AI maintenance frame building method and system capable of automatically generating maintenance construction drawings

Abstract

The invention belongs to the technical field of road maintenance engineering data processing, and discloses an AI maintenance frame building method and system capable of automatically generating a maintenance construction drawing. According to the method, a comprehensive visual decision support system is constructed, road network result information is obtained, and maintenance measure expert database construction standards are constructed according to actual requirements of different regions; performing priority ranking on the maintenance paragraphs according to target requirements, and completing optimization of maintenance countermeasures; realizing automatic generation of a maintenance design scheme based on an RAG application; and cross-platform intelligent interaction between the natural language and maintenance design scheme software is realized by utilizing a CAD software automation function of a large language model. According to the method, the first draft of the maintenance design report is finally and automatically generated, intelligentization and automation of highway maintenance design, decision making and implementation processes are realized, and the efficiency, accuracy and economical efficiency of the whole process can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of highway maintenance engineering data processing, and particularly relates to a method and system for building an AI maintenance framework for automatically generating maintenance construction drawings. Background Art

[0002] In recent years, with the completion of the highway road network, the tight use situation of transportation channels has been significantly improved, and the transportation capacity has been greatly enhanced. However, large-scale construction will inevitably bring heavy maintenance tasks, and highway maintenance has transitioned from the traditional "emergency repair era" to the "comprehensive maintenance era".

[0003] Highway maintenance engineering involves links such as inspection, design, construction, and post-evaluation. Highway maintenance design is the work of conducting survey and measurement on existing roads, and re-maintaining and designing old roads through content such as investigation and evaluation, disease diagnosis and maintenance countermeasure selection, technical design, and construction drawing design to restore their good service functions. As the first threshold for controlling construction quality, it is particularly important in each link. In other words, the quality of maintenance design documents often determines the quality of construction. For large-mileage maintenance design projects, compiling a large number of construction drawing plans is labor-intensive and time-consuming, and the current overall design efficiency is low.

[0004] For large-mileage maintenance design projects, compiling construction design drawings is labor-intensive and time-consuming. Considering factors such as the current low efficiency, strong subjectivity, and deviation in maintenance plan formulation of maintenance design, it is necessary to propose a method for building an AI maintenance design framework system based on multi-source data and large language models. Summary of the Invention

[0005] To overcome the problems existing in the related technologies, the disclosed embodiments of the present invention provide a method and system for building an AI maintenance framework for automatically generating maintenance construction drawings.

[0006] The technical solutions are as follows: A method for building an AI maintenance framework for automatically generating maintenance construction drawings includes the following steps:

[0007] S1, using the integrated application of high-resolution remote sensing technology and geographic information system GIS to construct a comprehensive visual decision support system, obtaining highway road network result information, and providing data support for maintenance design based on spatial data;

[0008] S2, according to the comprehensive visual decision support system, constructing the construction standard of a maintenance measure expert database for the actual needs of different regions;

[0009] S3, based on the constructed construction standard of the maintenance measure expert database, prioritizing maintenance sections according to the target requirements to complete the optimization of maintenance countermeasures;

[0010] S4. Based on the optimized results of the maintenance countermeasures, realize the automatic generation of the maintenance design plan based on the RAG application;

[0011] S5. According to the generated maintenance design plan, utilize the automatic function of the CAD software of the large language model to realize the cross-platform intelligent interaction between natural language and the maintenance design plan software, and complete the full-chain intelligent business closed-loop from pavement disease detection to maintenance design and then to the implementation plan.

[0012] In step S1, use the integrated application of high-resolution remote sensing technology and geographic information system GIS to construct a comprehensive visual decision support system, including:

[0013] Utilize high-resolution remote sensing images to obtain information on the earth's surface through sensors on satellites or high-altitude aircraft, and transmit it to the ground station via the data center; the information on the earth's surface includes traffic, meteorology, and geology;

[0014] Use image processing software for preprocessing such as radiometric correction and geometric correction; then extract information through image enhancement and classification recognition to generate an image map for analysis and application;

[0015] By overlaying the information of multiple thematic layers of traffic, meteorology, and geology, construct a comprehensive visual decision support system to provide data support for maintenance design based on spatial data.

[0016] Furthermore, the overlay of the information of multiple thematic layers of traffic, meteorology, and geology includes:

[0017] The first step is to divide the image of a certain thematic layer among traffic, meteorology, and geology into regions. The divided region image contains complete image information superposition units of a rows and b columns, or incomplete image information non-superposition units;

[0018] The second step is to identify the four edge node coordinates X1, X2, X3, X4, the total width Y, and the total height H of the a-row and b-column image information superposition unit on the map; identify the distances Y L 、Y R 、Y U 、Y D , and obtain the spacing h between two rows of image information superposition sub-pixels and the spacing y between two columns of image information superposition sub-pixels. Save the Y L 、Y R 、Y U 、Y D parameters locally, and extract the region divided by points X1, X2, X3, X4 from the map and save it locally in the form of a picture with a size of W×H;

[0019] Step 3: By calculating the relationships between X1, X2, X3, X4 and Y L 、Y R 、Y U 、Y D to dynamically obtain traffic, meteorological, and geological images acquired by sensors on satellites or high-altitude aircraft, so as to meet the requirements of image preprocessing during the next photo-taking; where Y L 、Y R 、Y U 、Y D The identification process of the parameters includes:

[0020] (1) Perform canny operator filtering on the first image of a certain thematic layer image after preprocessing to obtain a filtered image;

[0021] (2) Using the elastic ball model theory, analyze the filtered image with the starting point (0, 0) to obtain two sets of boundary line point sets; the image information superposition units within the picture are a×b, and based on the analysis of these two sets of boundary line point sets, b + 1 points are obtained. Then, starting from these b + 1 points respectively, use the elastic ball model theory to analyze and obtain (b + 1)×2 sets of boundary line point sets;

[0022] (3) Using the (b + 1)×2 sets of boundary line point sets, perform Hough line detection with a negative feedback mechanism to obtain (b + 1)×2 straight lines. Sort the (b + 1)×2 straight lines according to the magnitude of the X-axis coordinates;

