Urban drainage pipeline defect diagnosis and report automatic issuing method
Through automated pipeline robots and computer vision technology, the defects of urban drainage pipelines are identified and combined with artificial intelligence generation diagnostic opinions, the problems of low efficiency in defect image data collection and complex report preparation in traditional detection technology are solved, and efficient and accurate detection report generation is achieved.
Patent Information
- Application Number
- CN202410957619.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-17
- Publication Date
- 2024-11-12
AI Technical Summary
Traditional urban drainage pipeline detection technology has defects. The image data collection efficiency is low, the data accuracy and reliability are affected by human factors. The preparation of the test report takes time, is prone to errors, and the report format is inconsistent, resulting in inconsistent diagnostic opinions.
Automatic pipeline robots are used to collect data, identify defect types and levels through computer vision technology, combine artificial intelligence to generate diagnostic and repair opinions, and use automated report generation technology to unified report format to realize automatic collection, transmission and reporting output of defect data.
It improves the accuracy and efficiency of defect recognition rate and detection reports, simplifies data processing tasks, unifies diagnostic opinions, and ensures that the report format is beautiful and the data is accurate.
Smart Images

Figure CN118941515A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of pipeline detection, and in particular relates to a method for diagnosing urban drainage pipeline defects and automatically issuing reports. Background Art
[0002] The urban drainage system is one of the urban infrastructures and is called the blood vessels of the city. The underground drainage pipe is an important part of the urban drainage system. In the construction of urban drainage projects, it is necessary to control the function and structural stability of the pipes to ensure that the drainage system can be put into normal use. In the process of continuous advancement of urbanization, the aging and leakage of road drainage pipes have occurred repeatedly, resulting in serious waterlogging and landslides on some roads, which has greatly affected the quality of life of the people.
[0003] In terms of urban drainage pipeline inspection, traditional inspection technology can no longer meet the needs of pipeline inspection. The closed-circuit television system inspection technology and equipment are relatively advanced. With the assistance of cameras, lenses, lights and other equipment, the internal conditions of the pipeline can be recorded, providing accurate images and data for technical personnel to collect information. Conducting pipeline endoscopic video inspection (CCTV), pipeline periscope inspection (QV) and sonar inspection (Sonar) on drainage pipeline projects can efficiently complete pipeline defect detection and positioning, provide a scientific basis for pipeline repair and renovation by providing accurate images, and then ensure the quality of drainage pipeline construction;
[0004] With the advancement of science and technology, pipeline CCTV inspection robots and periscope QV inspection systems have been widely used in this field, bringing revolutionary changes to defect image collection. Although the introduction of new technologies has greatly improved the detection efficiency, there are still some problems that need to be solved. Among them, the most prominent is the low efficiency of defect image data collection: new technologies have brought many conveniences, but in terms of the identification and collection of defect image data, it is still impossible to achieve complete automation. At present, this process still needs to rely on technicians to make visual judgments and form defect image data by taking screenshots. This method is not only inefficient, but also affected by human interference, which affects the accuracy and reliability of the data;
[0005] When the test report is issued, the following problems still exist: ① The error rate of the test data set information is high. In the process of preparing the test report, a large amount of test data processing and calculation is involved, which is time-consuming, error-prone, complex result summary, and difficult manual verification. ② The report format is messy and lacks brand recognition. Manually prepared reports rely on the report preparers, which are subjective, and the problems of report standardization and inconsistent format are prominent. ③ Due to differences in the judgment of the severity of defects by technicians, there are often problems of inconsistent diagnosis and rectification opinions. Summary of the invention
[0006] In order to solve the above technical problems, the present invention provides a method for automatically diagnosing and reporting defects in urban drainage pipes, which can automatically collect and transmit image data, so as to use a computer to identify the defect type and defect level in the image, and use artificial intelligence to provide diagnostic and repair suggestions for the corresponding defects, and finally output the inspection reports in batches with beautiful report format and accurate data.
