Intelligent leak detection method and device coordinated by AI and robot, equipment and medium

Abstract

The application discloses an intelligent leak detection method and device cooperating with AI and robots, equipment and a medium, and belongs to the field of intelligent equipment manufacturing. The method comprises the following steps: receiving shape data of a product to be detected collected by a laser sensor, comparing the shape data with a product model database, and generating an optimal hole position layout scheme; generating an initial optimal detection path according to a pre-stored leakage knowledge graph, the shape data of the product to be detected and historical leakage data; calculating a comprehensive confidence score based on real-time detection data and a pre-set dynamic fusion algorithm; dynamically adjusting the initial optimal detection path according to the score and a pre-set dynamic adjustment strategy; comparing the detection result with actual maintenance data, updating parameters of a model and the leakage knowledge graph through a self-learning mechanism, and the model being a model corresponding to the dynamic fusion algorithm. The detection efficiency can be improved, and the detection has adaptability.

Description

Technical Field

[0001] This invention relates to the field of robot intelligent control technology, and in particular to an intelligent leak detection method, device, equipment and medium that integrates AI and robots. Background Technology

[0002] In current industrial production, especially for products with high sealing requirements, leak detection is a crucial step in ensuring product quality and safety. Traditional leak detection methods rely on manual judgment or testing equipment with fixed parameters, resulting in low detection efficiency, high false positive rates, and poor adaptability. Particularly when dealing with products of different models and complex structures, traditional methods struggle to achieve rapid switching and accurate detection, leading to increased testing costs and extended product delivery cycles.

[0003] Furthermore, existing leak detection systems suffer from severe data silos. Historical fault data, real-time detection data, and environmental parameters are not effectively integrated and analyzed, making it difficult to comprehensively reflect the leakage characteristics under complex operating conditions and effectively identify high-risk leak points, resulting in a high false alarm rate. Detection path planning is not intelligent enough, lacking real-time feedback and strategy adjustment capabilities, leading to poor detection results and low efficiency. Summary of the Invention

[0004] This invention provides an intelligent leak detection method, device, equipment, and medium that integrates AI and robots, aiming to solve the problems of low detection efficiency, high false detection rate, and poor adaptability in existing technologies.

[0005] In a first aspect, embodiments of the present invention provide an intelligent leak detection method based on AI and robot collaboration, applied to an intelligent leak detection system. The leak detection system includes a leak detection enclosure, a leak detection robot, and a control unit. The leak detection enclosure has a plurality of detection holes and detection hole assemblies for closing or opening the detection holes. The product to be tested is placed inside the leak detection enclosure. A laser sensor for scanning the product's shape is installed inside the leak detection enclosure. The laser sensor, the detection hole assemblies, and the leak detection robot are communicatively connected to the control unit. The leak detection method includes: The system receives the shape data of the product to be inspected collected by the laser sensor, compares the shape data with the product model database, generates an optimal hole layout scheme, and sends an opening command to the detection hole assembly according to the optimal hole layout scheme. Based on the pre-stored leakage knowledge graph, the shape data of the product to be inspected, and historical leakage data, an initial optimal detection path is generated, and a detection command is sent to the leak detection robot based on the optimal detection path. During the detection process, a comprehensive confidence score is calculated based on real-time detection data and a preset dynamic fusion algorithm; The initial optimal detection path is dynamically adjusted based on the scoring and the preset dynamic adjustment strategy. After the test is completed, the test results are compared with the actual maintenance data, and the parameters of the model and the leakage knowledge graph are updated through a self-learning mechanism. The model is the model corresponding to the dynamic fusion algorithm.

[0006] In some embodiments, the real-time detection data includes pressure data, temperature data, acoustic data, and visual data, which are collected by sensors mounted on the leak detection robot and / or the leak detection enclosure.

[0007] In some embodiments, generating an initial optimal detection path based on a pre-stored leakage knowledge graph, the shape data of the product to be inspected, and historical leakage data includes: Based on the shape data of the product to be inspected, determine the target product model corresponding to the product to be inspected from the product model database; Based on the target product model, retrieve historical leakage pattern data associated with the target product model from the leakage knowledge graph; Statistical processing of the historical leakage data yields the historical high-incidence areas of leakage for the product under inspection. Based on the historical leakage pattern data and the historical high-incidence areas of leakage, the detection priority of each detection point is determined; Based on the detection priority and the coordinate information of each detection point, an initial optimal detection path is generated, which includes the detection route, detection frequency, and detection time.