[0023] (4) Analyze any two sets of related point sets in the (b + 1)×2 sets of boundary line point sets, calculate a + 1 points, and starting from these a + 1 points respectively, use the elastic ball model theory to analyze and obtain (a + 1)×2 sets of boundary line point sets;

[0024] (5) Using the (a + 1)×2 sets of boundary line point sets, perform Hough line detection with a negative feedback mechanism to obtain (a + 1)×2 straight lines. Sort the (a + 1)×2 straight lines according to the magnitude of the Y-axis coordinates;

[0025] (6) After obtaining (b + 1)×2 vertical straight lines and (b + 1)×2 horizontal straight lines, calculate W L 、W R 、W U 、W D parameters;

[0026] Step 4: After all thematic layer images are saved, perform non-negative matrix fusion on the traffic thematic layer source image F1, meteorological thematic layer source image F2, and geological thematic layer source image F3 with a size of B×A to be superimposed to generate a temporary superimposed image C;

[0027] Construct the observation matrix P of the source images F1, F2, and F3 to be superimposed i , including storing the elements of each observation image row by row into a column vector to obtain the observation matrix P i ; Decompose the observation matrix P i using the non - negative matrix factorization algorithm to obtain the basis matrix L;

[0028]

[0029] In the formula, L ia is the basis matrix of the a - th row in the i source images to be superimposed, P i is the observation matrix of the i source images to be superimposed, (LH) i is the value of the basis matrix after the interval distance of the i source images to be superimposed, H a is the interval distance of the a - th row, L ja is the basis matrix of the a - th row in the j - th source image to be superimposed, i is the i source images to be superimposed, and j is the j - th source image to be superimposed;

[0030] Convert the basis matrix L into a matrix of size B×A, and the corresponding image of this matrix is the temporary superimposed image C;

[0031] Solve the multi - direction gradient fields of the source images F1, F2, F3 and the temporary superimposed image C to obtain the image gradient field diagram; Compare the gradient values at the same positions of the four image gradient fields, and select the image with the largest gradient value as the superimposed image selection value to obtain the decision matrix diagram of the visualization decision support system.

[0032] In step S2, construct the construction standards for the expert library of maintenance measures, including: By analyzing the differences in geographical divisions, meteorological hydrology, geological conditions, standards, and material price factors in different regions, using the route information as the main data, merging the detection data, highway maintenance historical data, highway grades, regional construction technology levels, material indicators, and regional material quota library maintenance basic materials, and constructing according to the actual needs of different regions.

[0033] In step S3, based on the constructed construction standards for the expert library of maintenance measures, prioritize the maintenance sections according to the target requirements to complete the optimization of maintenance countermeasures, including:

[0034] By accessing the pavement and traffic safety maintenance project cost quotas of the region, dividing the maintenance units, inputting key limiting conditions to adjust the vertical slope and elevation - limited section schemes of convex curves, then merging the scattered and discontinuous maintenance scheme sections and adding refined designs for the smooth slopes of level crossings. Finally, through estimating the project cost analysis, prioritize the maintenance sections according to the target requirements, and this ranking is based on the indicators of cost - performance priority, quality priority, and efficiency priority to achieve the optimized design of maintenance countermeasures.

[0035] Furthermore, the cost performance includes: benefit-cost ratio, benefit-energy consumption ratio, and benefit-carbon emission ratio;

[0036] The priority ranking includes:

[0037] (a) Obtain the cost performance f, the capital constraint vector g, the quality u, and the efficiency o indicators;

[0038] (b) Unify the dimensions of the measurement indicators, and use the normalization method to normalize the above indicator values to between 0 and 1 to obtain the normalized dimension values D of the above indicators;

[0039] (c) Use the analytic hierarchy process to define the within-layer importance weights for the above indicators, and obtain the corresponding weights of the above factors, including the cost performance weight Z f , the capital constraint vector weight Z g , the quality weight Z u , the efficiency weight Z o ;

[0040] (d) Calculate the comprehensive evaluation value of each node within each layer according to the above within-layer importance weights as:

[0041] J iQ = D f Z f + D g Z g + D u Z u + D o Z o

[0042] In the formula, J iQ is the comprehensive evaluation value, D f is the normalized dimension value under the cost performance constraint, Z f is the cost performance weight, D g is the normalized dimension value under the capital constraint vector, Z g is the capital constraint vector weight, D u is the normalized dimension value under the quality constraint, Z u is the quality weight, D o is the normalized dimension value under the efficiency constraint, Z o is the efficiency weight;

[0043] (e) Use the analytic hierarchy process to calculate the influence weights Zl of each layer respectively;

[0044] (f) Calculate the multi-layer structure evaluation value of the node maintenance countermeasure system;

[0045] (g) Rank the node importance from large to small according to the multi-layer structure evaluation value of the node maintenance countermeasure system.

[0046] In step (e), the analytic hierarchy process is used to calculate the influence weight Zl of each layer, including:

[0047] Step 1: Establish a judgment matrix B, compare evaluation indicators pairwise, and form a judgment matrix B with the initial weights. The elements in the judgment matrix B represent the scale coefficients obtained after comparing the indicators;

[0048] Step 2: Calculate the maximum eigenvalue and the corresponding eigenvector for each pairwise comparison judgment matrix, and perform consistency tests using the consistency index, random consistency index, and consistency ratio;

[0049] Step 3: The values of the normalized eigenvector {Z1, Z2, Z3, Z4} are the corresponding weights of the four factors. The weight of cost performance is Z f , the weight of the capital constraint vector is Z g , the weight of quality is Z u , the weight of efficiency is Z o , and the level is l;

[0050] Z1 = Dl f Z f + Dl g Z g + Dl u Z u + Dl o Z o

[0051] In the formula, D is the normalized dimension value, l f is the influence value under the cost performance constraint, l g is the influence value under the capital constraint vector, l u is the influence value under the quality constraint, l o is the influence value under the efficiency constraint;

[0052] In step (f), calculate the multi-layer structure evaluation value of the maintenance countermeasure system for the node, including:

[0053] (f - 1) First, calculate the product of the comprehensive evaluation value of the node at level Q and the influence weight of this layer, that is, the single-layer evaluation value;

[0054] (f - 2) The multi-layer network evaluation value of the node is the sum of the single-layer evaluation values.

