Self-adaptive power pipeline robot repairing method and system based on defect database

Abstract

The invention relates to the technical field of power pipeline maintenance, in particular to a self-adaptive power pipeline robot repairing method and system based on a defect database, and the method comprises the steps: obtaining power pipeline defect data in real time through a collection assembly, and extracting multi-dimensional defect feature parameters through the defect data; constructing an initial power pipeline defect model based on the multi-dimensional defect characteristic parameters and historical defect data, and performing similarity matching on real-time power pipeline defect data and data in the initial power pipeline defect model to obtain a power pipeline defect updating model; and after the repair is completed, the repair quality detection sensor comprehensively detects the repair area, and if the detection result is not ideal, the optimal repair strategy is regenerated by adopting a repair strategy adjustment algorithm based on the real-time power pipeline defect data. According to the method, through automatic defect identification and strategy generation and execution, the manual intervention requirement is remarkably reduced, and the repair efficiency is improved.

Description

Technical Field

[0001] The present application relates to the technical field of power pipeline maintenance, and in particular to an adaptive power pipeline robot repair method and system based on a defect database. Background Art

[0002] As an important part of the power transmission system, power pipelines have a complex operating environment and often face problems such as mechanical damage, corrosion, thermal expansion and contraction. They are prone to various defects such as cracks, holes, and surface corrosion. If these defects cannot be repaired in time, they will seriously affect the safe operation of power pipelines and even cause large-scale power outages.

[0003] Traditional power pipeline repair mainly relies on manual operation, and the repair process includes welding, filling, spraying, etc. However, manual repair has the following problems:

[0004] (1) Low efficiency: Manual operation takes a long time and is subject to greater environmental restrictions. For example, it is difficult to operate under high temperature and high pressure conditions.

[0005] (2) Unstable quality: The repair effect depends on the operator's experience and technical level, which easily leads to uneven repair quality.

[0006] (3) High risk: When repairing in-service power pipelines, operators face the threat of high voltage and harsh environment.

[0007] (4) Limitations: Traditional methods have difficulty dealing with complex and diverse defects, especially those involving high-precision and high-intensity repair tasks.

[0008] In recent years, with the rapid development of robotics and artificial intelligence, robot-based power pipeline repair technology has begun to receive widespread attention. These robots usually have automation and remote control functions, which can reduce the risk of manual participation and improve repair efficiency. However, the existing robot repair solutions still have the following technical deficiencies:

[0009] (1) Lack of intelligence: Many repair robots only have fixed repair strategies and cannot flexibly adjust the repair plan according to the specific characteristics of the defect.

[0010] (2) Lack of real-time optimization capability: It is difficult to optimize the repair parameters based on real-time feedback during the repair process, resulting in unstable repair effects.

[0011] (3) Lack of knowledge accumulation mechanism: repair experience is not systematically managed, and repair data cannot be used to optimize subsequent tasks. Summary of the invention

[0012] Purpose of the invention: The purpose of this application is to provide an adaptive power pipeline robot repair method based on a defect database, so as to solve the technical problems existing in the prior art such as insufficient intelligence and precision of the repair process, poor adaptability of the repair scheme, low repair efficiency, unstable repair quality, etc. The present invention also provides an adaptive power pipeline robot repair system based on a defect database.

[0013] Technical solution:

[0014] In view of the above problems, the present application provides an adaptive power pipeline robot repair method based on a defect database, the method comprising:

[0015] Acquire power pipeline defect data in real time through a collection component, and extract multi-dimensional defect characteristic parameters through the defect data;

[0016] The power pipeline defect initial model is constructed based on the multi-dimensional defect characteristic parameters and the historical defect data, and the real-time power pipeline defect data is similarly matched with the data in the power pipeline defect initial model, so as to obtain the power pipeline defect update model; and the optimal repair strategy for the real-time power pipeline defect data is obtained according to the similarity matching value;

[0017] According to the obtained optimal repair strategy, the movement path of the repair tool is planned to ensure that the repair operation covers all defect areas. The repair tool performs repair operations according to the planned path, and a repair trajectory model is constructed according to the obtained movement path and the corresponding multi-dimensional characteristic parameters of the power pipeline defect;

[0018] After the repair is completed, the repair quality detection sensor performs a comprehensive inspection of the repair area. If the inspection result is not ideal, the repair strategy adjustment algorithm is used based on the real-time power pipeline defect data to regenerate the best repair strategy, and update the power pipeline defect update model and repair trajectory model;

[0019] Repeat the above steps to identify and repair more power pipeline defects and continuously optimize the repair strategy.

[0020] Further, including:

[0021] The real-time acquisition of power pipeline defect data by the acquisition component includes:

[0022] The acquisition component is set at a suitable position to accurately collect the power pipeline defect data. The acquisition component includes a laser radar sensor, a visual camera and an ultrasonic sensor. The laser radar sensor is used to scan the geometric shape of the pipeline surface to obtain high-precision three-dimensional point cloud data. The visual camera is used to capture high-definition images of the pipeline surface to assist in identifying the visual features of defects. The ultrasonic sensor is used to detect defects inside the pipeline, thereby obtaining the geometric shape, size and material information of the defects in real time.

[0023] The collected data are filtered and denoised, and multimodal data are fused to generate unified defect description data.

[0024] Further, including:

[0025] The extracting multi-dimensional defect characteristic parameters through the defect data includes:

[0026] Performing edge detection and morphological analysis on the high-definition image captured by the visual camera to extract two-dimensional features of the defect;

[0027] The 3D point cloud data collected by the LiDAR is processed to build a model, and the 3D model of the corresponding defect is reconstructed to calculate the volume and depth of the defect;

[0028] A multidimensional feature model is constructed based on the two-dimensional features, the three-dimensional model and the distance and position information of the defect collected by the ultrasonic sensor, and the geometric shape, geometric features, material features and surface texture features of the defect are extracted from the multidimensional feature model.

[0029] Further, including:

[0030] The power pipeline defect update model includes:

[0031] The power pipeline defect initial model is the historical defect data stored in the database, and each record includes: defect type, geometric features, material properties, surface texture features, and the best repair strategy corresponding to the defect;

[0032] Updating the real-time multi-dimensional defect characteristic parameters corresponding to the power pipeline defect data into the power pipeline defect initial model;

[0033] The multi-dimensional defect feature parameters are matched with a feature vector formed by the features in each record for similarity, and the repair strategy corresponding to the historical record with the highest similarity result is used as the best repair strategy for the current defect.

[0034] Further, including:

[0035] The historical defect data is stored in a database, including:

[0036] The database has a two-level index structure, including: the first-level index is to use a clustering algorithm to preliminarily classify the historical defect data according to the defect type or material characteristics, and each category forms a sub-database; the second-level index is to build a multidimensional index structure inside each sub-database, and the multidimensional index structure recursively divides the feature space, reduces the dimension of the high-dimensional features, and performs fast neighbor search.

[0037] Further, including:

[0038] The multi-dimensional defect feature parameter is matched with a feature vector formed by the features in each record by similarity, including:

[0039] The feature vector of each historical defect data record is standardized to obtain the first feature vector.

[0040] The second eigenvector is obtained by standardizing and reducing the dimensionality of the eigenvector formed by the multi-dimensional defect characteristic parameters;

[0041] Setting a weight for a corresponding feature dimension, and calculating a corresponding weighted Euclidean distance according to the first feature vector and the second feature vector;

[0042] If the repair quality is low, the weights are updated through gradient descent to continuously optimize the accuracy of similarity matching.

[0043] Further, including:

[0044] The step of setting the weight of the corresponding feature dimension and calculating the corresponding weighted Euclidean distance according to the first feature vector and the second feature vector includes:

[0045] Pre-classifying the second feature vector according to the defect type and selecting the corresponding sub-database;

[0046] In the multidimensional index structure of the selected sub-database, a nearest neighbor search is performed with the second eigenvector as a query point to obtain k most similar historical defect records;

[0047] Sort these k results according to the weighted distance and select the first N results with the smallest distance;

[0048] Extract the corresponding repair strategy set from the most similar top N defect records;

[0049] If the set of repair strategies is greater than 1, a voting or weighted average strategy is used to select the best repair strategy.