[0008] In some embodiments, the calculation of the comprehensive confidence score based on real-time detection data and a preset dynamic fusion algorithm includes: The real-time detection data is normalized to obtain standardized detection data; The standardized detection data is subjected to time-frequency feature extraction using a pre-defined convolutional neural network to obtain leakage time-frequency feature data; Principal component analysis was performed on the leakage time-frequency characteristic data to reduce its dimensionality, resulting in dimensionality-reduced characteristic data. The dimensionality-reduced feature data are calculated based on a preset weighted fusion function to obtain the comprehensive confidence score.

[0009] In some embodiments, dynamically adjusting the optimal detection path based on the score and a preset dynamic adjustment strategy includes: The overall confidence score is compared with a preset first threshold and a second threshold, wherein the first threshold is greater than the second threshold; If the score is greater than the first threshold and the detection area corresponding to the score does not belong to the planned detection area in the current detection route, a priority detection path is generated, and the leak detection robot is controlled to interrupt the current detection route and perform detection according to the priority detection path. If the score is less than the second threshold multiple times in a row and matches the historical no-leakage feature of the corresponding detection point in the leakage knowledge graph, a path skipping instruction or a duration shortening instruction is generated to control the leak detection robot to skip secondary detection points near the detection point or reduce the dwell time at the detection point. If the score is greater than the second threshold and less than the first threshold, and the coordinate accuracy of the corresponding detection point meets the preset accuracy requirement, then an instruction to increase the collection frequency is generated to control the leak detection robot to increase the number of data collections and to perform statistical averaging on the data collected multiple times. If the score is greater than the second threshold and the coordinate accuracy of the corresponding detection point is insufficient, a grid-like scanning path centered on the detection point is generated, and the leak detection robot is controlled to perform high-density repeated scanning according to the grid-like scanning path.

[0010] In some embodiments, receiving the shape data of the product to be inspected collected by the laser sensor, comparing the shape data with a product model database, and generating an optimal hole layout scheme includes: The target product model of the product to be inspected is determined based on the comparison results. Obtain the standard shape parameters and required inspection locations corresponding to the target product model from the product model database; Based on the standard external shape parameters and the external shape data, calculate the distribution of the detection holes that need to be opened on the leak detection cover box; A detection hole status list is generated based on the location distribution. The detection hole status list is used to indicate the open or closed state of each detection hole.

[0011] In some embodiments, updating the model parameters and the leaked knowledge graph through a self-learning mechanism includes: The test results are compared with the actual maintenance data, and false positives and false negatives are marked. Based on the false positive and false negative samples, the model is retrained through incremental learning, the parameters of the model are updated, and the updated model is automatically deployed to the control unit. Based on the comparison results, update the association between product model, environmental conditions and leakage mode in the leakage knowledge graph.

[0012] Secondly, embodiments of the present invention also provide an intelligent leak detection device for performing the steps of the AI ​​and robot collaborative intelligent leak detection method described in the first aspect, the intelligent leak detection device comprising: The hole layout module is used to receive the shape data of the product to be inspected collected by the laser sensor, compare the shape data with the product model database, generate the optimal hole layout scheme, and send an opening command to the detection hole assembly according to the optimal hole layout scheme. The path generation module is used to generate an initial optimal detection path based on the pre-stored leakage knowledge graph, the shape data of the product to be inspected, and historical leakage data, and to send detection instructions to the leak detection robot based on the optimal detection path. The scoring calculation module is used to calculate the comprehensive confidence score based on real-time detection data and a preset dynamic fusion algorithm during the detection process. The path adjustment module is used to dynamically adjust the initial optimal detection path according to the score and the preset dynamic adjustment strategy. The model update module is used to compare the detection results with the actual maintenance data after the detection is completed, and update the parameters of the model and the leakage knowledge graph through a self-learning mechanism. The model is the model corresponding to the dynamic fusion algorithm.

[0013] Thirdly, embodiments of the present invention also provide a computer device, the device including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in memory, it implements the steps of the AI ​​and robot collaborative intelligent leak detection method described in the first aspect.

[0014] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the AI ​​and robot collaborative intelligent leak detection method as described in the first aspect.