[0055] In step S4, according to the optimization results of the maintenance countermeasures, realize the automatic generation of the maintenance design plan based on the RAG application, including:

[0056] S401, the knowledge base adopts a hierarchical structure, divides the information into multiple levels, including the file level, chapter level, and paragraph level. Each level contains information labels for locating the required knowledge;

[0057] S402. Design an index structure with an n-ary tree as the index, including establishing keyword indexes, entity indexes, and semantic indexes.

[0058] S403. Set up a semi-automated update process. By classifying new texts based on simple characters and text semantics and combining manual review, evaluate and verify newly added or modified knowledge points. At the same time, for existing knowledge entries, establish a feedback mechanism to improve the content of the knowledge base by using user-reported errors or improvement suggestions.

[0059] S404. For sensitive or private data included in the knowledge base, take data security measures, including md5 or rsa data encryption and user authentication access control, to ensure the implementation of the RAG system in the knowledge base.

[0060] In step S5, implement cross-platform intelligent interaction between natural language and the maintenance design plan software, including:

[0061] Describe the task to be executed through natural language. The AI will structure the instruction task, form a script or programming language, and automatically input it into the desktop design software. The desktop design software will execute various types of drawing tasks by executing the script and programming language.

[0062] The execution plan includes: The detection device will collect and process road data in real time, automatically fill in key content through a prefabricated report template to generate a disease report; Automatically generate a maintenance design plan based on the text content generated by the large language model, and realize the automation of drawing design through the interaction between natural language and the desktop design software; Implement road maintenance according to the automatically generated design plan.

[0063] Another object of the present invention is to provide an AI maintenance framework construction system for automatically generating maintenance construction drawings. This system implements the AI maintenance framework construction method for automatically generating maintenance construction drawings. The system includes:

[0064] A comprehensive visual decision support system construction module for using the integrated application of high-resolution remote sensing technology and geographic information system GIS to construct a comprehensive visual decision support system, obtain highway road network result information, and provide data support for maintenance design based on spatial data.

[0065] A maintenance measure expert library construction module for constructing the construction standard of a maintenance measure expert library according to the comprehensive visual decision support system and the actual needs of different regions.

[0066] A maintenance countermeasure optimization module for prioritizing maintenance paragraphs according to the target requirements based on the constructed construction standard of the maintenance measure expert library to complete the optimization of maintenance countermeasures.

[0067] A maintenance design plan automatic generation solution is used to automatically generate a maintenance design plan based on the optimization results of maintenance countermeasures and the RAG application;

[0068] A road surface disease detection module is used to realize cross-platform intelligent interaction between natural language and maintenance design plan software according to the generated maintenance design plan by using the CAD software automation function of the large language model, and complete the full-chain intelligent business closed-loop from road surface disease detection to maintenance design and then to the implementation plan.

[0069] Combining all the above technical solutions, the beneficial effects of the present invention are as follows: By integrating spatial information data, road condition detection equipment data, and artificial intelligence technology, the present invention aims to realize the intelligence of the entire process of highway maintenance. Its significance lies in enhancing the decision-making support ability, providing rich background information through visualization analysis tools to assist in more accurate decision-making; improving data processing efficiency, automatically processing data from different sources and ensuring data accuracy, reducing the time and resource consumption of manual review; optimizing the maintenance plan design, automatically outputting a preliminary maintenance plan according to the disease type and degree, and estimating the investment cost to help decision-makers evaluate the cost-benefit ratio; on the basis of the previous stage, with the help of the large language model, realizing the automatic generation of the maintenance design plan, and the automatic generation system of the road surface maintenance plan based on the RAG application recommends the best plan according to the economic and social benefit analysis, greatly improving the design automation level. Using methods such as large language model prompts and training fine-tuning to generate scripts to operate third-party design software to automatically generate maintenance design drawings, and combining high-resolution technology and GIS functions to generate distribution maps of road construction materials along the line and construction detour plans, and finally automatically generating the initial draft of the maintenance design report, realizing the intelligence and automation of the highway maintenance design, decision-making, and implementation processes, which can improve the efficiency, accuracy, and economy of the entire process. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] The drawings here are incorporated into the specification and form a part of this specification, showing embodiments consistent with the present disclosure, and are used together with the specification to explain the principles of the present disclosure;

[0071] Figure 1 It is a flowchart of the method for building an AI maintenance framework for automatically generating maintenance construction drawings provided by an embodiment of the present invention;

[0072] Figure 2 It is a schematic diagram of the system for building an AI maintenance framework for automatically generating maintenance construction drawings provided by an embodiment of the present invention;

[0073] Figure 3 It is a schematic diagram of the principle of the method for building an AI maintenance framework for automatically generating maintenance construction drawings provided by an embodiment of the present invention;

[0074] Figure 4 It is the schematic diagram of obtaining highway road network result information provided by the embodiment of the present invention;

[0075] Figure 5 It is the flow chart of pavement condition calibration, plan formulation and optimization based on multi-source data provided by the embodiment of the present invention;

[0076] Figure 6 It is the schematic diagram of the maintenance design software based on AI technology and large language model provided by the embodiment of the present invention;

[0077] Figure 7 It is the comparison diagram of the prior art and the technical solution of the present invention provided by the embodiment of the present invention;

[0078] In the figure: 1. Comprehensive visual decision support system construction module; 2. Maintenance measure expert library construction module; 3. Maintenance countermeasure optimization module; 4. Automatic generation scheme of maintenance design plan; 5. Pavement disease detection module. Specific implementation manner

[0079] In order to make the above objects, features and advantages of the present invention more obvious and understandable, the specific implementation manner of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific implementations disclosed below.

[0080] Example 1, as Figure 1 shown, the method for building an AI maintenance framework for automatically generating maintenance construction drawings provided by the embodiment of the present invention includes:

[0081] S1. Using the integrated application of high-resolution remote sensing technology and geographic information system GIS, construct a comprehensive visual decision support system, obtain highway road network result information, and provide data support for maintenance design based on spatial data;

[0082] Exemplarily, mainly use high-resolution remote sensing images to obtain earth surface information through high-resolution sensors on satellites or high-altitude aircraft, and transmit it to the ground station via the data center. The earth surface information includes traffic, meteorology, and geology;

[0083] Subsequently, use professional image processing software for preprocessing such as radiometric correction and geometric correction, and then extract useful information through technologies such as image enhancement and classification recognition, and finally generate a high-quality image map for analysis and application.

