diffusion model through the Zero Convolution layer; Train ControlNet to fine-tune the stable diffusion model.

[0010] According to the above scheme, in step 3), the stable diffusion model is fine-tuned using a control network (controlnet). During the fine-tuning process, only the weights of the Zero Convolution layer and the parameters of the trainable copy of the ControlNet are updated.

[0011] According to the above scheme, in step 5), verifying the accuracy of the model output results by changing the model's input conditions includes: Basic input validation uses the tiled image information and structural point information of the building foundation pit as input conditions to validate the model; it also validates the generated images of the support scheme within the building foundation pit in the test set. Extended input validation: Using images of building foundation pits (flat tiles), corner bracing of building foundation pits, and structural point information of building foundation pits as inputs to the model, the generated images of the support schemes within the building foundation pits in the validation test set are validated. To verify the validity of the image information, the image information of the building foundation pit was changed and used as the input of the model. The image information of the building foundation pit included the building size and shape. The verification was conducted to see if the model understood the physical meaning expressed by the image.

[0012] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above method.

[0013] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0014] The beneficial effects of this invention are: 1. This invention introduces a single-modal two-stage model and adopts a two-stage generation strategy. First, structural node information is generated from the foundation pit outline, and then a complete image of the foundation pit internal support scheme is generated from the constraints of the structural node information. This allows for the rapid acquisition of feasible design schemes for designers to refer to. Compared with the traditional design method of foundation pit internal support scheme that relies on engineers' experience and professional knowledge and involves manual calculation, reference to specifications, and previous engineering cases, this invention greatly improves design efficiency and reduces project costs. Attached Figure Description

[0015] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a flowchart of a method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the two-stage model network structure according to an embodiment of the present invention; Figure 4 This is a schematic diagram of a single-modal two-stage model generating a building foundation pit support scheme under flexible input conditions, according to an embodiment of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0017] like Figure 1 As shown, a method for generating an in-situ support scheme for a foundation pit based on a single-modal two-stage model includes the following steps: 1) Based on the drawings of the actual building foundation pit support scheme, obtain the information data of the building foundation pit support scheme; the information data includes building foundation pit pictures and foundation pit structural point information; the building foundation pit pictures include building foundation pit corner bracing pictures and building foundation pit flat tiling pictures; 2) Preprocess the information data on the support scheme in the building foundation pit, and construct an image dataset based on the preprocessed information data, which is divided into a training set and a test set; Preprocessing involves converting unstructured CAD drawings into single-modal images and standardized structural point data to meet the needs of subsequent model training. 3) Construct a two-stage model; 3.1) First stage: Generating structural point information; Using images of the foundation pit corner bracing and the foundation pit flat surface as input, structural point information is generated; in the first stage, the model is fine-tuned to control the generated model results. 3.2) Second stage: Generation of scheme diagrams based on structural points; Using the structural points output in the first stage as constraints, the model generation logic is adjusted to output images of the foundation pit support scheme that meet the structural point constraints. The second-stage model uses the Stable Diffusion model as its base model, taking images of the building foundation pit as input and the support scheme diagram inside the foundation pit as the generated image; the model includes two parts: diffusion and denoising. In the diffusion part of the model, noise is gradually added to the data, making the original data increasingly unclear until it becomes random noise. This simulates the gradual degradation of data on the support scheme within the foundation pit under the influence of natural or human factors. Through this process, the model can generate a series of data samples from the original data to a completely noisy state, which are then used to train the denoising model.

[0018] In the denoising part of the model, the model is trained to recover the original foundation pit structure information from noisy data and generate the support scheme within the foundation pit. The model first uses the noise data from the previous step and the conditional input information to predict the noise distribution information in the diffusion step, thereby obtaining the conditional probability of the denoising process. Finally, it gradually generates clearer images until the final scheme image is output.