[0015] Beneficial Effects: This invention provides an intelligent leak detection method, device, equipment, and medium that integrates AI and robotics. The method is applied to an intelligent leak detection system, which includes a leak detection enclosure, a leak detection robot, and a control unit. The leak detection enclosure has several detection holes and detection hole assemblies for closing or opening the detection holes. The product to be tested is placed inside the leak detection enclosure. A laser sensor for scanning the product's shape is located inside the leak detection enclosure. The laser sensor, detection hole assemblies, and leak detection robot are communicatively connected to the control unit. The leak detection method includes: receiving the shape data of the product to be tested collected by the laser sensor; comparing the shape data with a product model database; and generating... The system employs an optimal hole layout scheme, sending opening commands to the detection hole components based on this scheme. An initial optimal detection path is generated using a pre-stored leak knowledge graph, the product's shape data, and historical leak data. Detection commands are then sent to the leak detection robot based on this path. During detection, a comprehensive confidence score is calculated based on real-time detection data and a pre-set dynamic fusion algorithm. The initial optimal detection path is dynamically adjusted according to the score and a pre-set dynamic adjustment strategy. After detection, the results are compared with actual repair data, and the model's parameters and leak knowledge graph are updated through a self-learning mechanism. The model corresponds to the dynamic fusion algorithm. The intelligent flexible leak detection enclosure and self-closing leak detection hole structure enable the system to automatically match the optimal detection hole positions based on different product types and shapes, improving adaptability and flexibility, and achieving rapid switching and efficient detection. A built-in laser sensor in the leak detection enclosure scans the product's outline in real time, giving the system the ability to perceive the product structure and dynamically adjust detection parameters, thus improving detection matching accuracy and precision. By running a dynamic fusion algorithm, multi-source data such as historical fault data, real-time detection data, and environmental parameters are normalized and processed. Combined with a reinforcement learning mechanism, the fusion weights are automatically adjusted. The algorithm can autonomously identify high-risk leakage points and adjust the detection path according to different detection scenarios to improve detection efficiency and make the detection adaptive. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a hardware structure diagram of the intelligent detection system provided in an embodiment of the present invention; Figure 2 A flowchart of an AI and robot collaborative intelligent leak detection method provided in an embodiment of the present invention; Figure 3This is a schematic block diagram of an intelligent detection system provided in an embodiment of the present invention; Figure 4 A sub-flowchart of the AI ​​and robot collaborative intelligent leak detection method provided in an embodiment of the present invention; Figure 5 A sub-flowchart of the AI ​​and robot collaborative intelligent leak detection method provided in an embodiment of the present invention; Figure 6 A sub-flowchart of the AI ​​and robot collaborative intelligent leak detection method provided in an embodiment of the present invention; Figure 7 A sub-flowchart of the AI ​​and robot collaborative intelligent leak detection method provided in an embodiment of the present invention; Figure 8 A sub-flowchart of the AI ​​and robot collaborative intelligent leak detection method provided in an embodiment of the present invention; Figure 9 A schematic block diagram of an AI and robot collaborative intelligent leak detection device provided in an embodiment of the present invention; Figure 10 A schematic block diagram of a computer device provided for an embodiment of the present invention.

[0018] Figure label: 1. Leak detection enclosure; 11. Elastic sealing orifice plate; 12. Spring reset mechanism; 13. Electromagnetic drive module; 2. Leak detection robot; 3. Control unit. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0021] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0022] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0023] It should also be noted that, unless otherwise explicitly specified and limited, terms such as "installation," "connection," "linking," "fixing," and "setting" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. When an component is referred to as being "on" or "below" another component, the component can be located "directly" or "indirectly" on the other component, or there may be one or more intermediary components. The terms "first," "second," "third," etc., are only for the convenience of describing this technical solution and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, features defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. For those skilled in the art, the specific meaning of the above terms in this invention can be understood according to the specific circumstances.

[0024] Please see Figure 1 , Figure 2 and Figure 3 This invention provides an intelligent leak detection method that combines AI and robots, applied to an intelligent leak detection system. This invention belongs to the field of robot intelligent control technology, specifically to the field of intelligent equipment manufacturing.

[0025] The intelligent leak detection system includes a leak detection box 1, a leak detection robot 2, and a control unit 3. The leak detection box 1 is provided with a plurality of detection holes and a detection hole assembly for closing or opening the detection holes. The product to be tested is placed inside the leak detection box 1. The leak detection box 1 is provided with a laser sensor for scanning the shape of the product. The laser sensor, the detection hole assembly, and the leak detection robot 2 are communicatively connected to the control unit 3.

[0026] The leak detection cover 1 is an intelligent flexible leak detection cover, which has several detection holes, and each detection hole is equipped with a corresponding detection hole assembly.