[0084] This process is abbreviated as: Acquisition and Processing of High-Resolution Remote Sensing Images. By overlaying the information of various thematic layers such as transportation, meteorology, and geology, a comprehensive visual decision support system is constructed to provide more accurate and efficient data support for maintenance design based on spatial data.

[0085] Exemplarily, further, the overlay of the information of various thematic layers of transportation, meteorology, and geology includes:

[0086] First step, divide the area of an image of a certain thematic layer among the various thematic layers of transportation, meteorology, and geology. The divided area image contains complete image information superposition units of a rows and b columns, or incomplete image information non-superposition units;

[0087] Second step, identify the four edge node coordinates X1, X2, X3, X4 on the graph of the image information superposition unit of the a rows and b columns, the total width Y, and the total height H, and identify the distances Y L 、Y R 、Y U 、Y D , and obtain the spacing h between two rows of image information superposition sub-pixels and the spacing y between two columns of image information superposition sub-pixels. Save the Y L 、Y R 、Y U 、Y D parameters in accordance with certain rules locally, and extract the area divided by the points X1, X2, X3, X4 from the graph and save it locally in the form of a picture with a size of W*H;

[0088] Third step, by calculating the relationship between X1, X2, X3, X4 and Y L 、Y R 、Y U 、Y D , dynamically obtain high-resolution sensors on satellites or high-altitude aircraft to obtain transportation, meteorology, and geology images, so that the images meet the basic requirements of image preprocessing during the next photo-taking;

[0089] The Y L 、Y R 、Y U 、Y D parameter identification process includes:

[0090] (1) Perform canny operator filtering on the first image of a certain thematic layer image after preprocessing to obtain a filtered image;

[0091] (2) Using the pinball model theory, analyze the filtered image with the starting point (0, 0) to obtain two sets of boundary line point sets; the number of image information superposition units within the picture is a×b. Based on the analysis of these two sets of boundary line point sets, b + 1 points are obtained. Then, starting from these b + 1 points respectively, using the pinball model theory for analysis, (b + 1)×2 sets of boundary line point sets are obtained;

[0092] (3) Using the (b + 1)×2 sets of boundary line point sets, perform Hough line detection with a negative feedback mechanism to obtain (b + 1)×2 straight lines. Sort the (b + 1)×2 straight lines according to the magnitude of the X-axis coordinates;

[0093] (4) Analyze any two sets of relevant point sets among the (b + 1)×2 sets of boundary line point sets, calculate a + 1 points. Starting from these a + 1 points respectively, use the pinball model theory for analysis to obtain (a + 1)×2 sets of boundary line point sets;

[0094] (5) Using the (a + 1)×2 sets of boundary line point sets, perform Hough line detection with a negative feedback mechanism to obtain (a + 1)×2 straight lines. Sort the (a + 1)×2 straight lines according to the magnitude of the Y-axis coordinates;

[0095] (6) After obtaining (b + 1)×2 vertical straight lines and (b + 1)×2 horizontal straight lines, calculate W L 、W R 、W U 、W D parameters;

[0096] In the fourth step, after all thematic layer images are saved, perform non - negative matrix fusion on the traffic thematic layer source image F1, meteorological thematic layer source image F2, and geological thematic layer source image F3 with a size of B×A to be superimposed, generating a temporary superimposed image C;

[0097] Construct the observation matrix P i of the source images F1, F2, and F3 to be superimposed, including storing the elements of each observation image row by row into a column vector to obtain the observation matrix P i ; Use the non - negative matrix factorization algorithm to decompose the observation matrix P i to obtain the basis matrix L;

[0098]

[0099] In the formula, L ia is the basis matrix of the a - th row in the i source images to be superimposed, P i is the observation matrix of the i source images to be superimposed, and (LH) i is the basis matrix value after the interval distance of the i source images to be superimposed, Ha is the spacing distance for the a-th row, L ja is the base matrix for the a-th row in the j-th source image to be superimposed, i is the i source images to be superimposed, and j is the j-th source image to be superimposed;

[0100] Convert the base matrix L into a matrix of size B×A, and the image corresponding to this matrix is the temporary superimposed image C;

[0101] Perform multi-directional gradient field solution on the source images F1, F2, F3 and the temporary superimposed image C to obtain the image gradient field map; compare the gradient values at the same positions of the four image gradient fields, and select the image with the largest gradient value as the superimposed image selection value to obtain the decision matrix map of the visualization decision support system.

[0102] S2. According to the comprehensive visualization decision support system, construct the construction standard of the maintenance measure expert library according to the actual needs of different regions;

[0103] Specifically include: by analyzing the differences in geographical division, meteorological hydrology, geological conditions, standards and material prices in different regions, using the route information as the main data, merging the maintenance basic data of each provincial and municipal highway bureau such as detection data, highway maintenance history data, highway grade, regional construction technology level, main material indicators and regional material quota library, and formulating scientific and reasonable construction standards for the maintenance measure expert library according to the actual needs of different regions.

[0104] S3. Based on the constructed construction standard of the maintenance measure expert library, perform priority ranking on the maintenance sections according to the target requirements to complete the optimization of the maintenance countermeasures;

[0105] Through the access to the maintenance project cost quotas of the road surface and traffic safety in the area, divide the maintenance units, input the key limiting conditions to adjust the vertical slope and elevation limited section scheme of the convex curve, and then perform scheme merging on the scattered and discontinuous maintenance scheme sections and carry out refined designs such as adding smooth slopes at level crossings. Finally, through the estimated cost analysis, perform priority ranking on the maintenance sections according to the target requirements. This ranking can be sorted according to indicators such as cost-performance priority, quality priority, and efficiency priority to achieve the optimized design of the maintenance countermeasures.