[0007] The present invention provides a method for diagnosing urban drainage pipe defects and automatically issuing reports, comprising the following steps:
[0008] S1. Use an automated pipeline robot to collect data in the pipeline and convert the video into image frames. Use a positioning module and a communication module to achieve real-time data transmission.
[0009] S2. Use computer vision technology to identify the defect type of the image frame, and at the same time calculate the relative area of the defective pixel points to determine the defect level;
[0010] S3. Use automated data processing technology to process the data collected from the urban drainage and flood control survey project, form a pipe network inspection details summary table and a defect summary table, and batch generate a defect details table corresponding to each defect;
[0011] S4. Use large language model technology to propose rectification suggestions and repair plans for defects of different types and levels according to actual engineering conditions;
[0012] S5. Use automated report generation technology to determine the report layout, format, and fixed wording, lay out tables, images, rectification suggestions, and repair plans, and generate a unified format for urban drainage and flood control census reports.
[0013] Furthermore, in step S2, the following steps are included:
[0014] S21, marking the image data acquired in step S1;
[0015] S22, dividing the annotated image data into a training set and a test set, inputting the training set into a deep learning model for training, and obtaining an initial pipeline defect detection model;
[0016] S23, inputting the test set into the initial pipeline detection model to obtain an optimized pipeline detection model;
[0017] S24, performing pipeline defect detection and storing the result graph;
[0018] S25, pre-processing the image into grayscale and binarization, processing the image into black and white pixels, and the defect into white;
[0019] S26. Calculate the relative width and area of the white portion, and classify the defects according to their sizes.
[0020] Furthermore, in step S3, the following steps are also included:
[0021] S31. After the defect image is identified and extracted, the relevant information will be named on the image according to the established rules;
[0022] S32, while inserting the picture into the report template, extracting the naming information on the picture, and automatically filling the naming information into the corresponding placeholder position in the pipe network detection details summary table template to form a pipe network detection details summary table;
[0023] S33, filling the defect summary table template with the defect evaluation level according to the defect image to form a defect summary table;
[0024] S34. Introduce the defect summary pictures one by one to form a defect status table.
[0025] Furthermore, the defects include structural defects and functional defects.
[0026] Furthermore, the structural defects include: rupture, deformation, corrosion, interface, concealed connection of branch pipes and leakage, and the functional defects include: the presence of obstacles.
[0027] Furthermore, the default value of the defect is zero, and if a defect of a defect type is found in the cycle, the defect value is automatically increased by one.
[0028] Furthermore, in step S4, a large model base is used and the knowledge base is connected to the langchain framework to construct a town drainage pipeline detection model.
[0029] Furthermore, the input data for constructing the urban drainage pipe detection model includes: a knowledge base in the field of urban drainage pipe detection, a professional dataset after self-learning, and SFT supervised fine-tuning.
[0030] Furthermore, the knowledge base in the field of urban drainage pipe inspection includes: legal regulations and standards research, inspection technology, quality management, laboratory construction, personnel training, and risk management data.
[0031] Furthermore, the urban drainage pipe detection model training includes the following steps:
[0032] S41. Collect data on personnel's preferences for summaries;
[0033] S42, train the RM reward model, where formula 1 is the reward loss function,
[0034]
[0035] S43, optimizing the model by PPO algorithm; wherein, adding a β penalty term to Formula 2,
[0036]
[0037] The beneficial effect of the present invention is that the method for automatically issuing defects diagnosis and reports on urban drainage pipes of the present invention is based on controllable and data-transmitting detection equipment, computer vision technology that can accurately identify defects, data processing technology that can accurately and batch generate defect reports and summarize them, an artificial intelligence large model that can accurately put forward diagnostic and repair suggestions, and technology that can automatically generate urban drainage and flood control census reports using text templates. By combining these five parts, a method for automatically issuing defects diagnosis and reports on urban drainage pipes is established, which can automatically collect and transmit image data, thereby using a computer to identify the defect category and defect level in the image, with a high defect recognition rate, and through artificial intelligence, diagnostic and repair suggestions are put forward for the corresponding defects. Finally, the inspection reports are output in batches with beautiful report formats and accurate data.