[0027] The leak detection hole assembly includes an elastic sealing hole plate 11, a spring return mechanism 12, and an electromagnetic drive module 13. In the non-detection state, the electromagnetic drive module 13 is de-energized, and the spring pushes the elastic sealing hole plate 11 to close the leak detection hole, forming an airtight structure. When detection is initiated, the electromagnetic drive module 13 is energized, engaging the spring return mechanism 12 and opening the leak detection hole, allowing the leak detection robot 2 to pass through and perform detection. The electromagnetic drive module 13 is communicatively connected to the control unit 3 to receive control commands from the control unit.

[0028] The leak detection method includes steps S1 to S5.

[0029] S1. Receive the shape data of the product to be inspected collected by the laser sensor, compare the shape data with the product model database, generate the optimal hole layout scheme, and send an opening command to the detection hole assembly according to the optimal hole layout scheme.

[0030] In this embodiment, the laser sensor is a laser ranging sensor array, located inside the leak detection enclosure, used to scan the outline of the product under inspection in real time and acquire its shape data. The shape data includes the three-dimensional contour point cloud data of the product under inspection. After acquiring the three-dimensional contour point cloud data through the laser sensor, the control unit extracts the key shape feature parameters of the product under inspection. The key shape feature parameters may include the product's length, width, height, surface curvature, and interface location information.

[0031] The product model database is a pre-built database stored in the control unit. It includes the standard external dimensions and required inspection locations for each product model. The standard external dimensions include the standard length, standard width, standard height, and standard profile model for each model. The required inspection locations are the coordinates of the key areas for leak detection on each model, including weld seams, sealing interfaces, pipe connections, and vulnerable structural areas.

[0032] The shape data is compared with a product model database. This involves matching key shape feature parameters with the standard shape parameters of each product model in the database to determine the target product model corresponding to the product under inspection. The comparison process can employ a feature matching algorithm based on Euclidean distance to calculate the similarity between the key shape feature parameters and the standard shape parameters of each model, and the model with the highest similarity is determined as the target product model.

[0033] Please refer to the following: Figure 4 In one embodiment, receiving the shape data of the product to be inspected collected by the laser sensor, comparing the shape data with the product model database, and generating an optimal hole layout scheme includes sub-steps S11 to S14.

[0034] S11. Determine the target product model of the product to be inspected based on the comparison results.

[0035] In this embodiment, after the control unit acquires the shape data collected by the laser sensor, it extracts key shape feature parameters and compares the key shape feature parameters with the standard shape parameters of each model in the product model database one by one.

[0036] The comparison process may include: calculating the deviations of the length, width, and height of the product under inspection from the corresponding standard parameters of each model in the product model database; if the deviations of the standard shape parameters and key shape feature parameters of a certain model are all less than the preset tolerance threshold, then that model is determined to be the target product model. The preset tolerance threshold is set according to the product manufacturing tolerance and inspection accuracy requirements, for example, set to ±2mm.

[0037] If no model matching the key shape feature parameters exists in the product model database, meaning the deviation between the standard shape parameters and the key shape feature parameters of all models is greater than or equal to the preset tolerance threshold, the comparison is deemed a failure. In this case, the control unit switches to the default fully open hole mode, controlling all detection hole components to open to ensure that all surface areas of the product to be inspected can be detected; alternatively, the control unit generates a model entry prompt, prompting the operator to manually enter new model parameters, including the standard shape parameters of that model and the required detection location. After entry, the product model database is updated for automatic identification of subsequent products of the same model.

[0038] S12. Obtain the standard shape parameters and required detection locations corresponding to the target product model from the product model database.

[0039] In this embodiment, after determining the target product model, the control unit reads the standard shape parameters and required detection positions corresponding to that model from the product model database. The required detection positions are represented in coordinate form, which are three-dimensional coordinates relative to the product's reference coordinate system. The reference coordinate system is established based on the product's geometric center or positioning reference surface.

[0040] For example, the required inspection locations for a certain product model include: weld seam A (coordinates X1, Y1, Z1), sealing interface B (coordinates X2, Y2, Z2), and pipe connection point C (coordinates X3, Y3, Z3). Standard shape parameters are used to determine the spatial mapping relationship between various parts of the product surface and the leak detection enclosure when calculating the location distribution of the inspection holes.

[0041] S13. Calculate the distribution of the detection holes that need to be opened on the leak detection cover box based on the standard shape parameters and the shape data.