[0019] 4) Model training; The model is trained using the Dropout strategy to ensure that it can generalize to various types of input information, thereby completing the construction and training of an automatic generation model for building foundation pit support schemes based on a single-modal two-stage model.

[0020] 5) Model validation; The accuracy of the model's output results was verified by changing different input conditions.

[0021] include: Basic input validation uses the tiled image information and structural point information of the building foundation pit as input conditions to validate the model; it also validates the generated images of the support scheme within the building foundation pit in the test set. Extended input validation: Using images of building foundation pits laid flat, images of building foundation pit corner bracing (opposite bracing, circular bracing), and structural point information of building foundation pits as input, validate the generated images of building foundation pit support schemes in the test set; Furthermore, by changing the information of the building foundation pit image as the model input, such as different building sizes and shapes, we can verify whether the model can understand the physical meaning expressed by the image.

[0022] 6) Use the validated model to generate images of the internal support scheme for the foundation pit, and obtain the internal support scheme for the foundation pit.

[0023] like Figure 2 A method for generating in-situ support schemes for foundation pits based on a single-modal two-stage model includes the following steps: 1) Based on the CAD drawings of the actual building foundation pit support scheme, obtain the information data of the building foundation pit support scheme, including building foundation pit pictures and foundation pit structural point information; the building foundation pit pictures include building foundation pit corner bracing pictures and building foundation pit flat pictures; By converting unstructured CAD drawings into single-modal images and standardized structural point data, it can be adapted to the needs of subsequent model training. 2) Construct an image dataset based on the preprocessed information data, and divide it into training and test sets according to the proportions; 3) Construct a two-stage model; 3.1) First stage: Generating structural point information; Using images of the foundation pit corner bracing and the foundation pit flat tiling as input, the model is fine-tuned to control the generation result of the model and generate key structural point information. 3.2) Second stage: Generation of scheme diagrams based on structural points; Using the structural points output in the first stage as constraints, the model generation logic is adjusted to output images of the foundation pit support scheme that meet the structural point constraints. The second-stage model uses Stable Diffusion as its base model, taking images of the building foundation pit as input and a diagram of the support scheme within the foundation pit as the generated image. The model comprises two parts: diffusion and denoising. Figure 3 .

[0024] In the diffusion step, noise is gradually added to the data, making the original data increasingly unclear until it becomes random noise. This simulates the gradual degradation of data related to the support scheme within the foundation pit under the influence of natural or human factors. Through this process, the model can generate a series of data samples from the original data to a completely noisy state, which are then used to train the denoising model. Noise is gradually added to the original foundation pit image X0 to generate X1, X2, ... X... t …, X T Each step involves a transition probability q(X). t |X t-1 This probability describes how a state transitions from one state to the next, noisier state. By progressively adding Gaussian noise that matches the transition probability, the degradation process of data from a clear scheme diagram to a completely random noise diagram is simulated.

[0025] In the denoising step, the model is trained to recover the original foundation pit structure information from noisy data and generate the internal support scheme. The model first uses the noise data from the previous step and the conditional input information to predict the noise distribution information in the diffusion step, thereby obtaining the conditional probability of the denoising process. Finally, it gradually generates clearer images until the final scheme image is output. Figure 3 As shown, the denoising step uses a completely random noise map X. T Starting with the U-Net architecture, T is iteratively executed. One denoising operation: First, for X... T Perform preliminary denoising to generate an intermediate noise map X. 1000 ; iteratively denoising to obtain X 200 X 100 X 50 X 10 Each step is based on conditional probability p θ (x) t 1 | x t Predict the noise distribution and reverse the image details; finally, stitch together the feature information of all the denoising processes to output a clear diagram of the foundation pit support scheme X0 that is consistent with the initial input.