[0106] Exemplarily, the cost-performance includes benefit-cost ratio, benefit-energy consumption ratio, and benefit-carbon emission ratio;

[0107] The priority ranking includes:

[0108] (a) Obtain the cost-performance f, the capital restriction constraint vector g, the quality u, and the efficiency o indicators;

[0109] (b) Unify the dimension of the measurement indicators, and use the normalization method to normalize the above indicator values between 0 and 1 to obtain the normalized dimension values D of the above indicators;

[0110] (c) Use the Analytic Hierarchy Process to define the within-layer importance weights for the above indicators, and obtain the corresponding weights for the above factors, including the cost performance weight Z f , the fund restriction constraint vector weight Z g , the quality weight Z u , the efficiency weight Z o ;

[0111] (d) Calculate the comprehensive evaluation value of each node within each layer according to the above within-layer importance weights as:

[0112] J iQ = D f Z f + D g Z g + D u Z u + D o Z o

[0113] In the formula, J iQ is the comprehensive evaluation value, D f is the normalized dimension value under the cost performance constraint, Z f is the cost performance weight, D g is the normalized dimension value under the fund restriction constraint vector, Z g is the fund restriction constraint vector weight, D u is the normalized dimension value under the quality constraint, Z u is the quality weight, D o is the normalized dimension value under the efficiency constraint, Z o is the efficiency weight;

[0114] (e) Use the Analytic Hierarchy Process to calculate the influence weights Zl of each layer respectively;

[0115] (f) Calculate the multi-layer structure evaluation value of the node maintenance countermeasure system;

[0116] (g) Sort the node importance from large to small according to the multi-layer structure evaluation value of the node maintenance countermeasure system.

[0117] In step (e), the influence weights Zl of each layer calculated by using the Analytic Hierarchy Process include:

[0118] Step 1: Establish a judgment matrix B, compare the evaluation indicators pairwise, and form the initial weights into the judgment matrix B. The elements in the judgment matrix B represent the scale coefficients obtained after the index comparison;

[0119] Step 2: Calculate the maximum eigenvalue and its corresponding eigenvector for each pairwise comparison judgment matrix, and perform consistency tests using the consistency index, random consistency index, and consistency ratio.

[0120] In step 3, the values of the normalized eigenvectors {Z1, Z2, Z3, Z4} are the corresponding weights of the four factors, where the weight of cost performance is Z f , the weight of the capital constraint vector is Z g , the weight of quality is Z u , the weight of efficiency is Z o , and the level is l;

[0121] Z1 = Dl f Z f + Dl g Z g + Dl u Z u + Dl o Z o

[0122] In the formula, D is the normalized dimension value, and l f is the influence value under the cost performance constraint, and l g is the influence value under the capital constraint vector, and l u is the influence value under the quality constraint, and l o is the influence value under the efficiency constraint;

[0123] In step (f), calculate the multi-layer structure evaluation value of the maintenance countermeasure system of the node, including:

[0124] (f-1) First, calculate the product of the comprehensive evaluation value of the node at level Q and the influence weight of this layer, that is, the single-layer evaluation value;

[0125] (f-2) The multi-layer network evaluation value of the node is the sum of the single-layer evaluation values.

[0126] S4. According to the optimization results of the maintenance countermeasures, realize the automatic generation of the maintenance design scheme based on the RAG application; including:

[0127] S401. The knowledge base adopts a hierarchical structure, divides the information into multiple levels, including the file level, chapter level, and paragraph level, and each level contains specific information tags to facilitate quick positioning of the required knowledge.

[0128] S402. Design a powerful index structure, such as an n-ary tree as the index, to improve the retrieval efficiency. It includes, but is not limited to, establishing multiple index methods such as keyword index, entity index, and semantic index to meet the query requirements in different scenarios.

[0129] S403. Update the strategy to keep the knowledge base dynamic. Establish a semi-automated update process. By classifying new texts based on simple characters and text semantics, combined with manual review, evaluate and verify the newly added or modified knowledge points to ensure the accuracy and authority of the information stored in the database. At the same time, for existing knowledge entries, a feedback mechanism should also be established to encourage users to report errors or suggest improvements, thereby continuously improving the content of the knowledge base.

[0130] S404. Given that the knowledge base may contain sensitive or privacy data, strict data security measures need to be taken, including md5 or rsa data encryption, user authentication access control, etc. In addition, relevant laws and regulations, such as GDPR, need to be complied with to protect personal privacy from infringement. Achieve an efficient, secure, and easily maintainable knowledge base to ensure the successful implementation of the RAG system. Its design needs to comprehensively consider various factors to meet the diverse needs in practical applications.

[0131] S5. According to the generated maintenance design plan, utilize the CAD software automation function of the large language model to achieve cross-platform intelligent interaction between natural language and the maintenance design plan software, and complete the full-chain intelligent business closed-loop from road disease detection to maintenance design and then to the implementation plan;

[0132] Exemplarily, users can interact with the AI through natural language commands, that is, the user describes the task to be executed in natural language, and the AI structures the instruction task, forms a script or programming language, and automatically inputs it into the desktop design software (maintenance design software based on AI technology and large language model). The desktop design software executes no less than 30 types of drawing tasks by executing the script and programming language. This innovative interaction method can not only improve the design efficiency but also significantly enhance the human-machine collaboration experience in the design process.

[0133] Exemplarily, in this closed-loop system, the implementation plan includes: the detection device will collect and process road data in real time, and automatically fill in the key content through a prefabricated report template to generate an accurate disease report; automatically generate a maintenance design plan through the text content generated by the large language model, and realize the automation of drawing design through the interaction between natural language and the design software; finally, the construction unit can efficiently implement road maintenance work according to the automatically generated design plan. In addition, the design unit conducts scheme comparison for the maintenance project paragraphs through the maintenance measure library, automatically exports CAD drawings, and exports various economic and technical indicators such as the utilization rate of old materials and carbon emissions, and realizes the one-key generation of construction drawings, thereby significantly reducing the time and economic costs of traditional maintenance design.