[0050] Further, including:

[0051] If the repair quality is low, the weights are updated by gradient descent to continuously optimize the accuracy of similarity matching, including:

[0052] After each repair is completed, the repair quality is evaluated. It is assumed that the best strategy corresponding to the historical defect record matched in this repair is used to repair the current defect. If the repair quality is high, it means that the feature weight distribution is reasonable; if the evaluation result is that the repair quality is low, it means that the weights of some feature dimensions in the similarity measurement are not set properly, and the weights are updated through gradient descent;

[0053] A corresponding loss function is constructed based on the updated weights, which includes the expected value of the repair quality. When the repair quality is lower than the expected value, the relevant features in the matched defect records are analyzed. If a certain type of feature has a high correlation with the repair success rate in the database, the weight of the feature of this dimension is increased; if a certain type of feature shows a low correlation with the repair success rate in multiple repairs, its corresponding weight is reduced; this update process is carried out online, and the weight is recalculated after each batch of repair tasks is completed.

[0054] Further, including:

[0055] The method further includes:

[0056] If the detection result is ideal, the repair trajectory in the repair trajectory model and the corresponding repair effect are updated to the power pipeline defect update model to form a historical repair case record;

[0057] The power pipeline defect update model is continuously optimized and updated using a machine learning algorithm.

[0058] On the other hand, the present invention also provides an adaptive power pipeline robot repair system based on a defect database, the system comprising:

[0059] An acquisition module is used to acquire power pipeline defect data in real time through an acquisition component, and extract multi-dimensional defect characteristic parameters through the defect data;

[0060] A repair strategy generation module is used to construct the power pipeline defect initial model based on the multi-dimensional defect characteristic parameters and historical defect data, and perform similarity matching between the real-time power pipeline defect data and the data in the power pipeline defect initial model, so as to obtain the power pipeline defect update model; and obtain the best repair strategy for the real-time power pipeline defect data according to the similarity matching value;

[0061] A repair execution module is used to plan the movement path of the repair tool according to the obtained optimal repair strategy to ensure that the repair operation covers all defect areas. The repair tool performs repair operations according to the planned path, and constructs a repair trajectory model according to the obtained movement path and the corresponding multi-dimensional characteristic parameters of the power pipeline defect;

[0062] A repair quality detection and feedback module is used to perform a comprehensive inspection of the repair area by the repair quality detection sensor after the repair is completed. If the inspection result is not ideal, the repair strategy adjustment algorithm is used based on the real-time power pipeline defect data to regenerate the best repair strategy, and update the power pipeline defect update model and repair trajectory model;

[0063] The iteration module is used to cyclically execute the operations of the above modules, identify and repair more power pipeline defects, and continuously optimize the repair strategy.

[0064] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0065] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0066] The present invention first fuses multiple sensor data to obtain richer and more accurate defect feature parameters. Traditional technologies only rely on vision or lidar data to extract two-dimensional or three-dimensional parameters, while the present invention fuses multimodal data and combines the advantages of different sensors. This not only improves the accuracy and reliability of defect identification, but also provides high-quality input data for subsequent adaptive repair strategy generation. This allows defect feature extraction to be organically combined with subsequent adaptive repair strategy generation and quality feedback optimization to form a closed-loop system.

[0067] Specifically: First, the similarity matching of the present invention is not only for simple single parameters such as length and width, but is based on multi-dimensional and deep feature parameters obtained through multimodal data fusion. Then, the weight setting for each feature parameter is not fixed, but can be automatically optimized through a machine learning model, and the feature weights can be dynamically adjusted under long-term data accumulation to achieve more accurate similarity matching. Traditional technologies generally use fixed or empirically set similarity calculation methods, which lack such dynamic optimization and weight update mechanisms. Next, the present invention deeply couples similarity matching with continuous learning optimization of the defect database. The similarity matching results are not used only in one task, but the database and matching algorithm are continuously updated through subsequent repair feedback data. Finally, after similarity matching, an adaptive repair strategy is assisted in generating, rather than simply classifying or sorting. The present invention takes into account the correspondence between defects and repair solutions in the matching process, that is, similarity matching not only matches "similar defects", but also directly relates to "corresponding optimal repair solutions and their parameters".

[0068] After the repair is completed, the present invention detects the repair effect (such as surface integrity and adhesion strength) in real time, and uses the feedback results to optimize the repair strategy to ensure the reliability of the repair quality.

[0069] Therefore, the present invention supports the dynamic selection and switching of multiple repair tools and materials, and can be applied to various defect types such as cracks, holes, corrosion, etc. Through integrated modular design, the system has good scalability and can adapt to more complex repair needs in the future. In addition, the present invention uses intelligent technology and self-learning mechanism to achieve the generation and optimization of dynamic repair strategies, significantly improving the repair efficiency, quality and adaptability.

[0070] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented according to the contents of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically cited below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0072] Figure 1 This is a flow chart of the adaptive power pipeline robot repair method based on defect database of this application;

[0073] Figure 2 This is a schematic diagram of the structure of the adaptive power pipeline robot repair system based on the defect database of this application. DETAILED DESCRIPTION

[0074] The purpose of the present invention is to provide an adaptive power pipeline robot repair method and system based on a defect database, which realizes the generation and optimization of dynamic repair strategies through intelligent technology and self-learning mechanism, and significantly improves the repair efficiency, quality and adaptability. Specific purposes include: (1) Improving dynamic adjustment capability: The present invention collects defect characteristics in real time through built-in multimodal sensors, and dynamically generates repair strategies in combination with defect databases, which can flexibly adjust the repair method according to parameters such as defect shape, size and material, ensuring that the repair process is efficient and accurate. (2) Improving the level of intelligence: The present invention uses machine learning algorithms (such as neural networks and reinforcement learning) to conduct in-depth analysis of defect data, optimize repair strategies, and enable robots to have adaptive capabilities, so as to achieve high-quality repair in complex defect scenarios. (3) Optimizing the use of defect database: The present invention constructs a continuously updated defect database. The robot automatically records the repair data during the repair process and feeds it back to the database for updating, so as to achieve effective accumulation of historical experience and continuous optimization of future repair strategies. (4) Realizing dynamic feedback of repair quality: After the repair is completed, the present invention uses real-time detection of the repair effect (such as surface integrity and adhesion strength) to optimize the repair strategy and ensure the reliability of the repair quality. (5) Improved adaptability and scalability: The present invention supports the dynamic selection and switching of multiple repair tools and materials, and can be applied to various defect types such as cracks, holes, corrosion, etc. Through integrated modular design, the system has good scalability and can adapt to more complex repair needs in the future.

[0075] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of 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.

[0076] First, see the attached Figure 1 This application proposes an adaptive power pipeline robot repair method based on a defect database, which aims to achieve efficient, accurate and intelligent power pipeline repair by intelligently identifying pipeline defects, dynamically adjusting repair strategies and using a continuously updated defect database. The technical solution includes the following main parts:

[0077] (1) Defect information collection and analysis

[0078] a) Defect information collection

[0079] The robot scans the surface of the pipeline through built-in multimodal sensors (such as lidar, visual camera, ultrasonic sensor, etc.) to obtain the geometric shape, size, material and other information of the defects in real time. Defect characteristics include but are not limited to:

[0080] ① Crack: length, width, depth;

[0081] ②Hole: diameter, depth;

[0082] ③Corrosion: area, depth, and distribution range.

[0083] These collected raw data will be transmitted to the data processing module for subsequent analysis.

[0084] b) Defect feature extraction and classification

[0085] Based on data processing algorithms (such as edge detection, morphological analysis, 3D reconstruction, etc.), the robot extracts the key features of the defects. The defects will be classified into different repair categories (such as cracks, holes, corrosion, etc.) according to their geometry, size and material type.