[0038] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0040] Figure 1 A flow chart of a method for automatically issuing a report and diagnosing defects in urban drainage pipes according to the present invention;
[0041] Figure 2 The principle block diagram of the method for automatically issuing a defect diagnosis and report for urban drainage pipes of the present invention;
[0042] Figure 3 A flow chart of determining defect levels of the method for automatically issuing defect reports and diagnosing defects in urban drainage pipes of the present invention;
[0043] Figure 4 A principle block diagram of determining the defect level of the method for automatically issuing a defect diagnosis and report for urban drainage pipes of the present invention;
[0044] Figure 5 A schematic diagram of the backbone network of the method for automatically providing a defect diagnosis and report for urban drainage pipes of the present invention;
[0045] Figure 6 A schematic diagram of step S3 of the method for automatically issuing a defect diagnosis and report for urban drainage pipes of the present invention;
[0046] Figure 7 A flowchart of step S4 of the method for automatically issuing a defect diagnosis and report for urban drainage pipes of the present invention;
[0047] Figure 8 A flow chart of the urban drainage pipeline detection model training of the method for automatically issuing urban drainage pipeline defect diagnosis and report of the present invention;
[0048] Fig. 9 This is a schematic diagram of the catalog page of the urban drainage and flood prevention census report of the method for automatically issuing urban drainage pipe defect diagnosis and report of the present invention. DETAILED DESCRIPTION
[0049] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0050] like Figure 1 and Figure 2 As shown, the present invention provides a method for diagnosing urban drainage pipe defects and automatically issuing reports, comprising the following steps:
[0051] S1. Use an automated pipeline robot to collect data inside the pipeline and convert the video into image frames. Use a positioning module and a communication module to achieve real-time data transmission. A pipeline CCTV detection robot can be used to collect data inside the pipeline.
[0052] S2. Use computer vision technology to identify the defect type of the image frame, and at the same time calculate the relative area of the defective pixel points to determine the defect level;
[0053] Specifically, Figure 3 and Figure 4 As shown, in step S2, the following steps are included:
[0054] S21. Label the image data obtained in step S1. You can use deep learning image labeling tools such as labelme to label the frame. S22. Divide the labeled image data into a training set and a test set, input the training set into the deep learning model for training, and obtain the initial pipeline defect detection model. Divide the image in this step into five parts, four of which are used as training sets and one as a test set, and observe the training results, accuracy, and loss function. S23. Input the test set into the initial pipeline detection model to obtain an optimized pipeline detection model. S24. Perform pipeline defect detection and store the result map. S25. Pre-process the image into grayscale and binarization, process it into black and white pixels, and the defect is white. S26. Calculate the relative width and area of the white part, and classify the defect level according to the defect size.
[0055] like Figure 5 As shown in the figure, the main network processes the input image through a series of convolutional layers and pooling layers to retain and extract important features in the defect image. The Focus module is used to slice the image and extract the input defect features through eight convolutional layers: 4 convolutional layers (conv1, conv2, conv3, conv4) and 4 C3 layers (3*C3, 6*C3, 9*C3, 3*C3). The C3 layer is also a convolutional layer. There are three convolutional layers in the C3 module. The left convolution extracts half of the features, and the right convolution also extracts half of the features. After being processed by BottleNeck, the two parts are concat (joined), and then passed through a convolutional layer again, and SPPF is used for pyramid partitioning and pooling operations.
[0056] For the above sampling, the Upsample_Concat method is defined to amplify and connect the passed values. The basic architecture is: Conv1→Upsample1→Concat1→3*C3→Conv1→Upsample1→Concat1→3*C3. Then convolution and splicing are performed, and prediction heads are used for prediction. Here, two prediction heads are output in series, making the defect detection results more accurate.