[0134] Example 2, as Figure 2 shown, the AI maintenance framework building system for automatically generating maintenance construction drawings provided by the embodiments of the present invention includes:

[0135] Comprehensive Visualization Decision Support System Construction Module 1, which is used to construct a comprehensive visualization decision support system by integrating the application of high-resolution remote sensing technology and Geographic Information System (GIS), obtain highway road network result information, and provide data support for maintenance design based on spatial data;

[0136] Maintenance Measure Expert Database Construction Module 2, which is used to construct the construction standard of the maintenance measure expert database according to the actual needs of different regions based on the comprehensive visualization decision support system;

[0137] Maintenance Countermeasure Optimization Module 3, which is used to prioritize the maintenance sections according to the target requirements based on the constructed construction standard of the maintenance measure expert database, and complete the optimization of the maintenance countermeasures;

[0138] Automated Generation Scheme of Maintenance Design Plan 4, which is used to automatically generate the maintenance design plan based on the RAG application according to the optimization result of the maintenance countermeasures;

[0139] Pavement Disease Detection Module 5, which is used to realize the cross-platform intelligent interaction between natural language and the maintenance design plan software according to the generated maintenance design plan by using the automated function of the CAD software of the large language model, and complete the full-chain intelligent business closed-loop from pavement disease detection to maintenance design and then to the implementation plan.

[0140] Exemplarily, as Figure 3 shown, the principle of the AI maintenance framework construction method for automatically generating maintenance construction drawings provided by the embodiments of the present invention.

[0141] Embodiment 3, as another implementation manner of the present invention. In the AI maintenance framework construction method for automatically generating maintenance construction drawings provided by the embodiments of the present invention, constructing the visualization decision support system includes:

[0142] First, high-resolution satellite images are used as the basic data source. Through the processing and analysis of these images, the road surface areas that need to be maintained are accurately identified, and multi-dimensional data such as traffic flow, climate conditions, and geological structures are integrated with the remote sensing images to form a dataset that comprehensively reflects the conditions around the road. For example, by analyzing the traffic flow data passing through a specific area, it can be predicted that certain sections may have suffered greater wear; combined with the geological thematic map, potential risk areas prone to geological disasters such as settlement and landslides can be identified; using the meteorological thematic map helps to evaluate the impact degree of extreme weather on road facilities.

[0143] Based on the above data, further develop a visualization system to assist decision-makers in conducting more intuitive and effective analysis. This tool can not only quickly calculate the general location of the pavement to be treated, but also clearly display the sections that are difficult to construct due to terrain limitations, and provide a reference basis for future maintenance design. For example, when planning a repair plan, decision-makers can intuitively understand the specific location of each maintenance unit, the characteristics of the surrounding environment, and potential risk factors through this tool, so as to formulate a more reasonable and feasible repair plan.

[0144] Exemplarily, obtain highway road network result information, such as Figure 4 shown, specifically including:

[0145] High-resolution image data source;

[0146] Preprocessing: Geometric correction, projection transformation, image fusion;

[0147] Sample making: Conduct manual vectorization, divide into positive samples and negative samples, and then perform sample slicing;

[0148] Road network detection: Train the WGAN model, optimize parameters, and perform manual post-processing;

[0149] Accuracy verification: Use the SVM results and manual results to verify the road network results.

[0150] Example 4, as another implementation manner of the present invention. The method for building an AI maintenance framework for automatically generating maintenance construction drawings provided by the embodiments of the present invention further includes:

[0151] Calibration, plan formulation and optimization of pavement conditions based on multi-source data, such as Figure 5 shown.

[0152] This process is through the pavement quality index (PQI), pavement condition index (SCI), and pavement structure strength index, etc. Obtain more detailed pavement material property information through traditional methods such as core drilling to further verify the accuracy of the detection results. Use geographic information system (GIS) technology and high-resolution remote sensing images, combined with historical maintenance records and environmental factors, to establish a comprehensive data model. This model can evaluate the current pavement condition and predict the development trend of possible diseases in the future. By taking a route as the main data and integrating historical maintenance data, this multi-dimensional data fusion method can effectively improve the accuracy and reliability of data analysis, and provide a scientific basis for subsequent maintenance strategies.

[0153] Secondly, formulate a reasonable maintenance plan. With the assistance of intelligent algorithms and expert systems, automatically classify and prioritize the processing of the most urgent problem areas. At the same time, by integrating information from different sources, optimize resource allocation to ensure the maximization of the economic and social benefits of the maintenance work. The finally formed maintenance plan not only includes specific maintenance and repair plans but also long-term monitoring and maintenance strategies to ensure that the entire road network is in a relatively high service state.

[0154] Exemplarily, such as Figure 5 , specifically including:

[0155] Collect information on major diseases, highway grades, and traffic volumes;

[0156] Conduct a preliminary selection of pavement maintenance plans;

[0157] Conduct evaluations of benefit-cost ratio, benefit-energy consumption ratio, and benefit-carbon emission ratio;

[0158] Obtain comprehensive evaluation indicators for pavement maintenance plans;

[0159] Obtain the best maintenance plan;

[0160] Based on the constraint condition of capital limitation, optimize the allocation of maintenance funds;

[0161] Obtain the optimal result based on 0-1 linear programming.

[0162] Example 5, as another implementation manner of the present invention. The method for building an AI maintenance framework for automatically generating maintenance construction drawings provided by the embodiments of the present invention is further based on an AI technology and a maintenance design software of a large language model, such as Figure 6 shown.

[0163] Evaluate and select a large language model, such as Transformer or BERT, to meet the requirements of the highway design field. Use highway design-related data for model training and fine-tuning. For fine-tuning for specific tasks or fields, this usually involves a smaller but more targeted data set, and adjust the model parameters to better adapt to the specific application scenario, thereby improving performance and accuracy. The entire process needs to be iteratively optimized repeatedly to ensure that the model can accurately identify and generate design solutions.

[0164] Based on the existing large language model foundation and private knowledge base, develop a natural language-driven automated design system - the AIAgent plugin. The system will optimize and generate design solutions by combining task decomposition and execution with the large language model, and finally achieve the full automation and intelligence of highway detection and design tasks.

[0165] In the above embodiments, the descriptions of the various embodiments have their own focuses. For parts not detailed or recorded in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0166] As mentioned above, the above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall all be covered by the protection scope of the present invention.