[0086] c) Matching of defect database

[0087] The defect information after feature extraction will be matched with the pre-established defect database. The defect database contains various types of defect features, repair strategies, repair materials, process parameters and other information. The database selects the repair solution that is most similar to the current defect through a similarity matching algorithm and uses it as a preliminary repair suggestion.

[0088] (2) Adaptive Generation of Repair Strategies

[0089] a) Repair tools and material selection

[0090] Based on the defect type, shape, size and material, the system will automatically select the most suitable repair tools (such as filling tools, polishing tools, curing equipment, etc.) and repair materials (such as hot melt coatings, polymer fillers, etc.). Different defect types require different repair tools and materials, for example:

[0091] For cracks, choose the appropriate filling material;

[0092] For corrosion, choose a corrosion-resistant coating material.

[0093] b) Process parameter setting

[0094] The system automatically adjusts the process parameters of the repair process, including temperature, pressure, speed, etc., according to the defect characteristics and repair materials. For different repair tasks, precise control of process parameters can ensure the repair effect. For example, the coating temperature of hot melt coating, the application pressure of filling material, and the curing temperature.

[0095] c) Repair strategy optimization

[0096] Through machine learning (such as reinforcement learning or neural network) algorithms, the system can provide real-time feedback on the repair effect during multiple repair processes and automatically optimize the repair strategy based on the feedback results. The goal of optimization is to improve the quality of repair and reduce the repair time.

[0097] (3) Execution and feedback mechanism of the repair process

[0098] a) Repair task execution

[0099] When performing repair tasks, the repair robot operates according to the selected repair strategy. The robot performs repair operations such as coating, filling, welding, polishing, etc. through a precise motion control system.

[0100] b) Repair result feedback and real-time detection

[0101] After the repair is completed, the robot detects the repair effect through built-in sensors (such as force sensors, adhesion detectors, etc.). Specific feedback includes:

[0102] Repair surface integrity: Check whether the repair area is flat and has no cracks or peeling;

[0103] Repair adhesion strength: Ensure that the repair material is firmly attached to the original pipe surface.

[0104] c) Dynamic repair strategy adjustment

[0105] If the repair effect is not ideal, the robot will automatically adjust the repair strategy based on the detected repair quality problems (such as insufficient adhesion, uneven surface, etc.) and re-execute the repair task.

[0106] (4) Defect Database and Learning System

[0107] a) Database construction and maintenance

[0108] One of the keys of the present invention is to establish a defect database containing various defect types and repair solutions. The database contains information such as historical repair data, repair strategies and repair effects. The database not only stores the basic parameters of the defects, but also includes the best repair solution and its process parameters for each defect.

[0109] b) Self-learning mechanism

[0110] When the robot performs repair tasks, it continuously feeds back the data and results of the repair process to the database. Through self-learning and data accumulation, the repair strategies in the database will be continuously optimized, allowing the robot to improve future repair plans based on historical data.

[0111] c) Continuous Optimization

[0112] As the repair tasks continue, the defect database is gradually improved, and the robot's repair capability and intelligence level are constantly improving. The robot can not only quickly generate repair strategies based on current defects, but also extract experience from historical repair data and continuously improve adaptive repair strategies.

[0113] The above solution automatically selects appropriate repair tools (such as filling, coating, and curing devices) and materials according to the characteristics of defect type, material, and location, and adaptively adjusts the repair process parameters (such as temperature, pressure, and speed) to meet the repair needs of various complex pipeline defects. The specific steps include:

[0114] Step S1: acquiring power pipeline defect data in real time through a collection component, and extracting multi-dimensional defect characteristic parameters through the defect data.

[0115] This embodiment uses a similarity matching algorithm and a machine learning algorithm to dynamically generate a repair strategy that adapts to the current defect characteristics based on the defect characteristics (including defect shape, size, material, etc.) obtained in real time, and compares the historical data and repair strategies stored in the defect database, including:

[0116] The acquisition component is set at a suitable position to accurately collect the power pipeline defect data. The acquisition component includes a laser radar sensor, a visual camera and an ultrasonic sensor. The laser radar sensor is used to scan the geometric shape of the pipeline surface to obtain high-precision three-dimensional point cloud data. The visual camera is used to capture high-definition images of the pipeline surface to assist in identifying the visual features of defects. The ultrasonic sensor is used to detect defects inside the pipeline, thereby obtaining the geometric shape, size and material information of the defects in real time.

[0117] The collected data are filtered and denoised, and multimodal data are fused to generate unified defect description data.

[0118] Wherein, extracting multi-dimensional defect characteristic parameters through the defect data includes:

[0119] Edge detection and morphological analysis are performed on the high-definition images captured by the visual camera to extract the two-dimensional features of the defects.

[0120] Specifically, in this embodiment, edge detection and morphological analysis are performed on the image captured by the visual camera to extract the two-dimensional features of defects such as cracks and holes. In this embodiment, the Canny edge detection algorithm can be used to extract the edges of cracks and holes in the image. In the morphological analysis, this embodiment can use morphological operations (dilation, corrosion) to enhance the defect features for subsequent classification.

[0121] The 3D point cloud data collected by the LiDAR is processed to build a model, and the 3D model of the corresponding defect is reconstructed to calculate the volume and depth of the defect;

[0122] In this embodiment, the three-dimensional reconstruction can use PCL (Point Cloud Library) to perform three-dimensional reconstruction of point cloud data to generate a three-dimensional model of the defect, and the volume is calculated through a volume measurement algorithm to calculate the specific size parameters of the defect.

[0123] A multidimensional feature model is constructed based on the two-dimensional features, the three-dimensional model and the distance and position information of the defect collected by the ultrasonic sensor, and the geometric shape, geometric features, material features and surface texture features of the defect are extracted from the multidimensional feature model.

[0124] Specifically, in this embodiment, a feature extraction algorithm can be used in combination with image and point cloud data to extract multi-dimensional feature parameters such as the geometric shape, size (such as length, width, depth), and material of the defect.

[0125] This application uses multiple sensor fusion technologies such as lidar, visual cameras, ultrasonic sensors, etc. to detect and analyze pipeline defects in real time, extract multi-dimensional feature parameters of defects, and provide accurate data support for subsequent adaptive repair strategy generation.

[0126] Step S2: construct the initial model of the power pipeline defect based on the multi-dimensional defect characteristic parameters and historical defect data, and perform similarity matching between the real-time power pipeline defect data and the data in the initial model of the power pipeline defect, so as to obtain the updated model of the power pipeline defect; and obtain the best repair strategy for the real-time power pipeline defect data according to the similarity matching value. In addition, the present application performs quality inspection on the repair area after the repair operation is completed, returns the repair quality information to the defect database and learning system through the feedback mechanism, adjusts and optimizes the repair strategy, and realizes the closed-loop control of quality inspection-feedback-optimization.

[0127] Step S2 of this embodiment specifically includes:

[0128] Among them, the power pipeline defect update model includes:

[0129] The power pipeline defect initial model is the historical defect data stored in the database, and each record includes: defect type, geometric features, material properties, surface texture features, and the best repair strategy corresponding to the defect;

[0130] Updating the real-time multi-dimensional defect characteristic parameters corresponding to the power pipeline defect data into the power pipeline defect initial model;

[0131] The multi-dimensional defect feature parameters are matched with a feature vector formed by the features in each record for similarity, and the repair strategy corresponding to the historical record with the highest similarity result is used as the best repair strategy for the current defect.

[0132] Further, including:

[0133] The historical defect data is stored in a database, including:

[0134] The database has a two-level index structure, including: the first-level index is to use a clustering algorithm to preliminarily classify the historical defect data according to the defect type or material characteristics, and each category forms a sub-database; the second-level index is to build a multidimensional index structure inside each sub-database, and the multidimensional index structure recursively divides the feature space, reduces the dimension of the high-dimensional features, and performs fast neighbor search.