[0042] In this embodiment, the leak detection enclosure is a structure made of flexible material, and its surface has multiple detection holes distributed in a preset array. The positional distribution of the detection holes is predetermined based on the physical coordinate system of the leak detection enclosure, and each detection hole has a unique physical coordinate identifier.

[0043] The calculation of the distribution of the required detection holes on the leak detection box includes: determining the coordinates of each required detection position of the product under inspection in the reference coordinate system based on standard shape parameters; determining the actual pose of the product under inspection within the leak detection box based on shape data, i.e., the spatial transformation relationship between the reference coordinate system and the physical coordinate system of the leak detection box; mapping the coordinates of the required detection positions from the reference coordinate system to the physical coordinate system of the leak detection box through coordinate transformation, obtaining the projected coordinates of each required detection position on the surface of the leak detection box; and selecting the detection hole closest to each projected coordinate from the detection hole array as the detection hole to be opened, and calculating the distribution of the required detection hole positions.

[0044] The spatial transformation relationship is obtained by registering the three-dimensional contour point cloud data in the shape data with the standard contour model in the standard shape parameters. The registration calculation adopts the iterative nearest point algorithm to solve the rotation matrix and translation vector between the reference coordinate system and the physical coordinate system of the leak detection box, thereby achieving accurate mapping between the two coordinate systems.

[0045] S14. Generate a detection hole status list based on the location distribution. The detection hole status list is used to indicate the open or closed state of each detection hole.

[0046] In this embodiment, the control unit generates a detection hole status list based on the distribution of detection holes to be opened calculated in S13. The detection hole status list is a binary encoded list, where each element corresponds to a detection hole. An element value of "1" indicates that the detection hole needs to be opened, and an element value of "0" indicates that the detection hole remains closed.

[0047] The detection orifice assembly includes a resilient sealing orifice plate, a spring reset mechanism, and an electromagnetic drive module. The resilient sealing orifice plate is made of a resilient sealing material and is used to seal the detection orifice opening in the closed state; the spring reset mechanism includes a reset spring for providing the reset force of the resilient sealing orifice plate; and the electromagnetic drive module is used to generate electromagnetic attraction when energized.

[0048] When not in the detection state, the electromagnetic drive module is de-energized, and the spring reset mechanism is in the extended state, pushing the elastic sealing plate to fit tightly against the detection orifice, forming an airtight structure to prevent interference from the external environment; after receiving the opening command, the electromagnetic drive module is energized, generating electromagnetic attraction to compress the reset spring, causing the elastic sealing plate to detach from the detection orifice, thereby opening the detection hole.

[0049] Based on the detection hole status list, the control unit sends corresponding open or power-off commands to the electromagnetic drive modules of each detection hole assembly, controlling the detection holes that need to be opened to open and the detection holes that do not need to be opened to remain closed, thus completing the physical channel configuration before detection. The open command includes the physical coordinate identifier of the target detection hole and the power control signal.

[0050] In this way, the automatic model identification of the product to be inspected and the intelligent matching of the detection hole positions are realized, which solves the technical problems of low detection efficiency and poor adaptability caused by manual replacement of the leak detection mold and manual adjustment of the detection hole positions in the traditional leak detection system, and achieves the technical effect of rapid switching and efficient detection.

[0051] S2. Based on the pre-stored leakage knowledge graph, the shape data of the product to be inspected, and historical leakage data, generate an initial optimal detection path, and send a detection command to the leak detection robot based on the optimal detection path.

[0052] In this embodiment, the leakage knowledge graph is a pre-built graph-structured database stored in the control unit. The leakage knowledge graph uses product models as nodes, recording historical leakage patterns, leakage point distribution probabilities, and leakage characteristic association rules for each product model under different environmental conditions. Historical leakage data includes manufacturing process leakage data and after-sales maintenance records for the product under inspection. Manufacturing process leakage data consists of leakage detection records at each stage of the production line, while after-sales maintenance records are leakage faults and repair location information reported by users after the product leaves the factory.

[0053] The initial optimal detection path is a static path plan generated before the detection begins. It is used to indicate the detection route, detection frequency, and detection time of the leak detection robot within a single detection cycle. The detection route is the movement trajectory of the leak detection robot between the detection holes inside the leak detection enclosure. The detection frequency is the number of data collections at each detection point, and the detection time is the dwell time of the leak detection robot at each detection point.