[0057] S3. Use automated data processing technology to process the data collected from the urban drainage and flood control survey project, form a pipe network inspection details summary table and a defect summary table, and batch generate a defect details table corresponding to each defect;
[0058] Specifically, if Figure 6As shown, in step S3, the following steps are also included: S31. When the defect image is identified and extracted, the relevant information will be named on the image according to the established rules; S32. When inserting the image into the report template, the naming information on the image is extracted, and the naming information is automatically filled into the corresponding placeholder position in the pipeline inspection details summary table template to form a pipeline inspection details summary table; S33. According to the defect assessment level of the defect image, it is filled into the defect summary table template to form a defect summary table; S34. The defect summary images are introduced one by one to form a defect situation table.
[0059] In this embodiment, the image naming method is: defect number_pipe number_pipe diameter (mm)_detection length (m)_pipe material_defect name_severity (level)_location;
[0060] The corresponding placeholder symbol is: qxbh_gdbh_gj_cd_gc_qxmc_yzcd_wz
[0061] The basic information data of the pipe section is obtained after defect identification. The defect map output will be output in a fixed file output format, the image details will be identified, and the basic information of the pipe section will be extracted. The basic information will be output as a summary table in the format of the pipeline network inspection details summary table.
[0062] Table 1 Pipeline network inspection details summary table template
[0063] Serial number Pipe segment Pipe diameter(mm) Detection length (m) Detection Methods Pipe qx gdbh gj cd CCTV gc
[0064] The defects in the pipe network inspection details are counted to form a summary table, including two categories: structural defects and functional defects. The severity of the defects is divided into four types: mild 1, moderate 2, severe 3, and major 4. A defect summary table template is established, and placeholders are set for the cells that need to be summarized. Each placeholder is named by the defect type and defect level, and the default value is 0. In the loop, each defect level under a defect type is found, and the value is added by one. For example, pl_1 is a first-level rupture. If a first-level rupture is found, pl_1=pl_1+1.
[0065] Table 2 Defect summary table template
[0066]
[0067] Defects appearing in the pipe network inspection details are described in detail one by one, including defect number, pipe number, pipe material, pipe diameter, defect name, severity (level) and description. The description includes the pipe number, defect location and a fixed description statement based on the defect type and corresponding severity.
[0068] Table 3 Defect status table template
[0069]
[0070] S4. Use large language model technology to propose rectification suggestions and repair plans for defects of different types and levels according to actual engineering conditions; Figure 7 As shown in the figure, a large model base is used and the knowledge base is connected to the langchain framework to build a town drainage pipeline detection model, which has two application scenarios and can put forward rectification suggestions and repair plans. In this embodiment, the diagnosis and rectification opinions of existing reports are collected, and a knowledge base for urban drainage and flood control census diagnosis and rectification is built. Vectorized embedding is used, and the vector correlation algorithm is used to find the most matching knowledge base fragments. The ChatGLM language model is used to put forward more accurate and personalized rectification suggestions and repair plans according to the actual engineering conditions.
[0071] Table 4 Proposed repair solutions
[0072] Defect Number Pipe segment location Defect Name Defect level Correction suggestions Is it accurate? 001 Y1182_Y1183 rupture Severe (Grade 3) Should be fixed immediately yes 002 Y1189_Y1524 rupture Severe (Grade 3) Should be fixed immediately yes 003 Y1189_Y1806 rupture Severe (Grade 3) Should be fixed immediately yes 004 Y1226_Y1269 rupture Major (Level 4) Should be fixed immediately yes 005 Y1228_Y1139 rupture Severe (Grade 3) Should be fixed immediately yes 006 Y1241_Y1498 interface Severe (Grade 3) Should be fixed immediately yes 007 Y1259_Y1260 rupture Severe (Grade 3) Should be fixed immediately yes 008 Y1267_Y1002 rupture Severe (Grade 3) Should be fixed immediately yes 009 Y1340_Y1374 rupture Severe (Grade 3) Should be fixed immediately yes
[0073] Table 5 Proposed repair solutions
[0074]
[0075]
[0076] The input data for building the urban drainage pipeline detection model include: urban drainage pipeline detection field knowledge base, professional data sets after self-learning, and SFT supervised fine-tuning. The urban drainage pipeline detection field knowledge base includes: laws, regulations and standards research, detection technology, quality management, laboratory construction, personnel training, and risk management data. Figure 8 As shown in the figure, the urban drainage pipe detection model training includes the following steps:
[0077] S41. Collect data on personnel's preferences for summaries;
[0078] S42. Train the RM reward model. The reward is no longer considered as a given. Instead, the reward model is trained with high-quality labeled data. Formula 1 is the reward loss function, r θ (x,y) is the RM model, x is the prompt input, in actual training, the batch size is 64, which indicates the number of independent prompts in each batch, y is the predicted output of SFT, and D is the human comparison dataset.