[0135] The classification algorithm uses support vector machine (SVM): used to classify defects (such as cracks, holes, corrosion, etc.) according to the extracted feature parameters. The corresponding clustering algorithm: sub-classifies complex defects and refines the repair strategy.

[0136] In this embodiment, it includes:

[0137] The multi-dimensional defect feature parameter is matched with a feature vector formed by the features in each record by similarity, including:

[0138] The feature vector of each historical defect data record is standardized to obtain the first feature vector.

[0139] The second eigenvector is obtained by standardizing and reducing the dimensionality of the eigenvector formed by the multi-dimensional defect characteristic parameters;

[0140] Setting a weight for a corresponding feature dimension, and calculating a corresponding weighted Euclidean distance according to the first feature vector and the second feature vector;

[0141] If the repair quality is low, the weights are updated through gradient descent to continuously optimize the accuracy of similarity matching.

[0142] The step of setting the weight of the corresponding feature dimension and calculating the corresponding weighted Euclidean distance according to the first feature vector and the second feature vector includes:

[0143] Pre-classifying the second feature vector according to the defect type and selecting the corresponding sub-database;

[0144] In the multidimensional index structure of the selected sub-database, a nearest neighbor search is performed with the second eigenvector as a query point to obtain k most similar historical defect records;

[0145] Sort these k results according to the weighted distance and select the first N results with the smallest distance;

[0146] Extract the corresponding repair strategy set from the most similar top N defect records;

[0147] If the set of repair strategies is greater than 1, a voting or weighted average strategy is used to select the best repair strategy.

[0148] If the repair quality is low, the weights are updated through gradient descent to continuously optimize the accuracy of similarity matching, including:

[0149] After each repair is completed, the repair quality is evaluated. It is assumed that the best strategy corresponding to the historical defect record matched in this repair is used to repair the current defect. If the repair quality is high, it means that the feature weight distribution is reasonable; if the evaluation result is that the repair quality is low, it means that the weights of some feature dimensions in the similarity measurement are not set properly, and the weights are updated through gradient descent;

[0150] A corresponding loss function is constructed based on the updated weights, which includes the expected value of the repair quality. When the repair quality is lower than the expected value, the relevant features in the matched defect records are analyzed. If a certain type of feature has a high correlation with the repair success rate in the database, the weight of the feature of this dimension is increased; if a certain type of feature shows a low correlation with the repair success rate in multiple repairs, its corresponding weight is reduced; this update process is carried out online, and the weight is recalculated after each batch of repair tasks is completed.

[0151] Specifically, the similarity matching algorithm mentioned above includes: Euclidean distance calculation: Calculate the Euclidean distance between the current defect feature and the defect feature in the database, and select the most similar repair solution. Weighted matching: Assign different weights according to the important features of the defect (such as depth, material) to improve the accuracy of matching.

[0152] Step S3: According to the obtained optimal repair strategy, the movement path of the repair tool is planned to ensure that the repair operation covers all defective areas. The repair tool performs repair operations according to the planned path, and a repair trajectory model is constructed according to the obtained movement path and the corresponding multi-dimensional characteristic parameters of the power pipeline defect.

[0153] Step S3 of this embodiment specifically includes:

[0154] Filling tool: Automatic filling device is used to heat the filling material (such as hot melt coating) through the heating element and evenly apply it to the defect area through the nozzle. Coating equipment: Integrated spraying system can accurately control the coating thickness and uniformity, suitable for coating repair in corroded areas. Curing device: Use infrared curing lamp to quickly cure the applied repair material to ensure the quality of the repair.

[0155] Motion control system: Servo motor: Use high-precision servo motor to drive the repair tool to achieve precise position and speed control. Motion control software: Based on the ROS platform, write motion control algorithms to ensure that the repair tool moves accurately according to the planned path.

[0156] Step S4: After the repair is completed, the repair quality detection sensor performs a comprehensive inspection of the repair area. If the detection result is not ideal, the repair strategy adjustment algorithm is used based on the real-time power pipeline defect data to regenerate the optimal repair strategy, and update the power pipeline defect update model and repair trajectory model.

[0157] If the detection result is ideal, the repair trajectory in the repair trajectory model and the corresponding repair effect are updated to the power pipeline defect update model to form a historical repair case record;

[0158] The power pipeline defect update model is continuously optimized and updated using a machine learning algorithm.

[0159] In this embodiment, the machine learning algorithm includes: Neural network: construct a multi-layer perceptron (MLP) model, input defect features, and output optimized repair strategy parameters. Reinforcement learning: optimize the repair strategy through a reward mechanism, so that the robot can self-learn and improve in continuous repair tasks.

[0160] Therefore, the specific working principle and operation steps of step S4 in this embodiment include:

[0161] Quality Inspection: After the repair is completed, the repair quality inspection sensor conducts a comprehensive inspection of the repair area to evaluate the surface integrity, adhesion strength and material properties of the repair.

[0162] Feedback analysis: The feedback control unit analyzes the test data to determine whether the repair quality meets the standards.

[0163] Strategy adjustment: If the detection result is not ideal, the repair strategy adjustment algorithm regenerates the repair strategy based on the feedback information, adjusts the repair parameters (such as temperature and pressure), and instructs the repair execution module to re-perform the repair operation.

[0164] Step S5: The above operations are executed repeatedly to identify and repair more power pipeline defects, and the repair strategy is continuously optimized.

[0165] Specifically, the operation process of this method includes:

[0166] (1) Startup and initialization

[0167] After the robot is started, it performs a self-check to ensure that all sensors and actuators are working properly, initializes the defect database and learning system, and loads the latest repair strategy.

[0168] (2) Defect information collection

[0169] The robot moves inside or on the surface of the pipeline, scans the pipeline through the sensor system of the defect information collection module, and obtains defect data in real time.

[0170] (3) Defect identification and feature extraction

[0171] The collected raw data is transmitted to the defect recognition and feature extraction module to extract the defect’s geometric shape, size, material and other characteristic parameters and classify them.

[0172] (4) Repair strategy generation

[0173] According to the extracted defect features, the repair strategy generation module uses a similarity matching algorithm to find the most suitable repair solution in the defect database. The machine learning algorithm is used to optimize the repair strategy and generate an adaptive repair solution.

[0174] (5) Repair execution

[0175] The repair execution module selects appropriate repair tools and materials according to the generated repair strategy, plans the repair path and performs the repair operation.

[0176] (6) Repair quality inspection

[0177] After the repair is completed, the repair quality inspection and feedback module conducts a comprehensive inspection of the repair area and evaluates the repair effect.

[0178] (7) Feedback and Optimization

[0179] Feedback the inspection results to the defect database and learning system to record the repair data. The learning and optimization module analyzes the feedback information, optimizes the repair strategy, and improves the effectiveness of future repair tasks.

[0180] (8) Loop execution

[0181] The robot continues to scan the pipeline as needed, identifying and repairing more defects. The entire process forms a closed loop and continuously optimizes the repair strategy.

[0182] On the other hand, the present invention also provides an adaptive power pipeline robot repair system based on a defect database, wherein each module transmits and communicates data through an internal data bus to ensure coordinated operation of the system. Figure 2 The system comprises:

[0183] The acquisition module is used to acquire the power pipeline defect data in real time through the acquisition component, and extract multi-dimensional defect characteristic parameters through the defect data.

[0184] Specifically, the acquisition module includes: a defect information acquisition module and a defect recognition and feature extraction module. The defect information acquisition module is used to obtain pipeline defect data through sensors and transmit the data to the defect recognition and feature extraction module through an internal data bus.

[0185] The defect identification and feature extraction module is used to process and analyze the data, extract defect feature parameters, and pass these feature information to the repair strategy generation module through the internal data bus.

[0186] The defect information collection module in this embodiment is implemented by the following contents:

[0187] Defect information collection module components:

[0188] (1) LiDAR sensor: used to scan the geometric shape of the pipeline surface and obtain high-precision three-dimensional point cloud data.