[0026] In this embodiment, a control network (controlnet) is used to fine-tune the Stable Diffusion model; the details are as follows: Parameter migration involves locking the original model structure and parameters of the SD model and copying them to a trainable copy of ControlNet. For the input of building foundation pit images and foundation pit structural point information, the input of building foundation pit images (corner bracing and flat images) is as follows: the features are extracted by the image encoder of the SD model and then input into the convolutional layer of ControlNet for feature mapping; for structural point information, the structured coordinates are transformed into image-like features (e.g., heat map, with structural point positions marked as high-brightness areas) and then input into ControlNet to achieve image adaptation of structured constraints. The connection mechanism establishes a connection between the ControlNet trainable copy and the original SD model through the ZeroConvolution layer. The ZeroConvolution connects the copied model structure, and combined with the CLIP model and cross attention, the model can use text and images as input conditions. The model is fine-tuned so that it can simultaneously accept images of building foundation pit corner bracing (opposite bracing, circular bracing), images of the drilling outline inside the building foundation pit, or images presented by sketches, and finally generate an intuitive and accurate support scheme design drawing. During initialization, the weights of Zero Convolution are 0. At this time, ControlNet does not interfere with the generation of SD, ensuring that SD is still generated according to the original logic in the early stages of fine-tuning. During fine-tuning, only the weights of Zero Convolution and the parameters of the trainable copy of ControlNet are updated. Gradients are only propagated between the trainable copy and the Zero Convolution path, while the parameters of the original SD model remain unchanged.

[0027] The first stage of fine-tuning involves training only the trainable copy of ControlNet and Zero Convolution, with the input being the pit image and the output being the structure points, allowing the model to learn the constraints of mapping from the pit image to the structure points. The second stage of fine-tuning involves only fine-tuning the denoising module of SD. The input is a structural point feature map, and the output is a complete scheme map, allowing the model to learn the logic of generating scheme maps based on structural points.

[0028] 4) Model training; The model is trained using the Dropout strategy to ensure that it can generalize to various types of input information, thereby completing the construction and training of an automatic generation model for building foundation pit support schemes based on a single-modal two-stage model.

[0029] 5) Model validation; The accuracy of the model's output results was verified by changing different input conditions.

[0030] include: Basic input validation uses the tiled image information and structural point information of the building foundation pit as input conditions to validate the model; it also validates the generated images of the support scheme within the building foundation pit in the test set. Extended input validation: Using images of building foundation pits laid flat, images of building foundation pit corner bracing (opposite bracing, circular bracing), and structural point information of building foundation pits as input, validate the generated images of building foundation pit support schemes in the test set; Furthermore, the model input was changed to different images of building foundation pits, such as different building sizes and shapes, to verify whether the model could understand the physical meaning expressed in the images.

[0031] 6) Use the validated model to generate in-situ support schemes for the foundation pit, such as... Figure 4 .

[0032] Figure 4 In this embodiment, for the target irregular polygonal foundation pit, a combination of three types of input variables is used to achieve accurate generation of the solution: INPUT1 - Text Description: Clearly define the core physical parameters of the foundation pit, including area (20059㎡), length (169.7m), width (118.2m), and structural attributes such as internal perforations. The text description (INPUT 1) can be converted into structured parameters and used as numerical constraints for the model. INPUT2 - Outer Contour Image: Input a vector image of the actual outer contour of the foundation pit, enabling the model to directly identify the geometric features of the foundation pit; INPUT3 - Drill Outline Image: Input the specific shape of the internal drill (including composite structures such as rings and circles) to ensure that the support scheme avoids the drill area.

[0033] When only INPUT 1 and INPUT 2 are entered, the generated scheme corresponds to the support structure that covers the inner circular area in the generated scheme diagram; after adding INPUT 3, the support structure will maintain a safe distance from the drilling area inside the foundation pit.

[0034] It is evident that the model proposed in this application can combine text and images as input conditions, and can understand the physical meaning expressed by the input information, thereby quickly generating a solution that meets the requirements.