Claims

1. An AI maintenance framework construction method for automatically generating maintenance construction drawings, characterized in that The method includes the following steps: S1. By using the integrated application of high-resolution remote sensing technology and geographic information system (GIS), a comprehensive visual decision support system is constructed to obtain the result information of the highway network, providing data support for maintenance design based on spatial data; S2. According to the comprehensive visual decision support system, construction standards for an expert database of maintenance measures are constructed according to the actual needs of different regions; S3. Based on the constructed construction standards for the expert database of maintenance measures, the maintenance sections are prioritized according to the target requirements to complete the optimization of maintenance countermeasures; S4. According to the optimization result of the maintenance countermeasures, an automated generation of the maintenance design plan is realized based on the RAG application; S5. According to the generated maintenance design plan, by using the automated function of the CAD software of the large language model, cross-platform intelligent interaction between natural language and the maintenance design plan software is realized, and a full-chain intelligent business closed-loop from pavement disease detection to maintenance design and then to the implementation plan is completed.

2. The AI maintenance framework construction method for automatically generating maintenance construction drawings according to claim 1, characterized in that In step S1, by using the integrated application of high-resolution remote sensing technology and geographic information system (GIS), a comprehensive visual decision support system is constructed, including: Using high-resolution remote sensing images, the information on the earth's surface is obtained through sensors on satellites or high-altitude aircraft and transmitted to the ground station via the data center; the information on the earth's surface includes traffic, meteorology, and geology; Using image processing software for preprocessing such as radiometric correction and geometric correction; then, information is extracted through image enhancement and classification recognition to generate an image map for analysis and application; Through the information overlay of multiple thematic layers of traffic, meteorology, and geology, a comprehensive visual decision support system is constructed to provide data support for maintenance design based on spatial data.

3. The method for building an AI maintenance framework for automatically generating maintenance construction drawings according to claim 2, wherein, The information overlay of multiple thematic layers of traffic, meteorology, and geology includes: The first step is to divide the area of the image of a certain thematic layer among traffic, meteorology, and geology. The divided area image contains complete image information superimposable units of a rows and b columns, or incomplete image information non-superimposable units; Second step, identify the coordinates X1, X2, X3, X4 of the four edge nodes of the image information superposition unit of the a-row and b-column on the graph, the total width Y, and the total height H; identify the distances Y L , Y R , Y U , Y D from each boundary line of the image, and obtain the spacing h between two rows of image information superposition sub-pixels and the spacing y between two columns of image information superposition sub-pixels. Save the Y L , Y R , Y U , Y D parameters locally, and extract the area divided by the points X1, X2, X3, X4 from the graph and save it locally in the form of a picture with a size of W×H; Step 3: By calculating the relationships between X1, X2, X3, X4 and Y L , Y R , Y U , Y D , dynamically obtain traffic, meteorological, and geological images acquired by sensors on satellites or high-altitude aircraft, so that the requirements for image preprocessing are met during the next photo-taking; where Y L , Y R , Y U , Y D The identification process of the parameters includes: (1) Perform canny operator filtering on the first image of a certain preprocessed thematic layer image to obtain a filtered image; (2) Using the elastic ball model theory, analyze the filtered image with the starting point (0,0) to obtain two sets of boundary line point sets; the number of image information superimposable units within the picture is a×b. According to these two sets of boundary line point sets, b + 1 points are analyzed, and then, starting from these b + 1 points respectively, using the elastic ball model theory for analysis, (b + 1)×2 sets of boundary line point sets are obtained; (3) Using (b + 1)×2 sets of boundary line point sets, perform Hough line detection with a negative feedback mechanism to obtain (b + 1)×2 straight lines. Sort the (b + 1)×2 straight lines according to the magnitude of the X-axis coordinates; (4) Analyze any two sets of related point sets in the (b + 1)×2 sets of boundary line point sets, calculate a + 1 points, and starting from these a + 1 points respectively, use the elastic ball model theory for analysis to obtain (a + 1)×2 sets of boundary line point sets; (5) Using (a + 1) × 2 sets of boundary line point sets, and applying Hough line detection with a negative feedback mechanism, the number of lines obtained is (a + 1) × 2 lines. These lines are sorted according to the Y-axis coordinate size. After obtaining (b + 1) × 2 vertical lines and (b + 1) × 2 horizontal lines, calculate W L and W R and W U and W D parameters; In the fourth step, after all thematic layer images are saved, non-negative matrix fusion is performed on the traffic thematic layer source image F1, meteorological thematic layer source image F2, and geological thematic layer source image F3 with a size of B × A to be superimposed, generating a temporary superimposed image C. Construct the observation matrix P of the source images F1, F2, and F3 to be superimposed i , including storing the elements of each observation image row by row into a column vector to obtain the observation matrix P i ; use the non - negative matrix factorization algorithm to decompose the observation matrix P i to obtain the basis matrix L; Wherein, L ia is the base matrix of the a-th row in the i source images to be superimposed, P i is the observation matrix of the i source images to be superimposed, (LH) i is the value of the base matrix after the interval distance of the i source images to be superimposed, H a is the interval distance of the a-th row, L ja is the base matrix of the a-th row in the j-th source image to be superimposed, i is the i source images to be superimposed, and j is the j-th source image to be superimposed; Convert the base matrix L into a matrix with a size of B × A, and the image corresponding to this matrix is the temporary superimposed image C. Solve the multi-directional gradient fields of the source images F1, F2, F3 and the temporary superimposed image C to obtain the image gradient field map; compare the gradient values at the same positions of the four image gradient fields, and select the image with the largest gradient value as the superimposed image selection value to obtain the decision matrix map of the visualization decision support system.

4. The method for building an AI maintenance framework for automatically generating maintenance construction drawings according to claim 1, characterized in that In step S2, construct the construction standards for the maintenance measure expert database, including: by analyzing the differences in geographical division, meteorological hydrology, geological conditions, standards, and material price factors in different regions, using route information as the main data, merging inspection data, highway maintenance historical data, highway grades, regional construction technology levels, material indicators, and regional material quota library maintenance basic materials, and constructing according to the actual needs of different regions.

5. The method for building an AI maintenance framework for automatically generating maintenance construction drawings according to claim 1, characterized in that, In step S3, based on the constructed construction standards for the maintenance measure expert database, prioritize the maintenance sections according to the target requirements to complete the optimization of maintenance countermeasures, including: By accessing the pavement and traffic safety maintenance project cost quotas of the region, dividing the maintenance units, inputting key limiting conditions to adjust the vertical slope and elevation limited sections of the convex curve, and then merging the scattered intermittent maintenance plan sections and adding refined design of the smooth slopes at the level crossings. Finally, through the estimated cost analysis, prioritize the maintenance sections according to the target requirements. This ranking is based on the indicators of cost-effectiveness priority, quality priority, and efficiency priority to achieve the optimized design of maintenance countermeasures.