[0189] (2) Visual camera: used to capture high-definition images of the pipeline surface and assist in identifying the visual features of defects.

[0190] (3) Ultrasonic sensor: used to detect defects inside the pipeline, such as internal corrosion, voids, etc.

[0191] (4) Data processing unit: pre-processes the collected raw data, including filtering, noise reduction and data fusion.

[0192] Working principle and operation steps:

[0193] (1) Data acquisition: The sensor system (lidar, visual camera, ultrasonic sensor) carried by the robot simultaneously scans the surface and interior of the pipeline to obtain the geometric shape, size and material information of the defects in real time.

[0194] (2) Data preprocessing: The data processing unit filters and reduces noise on the original data, and fuses the multimodal data to generate unified defect description data.

[0195] Defect recognition and feature extraction module components:

[0196] (1) Image processing submodule: responsible for edge detection, morphological analysis and other processing on the images collected by the visual camera to extract the two-dimensional features of the defects.

[0197] (2) Point cloud processing submodule: responsible for processing the 3D point cloud data collected by the lidar, performing 3D reconstruction and defect volume calculation.

[0198] (3) Feature extraction algorithm: including shape recognition, dimension measurement, material analysis and other algorithms to extract key feature parameters of defects.

[0199] Working principle and operation steps:

[0200] (1) Image analysis: The image processing submodule performs edge detection and morphological analysis on the images captured by the visual camera to extract the two-dimensional features of defects such as cracks and holes.

[0201] (2) Point cloud analysis: The point cloud processing submodule processes the three-dimensional point cloud data collected by the lidar, reconstructs the three-dimensional model of the defect, and calculates the volume and depth of the defect.

[0202] (3) Feature extraction: The feature extraction algorithm combines image and point cloud data to extract multi-dimensional feature parameters such as the defect’s geometric shape, size (such as length, width, depth), and material.

[0203] The present invention not only improves the accuracy and reliability of defect identification in data fusion algorithm and feature extraction, but also provides high-quality input data for subsequent adaptive repair strategy generation. It also organically combines defect feature extraction with subsequent adaptive repair strategy generation and quality feedback optimization to form a closed-loop system.

[0204] The repair strategy generation module is used to construct the power pipeline defect initial model based on the multi-dimensional defect feature parameters and historical defect data, and to perform similarity matching between the real-time power pipeline defect data and the data in the power pipeline defect initial model, so as to obtain the power pipeline defect update model; and to obtain the best repair strategy for the real-time power pipeline defect data according to the similarity matching value. That is, by accessing the defect database and learning system to obtain historical repair experience and strategies, matching and optimizing the current defect features, and then generating the corresponding repair plan, which is sent to the repair execution module through the internal data bus.

[0205] The repair strategy generation module in this embodiment specifically includes the following parts:

[0206] (1) Defect database: stores the characteristic parameters, optimal repair strategies, repair materials and process parameters of various defects.

[0207] (2) Similarity matching algorithm: used to match the current defect with the known defects in the database and determine the most suitable repair strategy.

[0208] 1. Hierarchical database structure:

[0209] The defect database stores a large amount of historical defect data. Each record contains the defect type, geometric features (length, width, depth), material properties (conductivity, stress distribution parameters), surface texture features (edge ​​strength extracted based on image processing, local binary pattern LBP features), and the best repair strategy for the defect (including the type of repair material, heating temperature, filling pressure, curing time, etc.).

[0210] In order to improve the retrieval efficiency and matching accuracy, the present invention adopts a two-level index structure for the database:

[0211] (a) Primary clustering index: First, use clustering algorithms such as K-Means to group existing defect samples according to defect type or material characteristics. For example, crack type defects, corrosion type defects, and hole type defects are preliminarily classified, and each type forms a sub-database.

[0212] (b) Secondary spatial index structure: A multidimensional index structure (such as KD-Tree) is constructed inside each sub-database. KD-Tree recursively divides the feature space and reduces the dimension of high-dimensional features for fast nearest neighbor search.

[0213] 2. Feature standardization and dimensionality reduction:

[0214] Standardize the feature vector F_db(i) of each defect record:

[0215]

[0216] Among them, μ j and σ j They are the mean and standard deviation of feature dimension j in historical data. Through standardization, the influence of features of different dimensions and magnitudes is eliminated.

[0217] In addition, when the feature dimension is too high, PCA (Principal Component Analysis) can be used to reduce the dimension of the feature, and the first few principal components with cumulative variance contribution rate of more than 90% can be selected to construct the KD-Tree index. This dimensionality reduction process can significantly improve the retrieval speed.

[0218] 3. Similarity measures and algorithm steps

[0219] (1) Definition of similarity distance:

[0220] The current defect feature vector is recorded as F_current, which is processed by the same standardization and dimension reduction to obtain F_current'. The database record is F_db(i)'. Define the weighted Euclidean distance:

[0221]

[0222] Among them, m is the number of features after dimensionality reduction, w j is the weight of feature dimension j. These initial weight values ​​can be obtained by expert experience or by using linear regression or neural network training in a special training set.

[0223] (2) Adaptive weight update and feedback mechanism:

[0224] After each repair is completed, the present invention evaluates the repair quality Q (determined based on comprehensive indicators such as adhesion, surface smoothness, electrical performance scoring, etc.). Suppose the historical defect record matched in this repair is F_db(k), and its corresponding strategy S_k is used to repair the current defect. If the repair quality Q is high, it indicates that the feature weight allocation is reasonable; if Q is low, it means that the weight settings of some feature dimensions in the similarity metric are inappropriate. The present invention updates w through gradient descent j as follows:

[0225]

[0226] where the loss function L can be defined as:

[0227]

[0228] When Q < Q_ideal (lower than the expected value), analyze the relevant features in the matched defect records. If a certain type of feature (such as the material conductivity feature) has a high correlation with the repair success rate in the database, moderately increase the weight w of this dimension of the feature j ; if a certain feature (such as the texture feature) shows unimportance or misleadingness in multiple repairs, reduce its weight.

[0229] This update process is carried out online. After each batch of repair tasks (such as N repairs) is completed, the weights are recalculated to continuously optimize the accuracy of similarity matching.

[0230] 4. Retrieval and Matching Process

[0231] (1) Retrieval Process:

[0232] (a) Input: The multi-modal fusion feature F_current of the current defect data.

[0233] (b) Preprocessing: Standardize F_current and multiply it by the PCA matrix to achieve dimensionality reduction, obtaining F_current'.

[0234] (c) First-level Classification: Pre-classify F_current' according to the defect type and select the corresponding sub-database (such as the crack class sub-database).

[0235] (d) Second-level Index Search: In the KD-Tree of the selected sub-database, perform a nearest neighbor search (k-NN query) with F_current' as the query point to obtain k most similar historical defect records F_db(i)'.

[0236] (e) Similarity Ranking: Sort these k results according to the weighted distance d(F_current', F_db(i)') and select the top N results with the smallest distance.

[0237] (2) Strategy selection and comprehensive decision-making:

[0238] Extract the corresponding repair strategy set S = {S_1, S_2, ..., S_N} from the most similar top N defect records.

[0239] If N>1, a voting or weighted average strategy can be used to select the final repair solution. For example, each strategy is weighted according to its historical repair success rate and average repair quality score, and the strategy S_best with the highest expected repair effect is selected.

[0240] 5. Feedback and Updates:

[0241] After the repair is completed, the matching error and feature weight change trend corresponding to the repair effect Q and strategy S_best are recorded. This information is sent back to the database for the next round of weight update and strategy optimization to gradually improve the performance of the matching algorithm.

[0242] This embodiment makes special designs and improvements in the selection of characteristic parameters and the matching degree calculation method.

[0243] (a) Multi-dimensional feature input: The similarity matching of the present invention is not only based on simple single parameters such as length and width, but is based on multi-dimensional and deep feature parameters obtained through multimodal data fusion (such as defect shape complexity, material properties, corrosion depth distribution, stress concentration degree, etc.).