[0035] This application also provides a computer-readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, server, App application store, etc., which stores a computer program. When the program is executed by a processor, it implements the method for generating a foundation pit support scheme based on a single-modal two-stage model in the method embodiment.

[0036] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for generating an in-situ support scheme for a foundation pit based on a single-modal two-stage model, characterized in that, Includes the following steps: 1) Based on the drawings of the actual building foundation pit support scheme, obtain the information data of the building foundation pit support scheme; the information data includes building foundation pit pictures and foundation pit structural point information; The images of the building foundation pit include images of the corner bracing of the building foundation pit and images of the building foundation pit laid flat. 2) Preprocess the information data on the support scheme in the building foundation pit, and construct an image dataset based on the preprocessed information data, which is divided into a training set and a test set; 3) Construct a two-stage model; The two-stage model uses images of the building foundation pit as input and a diagram of the support scheme within the foundation pit as the generated image. The specific construction of the two-stage model is as follows: 3.1) First stage: Using the images of the corner bracing and the flat surface of the foundation pit as input, generate structural point information; 3.2) Second stage: Using the structural points generated in the first stage as constraints, output images of the internal support schemes for the foundation pit that meet the structural point constraints; 4) Use the training set for model training; 5) Use the test set to validate the model; 6) Use the validated model to generate images of the support scheme within the foundation pit.

2. The method for generating an internal support scheme for a foundation pit based on a single-modal two-stage model according to claim 1, characterized in that, In step 2), the preprocessing involves converting unstructured drawings into single-modal images and standardized structural point data.

3. The method for generating an internal support scheme for a foundation pit based on a single-modal two-stage model according to claim 1, characterized in that, In step 3), the model includes a diffusion module and a denoising module; The diffusion module is used to gradually add noise to the input building foundation pit image until it becomes random noise, simulating the data degradation process; The denoising module is used to train the model so that it can recover the original foundation pit structure information from the foundation pit image data that is disturbed by noise, predict the noise distribution and gradually restore the clear image, and output the structure point information.

4. The method for generating an in-situ support scheme for a foundation pit based on a single-modal two-stage model according to claim 1, characterized in that, In step 3), the basic model used in the second-stage model is the stable diffusion model.

5. The method for generating an internal support scheme for a foundation pit based on a single-modal two-stage model according to claim 1, characterized in that, In step 3), the stable diffusion model is fine-tuned using a control network (controlnet).

6. The method for generating an internal support scheme for a foundation pit based on a single-modal two-stage model according to claim 5, characterized in that, In step 3), a control network (controlnet) is used to fine-tune the stable diffusion model; specifically as follows: Lock the original model structure and parameters of the stable diffusion model, and copy the model structure and parameters to a trainable copy of ControlNet; A connection is established between the trainable copy of ControlNet and the stable diffusion model through the Zero Convolution layer; Train ControlNet to fine-tune the stable diffusion model.

7. The method for generating an internal support scheme for a foundation pit based on a single-modal two-stage model according to claim 6, characterized in that, In step 3), the stable diffusion model is fine-tuned using a control network (controlnet); during the fine-tuning process, only the weights of the Zero Convolution layer and the parameters of the trainable copy of the ControlNet are updated.

8. The method for generating an internal support scheme for a foundation pit based on a single-modal two-stage model according to claim 1, characterized in that, In step 5), the accuracy of the model output is verified by changing the model's input conditions, including: Basic input validation uses the tiled image information and structural point information of the building foundation pit as input conditions to validate the model; it also validates the generated images of the support scheme within the building foundation pit in the test set. Extended input validation: Using images of building foundation pits (flat tiles), corner bracing of building foundation pits, and structural point information of building foundation pits as inputs to the model, the generated images of the support schemes within the building foundation pits in the validation test set are validated. To verify the validity of the image information, the image information of the building foundation pit was changed and used as the input of the model. The image information of the building foundation pit included the building size and shape. The verification was conducted to see if the model understood the physical meaning expressed by the image.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 8.

Images