[0079] S43. With the reward model, based on the premise of maximizing the score given by RM, the model strategy is optimized through the PPO algorithm. In order to avoid the RM reward model being too absolute, a β penalty term is added to the RM reward model to prevent the strategy in the iteration from deviating too far from the initial SFT strategy. The addition of the β penalty term is shown in Formula 2, where π SFT is the baseline strategy, For the new strategy, S42 has obtained the RM model learned based on human preferences, so there is only one r θ (x,y), instead of retraining by comparison and sorting.
[0080]
[0081] S5. Use automated report generation technology to determine the report layout, format, and fixed words, typeset tables, images, rectification suggestions, and repair plans, and generate a unified urban drainage and flood control census report. In this embodiment, a report form template can be written, and the data automatic filling function can be implemented. Compared with the need to first enter the information into the management system and then output it according to the template, this method reduces one manual step. According to one embodiment of the present invention, the python-docx library is used to process the word solution document, and the first-level title and the second-level title are distinguished by the special format of the title directory. The report layout, format, and fixed words are used to typeset tables, images, rectification suggestions, and repair plans to generate a unified urban drainage and flood control census report template. The data generated in step S3 and step S4 are automatically filled into the urban drainage and flood control census report template. The directory is as follows Fig. 9 shown.
[0082] The method for automatically issuing defect diagnosis and reports on urban drainage pipelines of the present invention is based on controllable and data-transmitting detection equipment, computer vision technology that can accurately identify defects, data processing technology that can accurately and batch generate defect reports and summarize them, an artificial intelligence large model that can accurately put forward diagnostic and repair suggestions, and technology that can automatically generate urban drainage and flood control census reports using text templates. By combining these five parts, a method for automatically issuing defect diagnosis and reports on urban drainage pipelines is established, which can automatically collect and transmit image data, thereby using a computer to identify the defect type and defect level in the image, with a high defect recognition rate, and put forward diagnostic and repair suggestions for the corresponding defects through artificial intelligence. Finally, the detection reports are output in batches with beautiful report formats and accurate data.
[0083] In summary, the method of automatic diagnosis and reporting of urban drainage pipe defects has improved the efficiency and accuracy of urban drainage pipe inspection data collection and defect diagnosis, simplified the defect image processing task, unified the diagnosis and rectification opinions, and thus reduced the occurrence of flood disasters, which will be conducive to improving social stability, maintaining urban social order, and protecting people’s lives and property.
[0084] The various devices selected in this application are all universal standard parts or parts known to those skilled in the art, and their structures and principles can be known to those skilled in the art through technical manuals or conventional experimental methods.
[0085] In the description of the embodiments of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or a connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0086] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.
[0087] Based on the above ideal embodiments of the present invention, the relevant staff can make various changes and modifications without departing from the technical concept of the present invention through the above description. The technical scope of the present invention is not limited to the contents of the specification, and its technical scope must be determined according to the scope of the claims.