[0244] (b) Adaptive weight allocation: In the present invention, the weight setting for each feature parameter is not fixed, but can be automatically optimized through a machine learning model, dynamically adjusting the feature weight under long-term data accumulation to achieve more accurate similarity matching. Traditional technologies generally use fixed or empirically set similarity calculation methods, lacking such dynamic optimization and weight update mechanisms.

[0245] The closed-loop combination of similarity matching and learning system of the present invention specifically includes:

[0246] The present invention deeply couples similarity matching with continuous learning optimization of the defect database. The similarity matching result is not only used in one task, but the database and matching algorithm are continuously updated through subsequent repair feedback data.

[0247] (a) Continuously optimized matching model: Traditional technologies use fixed matching rules and algorithms in similarity matching. The present invention uses machine learning and reinforcement learning to continuously adjust the matching algorithm based on historical repair results, making the matching results more relevant and instructive to the actual repair effect.

[0248] (b) Dynamic evolution of database: Different from the prior art that relies on static feature templates, the defect database of the present invention is dynamically updated, and the matching algorithm continues to improve its accuracy and practicality under the continuous enrichment of training data, so that similarity matching is no longer a simple call to existing methods, but a core link throughout the entire system self-learning cycle.

[0249] The present invention is linked with specific repair strategies and process parameters, specifically including:

[0250] The ultimate goal of similarity matching is to assist in generating adaptive repair strategies, rather than simply classifying or sorting. The present invention considers the correspondence between defects and repair solutions during the matching process, that is, similarity matching not only matches "similar defects", but also directly relates to "corresponding optimal repair solutions and their parameters".

[0251] (a) Personalized matching indicators: The matching process introduces considerations of subsequent repair process parameters, material selection, tool adaptability, etc., so that the similarity matching results can be directly used for repair strategy generation. Traditional similarity matching is mostly limited to the comparison of feature similarity. The present invention incorporates effectiveness indicators such as the applicability of executable strategies and historical repair success rates into the matching calculation, which far exceeds the level of traditional pure feature matching.

[0252] (b) Dynamic evolution of database: Different from the prior art that relies on static feature templates, the defect database of the present invention is dynamically updated, and the matching algorithm continues to improve its accuracy and practicality under the continuous enrichment of training data, so that similarity matching is no longer a simple call to existing methods, but a core link throughout the entire system self-learning cycle.

[0253] The repair execution module is used to plan the movement path of the repair tool according to the obtained optimal repair strategy to ensure that the repair operation covers all defective areas. The repair tool performs repair operations according to the planned path and constructs a repair trajectory model based on the obtained movement path and the corresponding multi-dimensional characteristic parameters of the power pipeline defect.

[0254] The repair execution module of this embodiment specifically includes the following contents:

[0255] (1) Repair tool system: including filling tools, coating equipment, curing devices, etc., which perform specific operations according to the repair strategy.

[0256] (2) Motion control system: controls the motion path and operation accuracy of the repair tool.

[0257] (3) Energy supply system: Provides the required energy for repair tools and control systems.

[0258] Working principle and operation steps:

[0259] (1) Tool configuration: Select the appropriate repair tool according to the repair strategy (such as filling tools for crack repair and coating equipment for corrosion repair).

[0260] (2) Path planning: The motion control system plans the motion path of the repair tool based on the geometry and location of the defect to ensure that the repair operation covers all defect areas.

[0261] (3) Perform repair: The repair tool operates according to the planned path, applies the repair material (such as hot-melt coating, polymer filler) and performs a curing process to complete the repair of the defect.

[0262] The repair quality detection and feedback module is used to perform a comprehensive inspection of the repair area by the repair quality detection sensor after the repair is completed. If the inspection result is not ideal, the repair strategy adjustment algorithm is used based on the real-time power pipeline defect data to regenerate the optimal repair strategy, and update the power pipeline defect update model and repair trajectory model.

[0263] Specifically, the repair quality detection and feedback module in this embodiment specifically includes the following contents:

[0264] (1) Repair quality detection sensors: including force sensors, adhesion force detectors, surface smoothness detectors, etc., used to detect the repair effect.

[0265] (2) Feedback control unit: receives data from detection sensors, evaluates repair quality and generates feedback information.

[0266] (3) Repair strategy adjustment algorithm: Adjust the repair strategy based on feedback information and optimize subsequent repair operations.

[0267] Working principle and operation steps:

[0268] (1) Quality inspection: After the repair is completed, the repair quality inspection sensor conducts a comprehensive inspection of the repair area to evaluate the surface integrity, adhesion strength and material properties of the repair.

[0269] (2) Feedback analysis: The feedback control unit analyzes the test data to determine whether the repair quality meets the standards.

[0270] (3) Strategy adjustment: If the detection result is not ideal, the repair strategy adjustment algorithm regenerates the repair strategy based on the feedback information, adjusts the repair parameters (such as temperature and pressure), and instructs the repair execution module to re-perform the repair operation.

[0271] Specifically, the technical implementation of the repair quality detection and feedback module includes:

[0272] Repair quality detection sensors: ① Force sensor: measures the force of the repair material to ensure the mechanical properties during filling and coating. ② Adhesion detector: detects the adhesion strength between the repair material and the pipe surface to prevent material peeling. ③ Surface smoothness detector: detects the surface smoothness of the repair area through laser scanning or tactile sensor to ensure that there are no obvious defects on the repaired surface.

[0273] Feedback control unit: ① Data analysis module: Analyze the data of the detection sensor and evaluate whether the repair quality meets the standard. ② Decision module: Based on the analysis results, decide whether to adjust the repair strategy and re-execute the repair task.

[0274] Repair strategy adjustment algorithm: ① Adaptive adjustment: According to the feedback results, dynamically adjust the repair parameters (such as temperature, pressure, speed) to optimize the repair process. ② Re-repair instructions: If the detection results are not ideal, generate new repair instructions to guide the repair execution module to re-perform the repair operation.

[0275] The defect database and learning module are used to store and manage historical data of various defect types and repair strategies, and continuously optimize and update the repair strategies through machine learning algorithms to provide better decision support for the next repair task. The present invention constructs and uses a continuously updated defect database to record and analyze the repair results, and self-learns and optimizes through the continuously accumulated historical data and feedback information, so that the repair strategy can be continuously improved as the repair experience is enriched.

[0276] Specifically, the defect database and learning system in this embodiment comprises the following components:

[0277] (1) Database management system: used to store and manage defect data and repair strategies.

[0278] (2) Data collection module: collects detailed data of each repair operation, including defect characteristics, repair strategies, repair effects, etc.

[0279] (3) Learning and optimization module: Use machine learning algorithms to analyze and optimize data in the database to improve the intelligence level of the repair strategy.

[0280] Working principle and operation steps:

[0281] (1) Data recording: After each repair operation, the repair quality detection and feedback module feeds back the repair data (defect characteristics, repair strategy, and repair effect) to the data acquisition module.

[0282] (2) Data storage: The database management system stores the collected data in the defect database to form a historical repair case record.

[0283] (3) Data analysis and learning: The learning and optimization module analyzes the stored data, identifies the effectiveness of the repair strategy, and uses machine learning algorithms to continuously optimize and update the repair strategy database.

[0284] (4) Continuous optimization: Through continuous learning and optimization, the repair strategies in the defect database are gradually improved, enhancing the system's adaptive and intelligent capabilities.

[0285] In this embodiment, the technical implementation of the defect database and the learning module includes:

[0286] Database management system: ① Data storage: Use a relational database (such as MySQL) to store defect data, repair strategies, and repair results. ② Data indexing: Establish an efficient indexing mechanism to speed up defect matching and strategy retrieval.

[0287] Data acquisition module: ① Data recording: Automatically record detailed data of each repair operation, including defect characteristics, repair strategy, repair effect, etc. ② Data synchronization: Synchronize data to the cloud database through wireless communication (such as Wi-Fi, 4G / 5G) to achieve centralized data management and backup.