6. The method for building an AI maintenance framework for automatically generating maintenance construction drawings according to claim 5, wherein, The cost-effectiveness includes: Benefit-cost ratio, benefit-energy consumption ratio, benefit-carbon emission ratio; The priority ranking includes: (a) Obtain the cost-effectiveness f, capital constraint vector g, quality u, and efficiency o indicators. (b) Unify the dimension of the measurement indicators, and use the normalization method to normalize the above indicator values between 0 and 1 to obtain the normalized dimension values D of the above indicators. (c) Use the Analytic Hierarchy Process to define the importance weights within the layer for the above indicators, and obtain the corresponding weights of the above factors, including the cost performance weight Z f , the capital constraint vector weight Z g , the quality weight Z u , the efficiency weight Z o ; (d) According to the importance weights within the above layer, calculate the comprehensive evaluation value of each node within each layer as: J iQ = D f Z f + D g Z g + D u Z u + D o Z o In the formula, J iQ is the comprehensive evaluation value, D f is the normalized dimension value under the performance-price ratio constraint, Z f is the performance-price ratio weight, D g is the normalized dimension value under the capital limit constraint vector, Z g is the capital limit constraint vector weight, D u is the normalized dimension value under the quality constraint, Z u is the quality weight, D o is the normalized dimension value under the efficiency constraint, Z o is the efficiency weight; (e) Use the analytic hierarchy process to calculate the influence weights Zl of each layer respectively. (f) Calculate the multi-layer structure evaluation value of the node's maintenance countermeasure system. (g) Rank the importance of nodes from large to small according to the multi-layer structure evaluation value of the node's maintenance countermeasure system.

7. The method for building an AI maintenance framework for automatically generating maintenance construction drawings according to claim 6, wherein In step (e), use the analytic hierarchy process to calculate the influence weights Zl of each layer respectively, including: In the first step, establish a judgment matrix B, compare the evaluation indicators pairwise, and form the initial weights into the judgment matrix B. The elements in the judgment matrix B represent the scale coefficients obtained after the indicator comparison. Step 2: Calculate the maximum eigenvalue and the corresponding eigenvector for each pairwise comparison judgment matrix, and conduct consistency tests using the consistency index, random consistency index, and consistency ratio; Step 3, the values of the normalized eigenvectors {Z1, Z2, Z3, Z4} are the corresponding weights of the four factors, and the weight of cost performance Z f , the weight of the capital constraint vector Z g , the weight of quality Z u , the weight of efficiency Z o , with the level being l; Z1 = Dl f Z f + Dl g Z g + Dl u Z u + Dl o Z o where D is the normalized dimension value, l f is the influence value under the cost performance constraint, l g is the influence value under the fund limit constraint vector, l u is the influence value under the quality constraint, l o is the influence value under the efficiency constraint; In step (f), calculate the multi-layer structure evaluation value of the maintenance countermeasure system for the node, including: (f-1) First, calculate the product of the comprehensive evaluation value of the node at level Q and the influence weight of this layer, that is, the single-layer evaluation value; (f-2) The multi-layer network evaluation value of the node is the sum of the single-layer evaluation values.

8. The method for building an AI maintenance framework for automatically generating maintenance construction drawings according to claim 1, characterized in that In step S4, according to the optimization result of the maintenance countermeasure, based on the RAG application, realize the automatic generation of the maintenance design plan, including: S401, The knowledge base adopts a hierarchical structure, divides the information into multiple levels, including the file level, chapter level, and paragraph level. Each level contains information tags for locating the required knowledge; S402, Design an index structure, with an n-ary tree as the index, including establishing keyword indexes, entity indexes, and semantic indexes; S403, Set up a semi-automatic update process. By segmenting and classifying the new text based on simple characters and text semantics, combined with manual review, evaluate and verify the newly added or modified knowledge points. At the same time, for the existing knowledge entries, establish a feedback mechanism to improve the content of the knowledge base by using user-reported errors or improvement suggestions; S404, For the sensitive or private data contained in the knowledge base, take data security measures, including md5 or rsa data encryption, user authentication access control, to ensure the implementation of the RAG system for the knowledge base.

9. The AI maintenance framework building method for automatically generating maintenance construction drawings according to claim 1, characterized in that In step S5, realize cross-platform intelligent interaction between natural language and the maintenance design plan software, including: Describe the task to be executed through natural language, and the AI will structure the instruction task, form a script or programming language, and automatically input it into the desktop design software. The desktop design software executes various types of drawing tasks by executing the script and programming language; The execution plan includes: The detection device will collect and process road data in real time, automatically fill in the key content through a prefabricated report template to generate a disease report; Automatically generate a maintenance design plan based on the text content generated by the large language model, and realize the automation of drawing design through the interaction between natural language and the desktop design software; Implement road maintenance according to the automatically generated design plan.

10. An AI maintenance framework building system for automatically generating maintenance construction drawings, characterized in that, The system implements the AI maintenance framework construction method for automatically generating maintenance construction drawings as described in any one of claims 1-9. The system includes: A comprehensive visual decision support system construction module (1), used to construct a comprehensive visual decision support system by integrating the application of high-resolution remote sensing technology and geographic information system GIS, obtain the highway road network result information, and provide data support for maintenance design based on spatial data; A maintenance measure expert database construction module (2), used to construct the construction standard of the maintenance measure expert database according to the comprehensive visual decision support system for the actual needs of different regions; A maintenance countermeasure optimization module (3), used to prioritize the maintenance sections according to the target requirements based on the constructed maintenance measure expert database construction standard, and complete the optimization of the maintenance countermeasures; Automated Generation Solution for Maintenance Design Plan (4), which is used to automatically generate a maintenance design plan based on the optimization results of maintenance countermeasures and through the application of RAG; Pavement Disease Detection Module (5), which is used to realize cross-platform intelligent interaction between natural language and maintenance design plan software according to the generated maintenance design plan by using the automated function of CAD software of large language models, and complete the full-chain intelligent business closed-loop from pavement disease detection to maintenance design and then to the implementation plan.

Images