Claims
1. A method for diagnosing and automatically issuing reports on urban drainage pipe defects, characterized in that: The following steps are involved: S1. Use an automated pipeline robot to collect data in the pipeline and convert the video into image frames. Use a positioning module and a communication module to achieve real-time data transmission. S2. Use computer vision technology to identify the defect type of the image frame, and at the same time calculate the relative area of the defective pixel points to determine the defect level; S3. Use automated data processing technology to process the data collected from the urban drainage and flood control survey project, form a pipe network inspection details summary table and a defect summary table, and batch generate a defect details table corresponding to each defect; S4. Use large language model technology to propose rectification suggestions and repair plans for defects of different types and levels according to actual engineering conditions; S5. Use automated report generation technology to determine the report layout, format, and fixed wording, and lay out tables, images, rectification suggestions, and repair plans.
2. The method for automatically diagnosing and reporting urban drainage pipe defects as claimed in claim 1, characterized in that: In step S2, the following steps are included: S21, marking the image data acquired in step S1; S22, dividing the annotated image data into a training set and a test set, inputting the training set into a deep learning model for training, and obtaining an initial pipeline defect detection model; S23, inputting the test set into the initial pipeline detection model to obtain an optimized pipeline detection model; S24, performing pipeline defect detection and storing the result graph; S25, pre-processing the image into grayscale and binarization, processing the image into black and white pixels, and the defect into white; S26. Calculate the relative width and area of the white portion, and classify the defects according to their sizes.
3. The method for automatically diagnosing and reporting urban drainage pipe defects as claimed in claim 1, characterized in that: In step S3, the following steps are also included: S31. After the defect image is identified and extracted, the relevant information will be named on the image according to the established rules; S32, while inserting the picture into the report template, extracting the naming information on the picture, and automatically filling the naming information into the corresponding placeholder position in the pipe network detection details summary table template to form a pipe network detection details summary table; S33, filling the defect summary table template with the defect evaluation level according to the defect image to form a defect summary table; S34. Introduce the defect summary pictures one by one to form a defect status table.
4. The method for automatically diagnosing and reporting urban drainage pipe defects as claimed in claim 3, characterized in that: The defects include structural defects and functional defects.
5. The method for automatically diagnosing and reporting urban drainage pipe defects as claimed in claim 4, characterized in that: The structural defects include: rupture, deformation, corrosion, interface, concealed connection of branch pipes and leakage; the functional defects include: the presence of obstacles.
6. The method for automatically diagnosing and reporting urban drainage pipe defects as claimed in claim 5, characterized in that: The default value of the defect is zero. If a defect of a defect type is found in the cycle, the defect value will be automatically increased by one.
7. The method for automatically diagnosing and reporting urban drainage pipe defects as claimed in claim 1, characterized in that: In step S4, a large model base is used and the knowledge base is connected to the langchain framework to build a town drainage pipeline detection model.
8. The method for automatically diagnosing and reporting urban drainage pipe defects as claimed in claim 7, characterized in that: The input data for building the urban drainage pipe detection model include: urban drainage pipe detection field knowledge base, professional data set after self-learning, and SFT supervised fine-tuning.
9. The method for automatically diagnosing and reporting urban drainage pipe defects as claimed in claim 8, characterized in that: The knowledge base in the field of urban drainage pipeline inspection includes: legal regulations and standards research, inspection technology, quality management, laboratory construction, personnel training, and risk management data.
10. The method for diagnosing and automatically issuing reports on urban drainage pipe defects according to claim 7, characterized in that: The urban drainage pipe detection model training includes the following steps: S41. Collect data on personnel's preferences for summaries; S42, train the RM reward model, where formula 1 is the reward loss function, S43, optimizing the model by PPO algorithm; wherein, adding a β penalty term to Formula 2,
Citation Information
Patent Citations
Underground drainage pipeline endoscopic detection and evaluation system
CN114897371A
Pipeline detection method and system, terminal equipment and storage medium
CN117173130A
Gas pipeline monitoring system and method based on artificial intelligence mixed large model
CN117523177A