[0288] Learning and optimization module: ① Data analysis: Use big data analysis technology to conduct in-depth analysis of stored data to discover the pros and cons of repair strategies. ② Model training: Based on the collected repair data, train machine learning models to improve the intelligence level of repair strategies. ③ Strategy update: Update the optimized repair strategy to the defect database for reference in subsequent repair tasks.

[0289] To facilitate understanding of the above solution, this embodiment is given as follows:

[0290] Crack repair

[0291] (1) Startup and initialization

[0292] After the robot starts, it performs a self-check to ensure that the lidar, visual camera, and ultrasonic sensor are functioning properly. The defect database and learning system are loaded with the latest repair strategies.

[0293] (2) Defect information collection

[0294] The robot moves across the pipe surface, scanning with lidar and vision cameras to identify linear cracks with a length of 20 mm and a width of 2 mm.

[0295] (3) Defect identification and feature extraction

[0296] The image processing submodule extracts the crack edge using the Canny edge detection algorithm. The point cloud processing submodule reconstructs the three-dimensional model of the crack and calculates the crack depth to be 3 mm.

[0297] (4) Repair strategy generation

[0298] The similarity matching algorithm finds similar thermal stress crack repair cases in the defect database. The machine learning algorithm optimizes the repair strategy and determines the use of high-strength hot-melt filling material, sets the filling temperature to 200°C, and the filling path is to gradually expand from the crack center to both sides.

[0299] (5) Repair execution

[0300] The repair execution module selects the filling tool and evenly applies the hot melt filling material through the nozzle. The motion control system controls the repair tool to move along the crack path and applies pressure to ensure uniform filling of the material.

[0301] (6) Repair quality inspection and feedback

[0302] The repair quality detection sensor detects the adhesion strength of the filling area and finds that some areas are not sufficiently adhered. The feedback control unit analyzes the results and instructs the repair strategy to adjust, increase the filling temperature to 220°C, and re-execute the repair operation.

[0303] (7) Strategy Optimization

[0304] After the repair is completed, the repair quality detection sensor confirms that the repair effect meets the standard. The data acquisition module records the repair data, and the learning and optimization module updates the repair strategy in the defect database.

[0305] Corrosion repair

[0306] (1) Startup and initialization

[0307] After the robot is started, it completes self-inspection and loads the defect database and learning system.

[0308] (2) Defect information collection

[0309] The robot scanned with ultrasonic sensors and detected a circular corrosion area with a diameter of 50 mm, which was made of carbon steel.

[0310] (3) Defect identification and feature extraction

[0311] The image processing submodule extracts the edge of the corrosion area through morphological analysis. The point cloud processing submodule calculates the area of ​​the corrosion area as 200 square millimeters and the depth as 5 millimeters.

[0312] (4) Repair strategy generation

[0313] The similarity matching algorithm found similar carbon steel corrosion repair cases in the defect database. The machine learning algorithm optimized the repair strategy and determined that the corrosion-resistant coating should be applied in layers, with the first layer of coating thickness being 0.5 mm and the second layer being 0.8 mm, and the curing temperature set to 180°C.

[0314] (5) Repair execution

[0315] The repair execution module selects the coating equipment and evenly applies the first layer of corrosion-resistant coating through the spray system. The curing device starts the infrared curing lamp to cure the first layer of coating. Repeat the above steps to apply the second layer of coating and perform curing.

[0316] (6) Repair quality inspection and feedback

[0317] The surface smoothness tester detects the coating thickness and uniformity to confirm that it meets the set standards. The data acquisition module records the repair data, and the learning and optimization module updates the repair strategy to optimize the coating thickness and curing parameters for subsequent similar repair tasks.

[0318] In summary, the adaptive power pipeline robot repair method and system based on defect database provided by this application have the following technical effects:

[0319] (1) Significant improvement in intelligence and adaptive capabilities

[0320] Disadvantages of existing technologies: Existing power pipeline repair robots usually adopt fixed repair strategies and cannot dynamically adjust the repair methods according to the specific characteristics of different defects (such as shape, size, and material). This results in unstable repair effects when facing complex and diverse defects, making it difficult to meet the needs of high-quality repair.

[0321] Advantages of the invention: The invention can automatically identify the specific characteristics of defects and dynamically generate the optimal repair strategy through an adaptive repair strategy generation method based on defect characteristics, combined with multimodal sensor data and machine learning algorithms. In this way, the robot can flexibly respond to various complex defects and ensure the accuracy and consistency of each repair operation.

[0322] (2) Efficient data utilization and continuous optimization

[0323] Disadvantages of existing technologies: Although existing repair robots may be equipped with defect databases, these databases are often static and lack real-time updating and self-learning capabilities. The selection of repair strategies depends on pre-set rules and cannot be optimized and improved through actual repair data.

[0324] Advantages of the present invention: The present invention introduces a defect database and learning system, continuously records the data and results of each repair operation, and uses machine learning algorithms to analyze and optimize the database. This enables the repair strategy to be continuously optimized based on historical data and actual results, improves the system's self-learning ability and intelligence level, and significantly improves the repair efficiency and quality.

[0325] (3) Multimodal sensor fusion enhances defect recognition accuracy

[0326] Disadvantages of existing technology: Repair robots in existing technology usually rely on a single type of sensor (such as a camera or lidar) for defect detection, resulting in incomplete recognition of defect features, affecting the accuracy of the repair strategy.

[0327] Advantages of the invention: The invention uses multimodal sensor fusion technology, including laser radar, visual camera and ultrasonic sensor, to comprehensively collect the geometric shape, size and material information of the defect. By fusing the data of different sensors, the accuracy of defect identification and feature extraction is improved, providing reliable data support for the subsequent generation of adaptive repair strategies.

[0328] (4) Real-time repair quality detection and closed-loop feedback control

[0329] Disadvantages of existing technology: After completing the repair operation, the existing repair robots lack an effective quality inspection and feedback mechanism, and are unable to promptly discover and correct deficiencies in the repair, resulting in uneven repair quality and difficulty in achieving high-standard repair results.

[0330] Advantages of the present invention: The present invention designs a repair quality detection and feedback module. After the repair is completed, a variety of sensors such as force sensors, adhesion detectors and surface smoothness detectors are used to comprehensively evaluate the repair effect. The detection results are transmitted to the defect database and learning system through the feedback control unit to achieve real-time adjustment and optimization of the repair strategy. This closed-loop control mechanism ensures the quality of each repair operation and significantly improves the reliability and consistency of the repair.

[0331] (5) Highly modular design and strong system scalability

[0332] Disadvantages of existing technologies: Existing repair robots often lack modularity in design, making it difficult to flexibly configure and expand them according to different repair needs, which limits their scope of application and adaptability.

[0333] Advantages of the invention: The invention adopts a modular system architecture, and each functional module (such as defect information collection module, repair execution module, etc.) can be independently upgraded and replaced, which is convenient for system expansion and maintenance. At the same time, the system supports dynamic selection and switching of multiple repair tools and materials, which enhances the robot's adaptability to diverse defects and expands its application scenarios and service life.

[0334] (6) Improve repair efficiency and reduce operating costs

[0335] Disadvantages of existing technologies: Existing repair methods rely on manual operations or require frequent manual intervention, resulting in low repair efficiency and high labor costs. In addition, due to unstable repair quality, multiple repair operations may be required, further increasing operating costs.

[0336] Advantages of the present invention: The present invention significantly reduces the need for manual intervention and improves repair efficiency through automated defect identification, strategy generation and execution. At the same time, the continuously optimized repair strategy reduces the probability of repair failure and rework, and reduces operating costs. Overall, the present invention can complete more repair tasks in a shorter time, improving the economy and sustainability of power pipeline maintenance.

[0337] (7) Wide applicability and good environmental adaptability

[0338] Disadvantages of existing technology: Existing repair robots have poor adaptability when facing power pipelines of different types and materials, and it is difficult for them to work stably under various environmental conditions.

[0339] Advantages of the invention: The invention can adapt to different types and materials of power pipelines through adaptive repair strategies and multimodal sensors, and has good environmental adaptability. In addition, the robot design takes into account special environmental requirements such as high voltage and corrosion resistance, enabling it to operate stably under various complex working conditions, expanding its application range.

[0340] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0341] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalent technology, the present application is also intended to include these modifications and variations.

Claims

1. An adaptive power pipeline robot repair method based on defect database, characterized in that: The method includes: Acquire power pipeline defect data in real time through a collection component, and extract multi-dimensional defect characteristic parameters through the defect data; The power pipeline defect initial model is constructed based on the multi-dimensional defect characteristic parameters and the historical defect data, and the real-time power pipeline defect data is similarly matched with the data in the power pipeline defect initial model, so as to obtain the power pipeline defect update model; and the optimal repair strategy for the real-time power pipeline defect data is obtained according to the similarity matching value; According to the obtained optimal repair strategy, the movement path of the repair tool is planned to ensure that the repair operation covers all defect areas. The repair tool performs repair operations according to the planned path, and a repair trajectory model is constructed according to the obtained movement path and the corresponding multi-dimensional characteristic parameters of the power pipeline defect; After the repair is completed, the repair quality detection sensor performs a comprehensive inspection of the repair area. If the inspection result is not ideal, the repair strategy adjustment algorithm is used based on the real-time power pipeline defect data to regenerate the best repair strategy, and update the power pipeline defect update model and repair trajectory model; Repeat the above steps to identify and repair more power pipeline defects and continuously optimize the repair strategy.

2. The adaptive power pipeline robot repair method based on defect database according to claim 1 is characterized in that: The real-time acquisition of power pipeline defect data by the acquisition component includes: The acquisition component is set at a suitable position to accurately collect the power pipeline defect data. The acquisition component includes a laser radar sensor, a visual camera and an ultrasonic sensor. The laser radar sensor is used to scan the geometric shape of the pipeline surface to obtain high-precision three-dimensional point cloud data. The visual camera is used to capture high-definition images of the pipeline surface to assist in identifying the visual features of defects. The ultrasonic sensor is used to detect defects inside the pipeline, thereby obtaining the geometric shape, size and material information of the defects in real time. The collected data are filtered and denoised, and multimodal data are fused to generate unified defect description data.

3. The adaptive power pipeline robot repair method based on defect database according to claim 2 is characterized in that: The extracting multi-dimensional defect characteristic parameters through the defect data includes: Performing edge detection and morphological analysis on the high-definition image captured by the visual camera to extract two-dimensional features of the defect; The 3D point cloud data collected by the LiDAR is processed to build a model, and the 3D model of the corresponding defect is reconstructed to calculate the volume and depth of the defect; A multidimensional feature model is constructed based on the two-dimensional features, the three-dimensional model and the distance and position information of the defect collected by the ultrasonic sensor, and the geometric shape, geometric features, material features and surface texture features of the defect are extracted from the multidimensional feature model.

4. The adaptive power pipeline robot repair method based on defect database according to claim 1 is characterized in that: The power pipeline defect update model includes: The power pipeline defect initial model is the historical defect data stored in the database, and each record includes: defect type, geometric features, material properties, surface texture features, and the best repair strategy corresponding to the defect; Updating the real-time multi-dimensional defect characteristic parameters corresponding to the power pipeline defect data into the power pipeline defect initial model; The multi-dimensional defect feature parameters are matched with a feature vector formed by the features in each record for similarity, and the repair strategy corresponding to the historical record with the highest similarity result is used as the best repair strategy for the current defect.

5. The adaptive power pipeline robot repair method based on defect database according to claim 4 is characterized in that: The historical defect data is stored in a database, including: The database has a two-level index structure, including: the first-level index is to use a clustering algorithm to preliminarily classify the historical defect data according to the defect type or material characteristics, and each category forms a sub-database; the second-level index is to build a multidimensional index structure inside each sub-database, and the multidimensional index structure recursively divides the feature space, reduces the dimension of the high-dimensional features, and performs fast neighbor search.

6. The adaptive power pipeline robot repair method based on defect database according to claim 4 is characterized in that: The multi-dimensional defect feature parameter is matched with a feature vector formed by the features in each record by similarity, including: The feature vector of each historical defect data record is standardized to obtain the first feature vector. The second eigenvector is obtained by standardizing and reducing the dimensionality of the eigenvector formed by the multi-dimensional defect characteristic parameters; Setting a weight for a corresponding feature dimension, and calculating a corresponding weighted Euclidean distance according to the first feature vector and the second feature vector; If the repair quality is low, the weights are updated through gradient descent to continuously optimize the accuracy of similarity matching.

7. The adaptive power pipeline robot repair method based on defect database according to claim 6 is characterized in that: The step of setting the weight of the corresponding feature dimension and calculating the corresponding weighted Euclidean distance according to the first feature vector and the second feature vector includes: Pre-classifying the second feature vector according to the defect type and selecting the corresponding sub-database; In the multidimensional index structure of the selected sub-database, a nearest neighbor search is performed with the second eigenvector as a query point to obtain k most similar historical defect records; Sort these k results according to the weighted distance and select the first N results with the smallest distance; Extract the corresponding repair strategy set from the most similar top N defect records; If the set of repair strategies is greater than 1, a voting or weighted average strategy is used to select the best repair strategy.

8. The adaptive power pipeline robot repair method based on defect database according to claim 6 is characterized in that: If the repair quality is low, the weights are updated by gradient descent to continuously optimize the accuracy of similarity matching, including: After each repair is completed, the repair quality is evaluated. It is assumed that the best strategy corresponding to the historical defect record matched in this repair is used to repair the current defect. If the repair quality is high, it means that the feature weight distribution is reasonable; if the evaluation result is that the repair quality is low, it means that the weights of some feature dimensions in the similarity measurement are not set properly, and the weights are updated through gradient descent; A corresponding loss function is constructed based on the updated weights, which includes the expected value of the repair quality. When the repair quality is lower than the expected value, the relevant features in the matched defect records are analyzed. If a certain type of feature has a high correlation with the repair success rate in the database, the weight of the feature of this dimension is increased; if a certain type of feature shows a low correlation with the repair success rate in multiple repairs, its corresponding weight is reduced; this update process is carried out online, and the weight is recalculated after each batch of repair tasks is completed.

9. The adaptive power pipeline robot repair method based on defect database according to claim 1 is characterized in that: The method further includes: If the detection result is ideal, the repair trajectory in the repair trajectory model and the corresponding repair effect are updated to the power pipeline defect update model to form a historical repair case record; The power pipeline defect update model is continuously optimized and updated using a machine learning algorithm.

10. An adaptive power pipeline robot repair system based on defect database, characterized in that: The system includes: An acquisition module is used to acquire power pipeline defect data in real time through an acquisition component, and extract multi-dimensional defect characteristic parameters through the defect data; A repair strategy generation module is used to construct the power pipeline defect initial model based on the multi-dimensional defect characteristic parameters and historical defect data, and perform similarity matching between the real-time power pipeline defect data and the data in the power pipeline defect initial model, so as to obtain the power pipeline defect update model; and obtain the best repair strategy for the real-time power pipeline defect data according to the similarity matching value; A repair execution module is used to plan the movement path of the repair tool according to the obtained optimal repair strategy to ensure that the repair operation covers all defect areas. The repair tool performs repair operations according to the planned path, and constructs a repair trajectory model according to the obtained movement path and the corresponding multi-dimensional characteristic parameters of the power pipeline defect; A repair quality detection and feedback module is used to perform a comprehensive inspection of the repair area by the repair quality detection sensor after the repair is completed. If the inspection result is not ideal, the repair strategy adjustment algorithm is used based on the real-time power pipeline defect data to regenerate the best repair strategy, and update the power pipeline defect update model and repair trajectory model; The iteration module is used to cyclically execute the operations of the above modules, identify and repair more power pipeline defects, and continuously optimize the repair strategy.

Images