[0054] Please refer to the following: Figure 5 In one embodiment, the step of generating an initial optimal detection path based on a pre-stored leakage knowledge graph, the shape data of the product to be inspected, and historical leakage data includes sub-steps S21 to S25.

[0055] S21. Based on the shape data of the product to be inspected, determine the target product model corresponding to the product to be inspected from the product model database.

[0056] In this embodiment, the control unit directly calls the target product model obtained in step S11, without repeating the comparison operation. The target product model serves as an index identifier for subsequent retrieval of related data from the leaked knowledge graph and product model database.

[0057] S22. Based on the target product model, retrieve historical leakage pattern data associated with the target product model from the leakage knowledge graph.

[0058] In this embodiment, the control unit uses the target product model as the search keyword to perform a graph traversal query in the leakage knowledge graph to obtain historical leakage pattern data associated with that product model. The historical leakage pattern data includes the location coordinates of leaks that occurred during each historical detection period for that product model, the range of environmental parameters at the time of the leak, the characteristic parameters of the leak signal, and the severity level of the leak. The leakage knowledge graph is used to characterize the relationship between product model, environmental conditions, and leakage patterns.

[0059] S23. Perform statistical processing on the historical leakage data to obtain the historical high-incidence areas of leakage of the product to be inspected.

[0060] In this embodiment, the control unit acquires manufacturing process leakage data and after-sales maintenance records from historical leakage data, extracts the recorded leakage location coordinates, and maps these coordinates to the reference coordinate system of the product under inspection. Spatial clustering analysis is then performed on the mapped leakage location coordinates using a density-based clustering algorithm, such as the DBSCAN algorithm. Leakage location coordinates with a spatial distance less than a preset cluster radius are grouped into the same cluster. The centroid coordinates and leakage frequency within each cluster are calculated. The area covered by clusters with leakage frequencies greater than a preset frequency threshold is identified as a historical high-incidence leakage area, and the center coordinates and boundary range of each historical high-incidence leakage area are recorded.

[0061] S24. Determine the detection priority of each detection point based on the historical leakage pattern data and the historical high-incidence areas of leakage.

[0062] In this embodiment, the control unit obtains the required detection locations corresponding to the product model from the product model database and spatially matches each required detection location with historical high-incidence leakage areas. If a required detection location falls within the boundary range of a historical high-incidence leakage area, the detection priority of that detection point is increased to Level 1 priority; if the distance between a required detection location and the center coordinates of a historical high-incidence leakage area is less than a preset distance threshold, the detection priority of that detection point is increased to Level 2 priority; the detection priority of the remaining detection points is Level 3 priority. Level 1 priority is higher than Level 2 priority, and Level 2 priority is higher than Level 3 priority. In addition, the control unit further sorts the Level 1 and Level 2 priority detection points according to the leakage severity level in the historical leakage pattern data, with detection points of higher leakage severity level having a higher detection priority.

[0063] S25. Based on the detection priority and the coordinate information of each detection point, generate an initial optimal detection path that includes the detection route, detection frequency and detection time.

[0064] In this embodiment, the control unit generates a detection route using a preset path planning algorithm based on the coordinate information and detection priority of each detection point. The path planning algorithm is an improved fast exploration random tree star algorithm, which takes the initial position of the leak detection robot as the starting point, the coordinates of each detection point as the waypoints, and aims to minimize the total movement distance and maximize the coverage of high-priority detection points to generate the optimal movement trajectory connecting each detection point as the detection route.

[0065] The detection frequency and duration for each detection point are determined based on detection priority: Level 1 priority detection points are assigned the first detection frequency (e.g., 10 data points per point) and the first detection duration (e.g., 30 seconds); Level 2 priority detection points are assigned the second detection frequency (e.g., 6 data points per point) and the second detection duration (e.g., 20 seconds); Level 3 priority detection points are assigned the third detection frequency (e.g., 3 data points per point) and the third detection duration (e.g., 10 seconds). The first detection frequency is greater than the second detection frequency, and the second detection frequency is greater than the third detection frequency; the first detection duration is greater than the second detection duration, and the second detection duration is greater than the third detection duration.

[0066] After generating the initial optimal detection path, the control unit sends a detection command corresponding to the initial optimal detection path to the leak detection robot. After receiving the detection command, the leak detection robot starts to execute the detection task according to the initial optimal detection path.

[0067] In this way, by using S2 to pre-plan the detection path based on historical data, the technical problem of low detection efficiency caused by fixed detection paths and lack of targeting in traditional leak detection systems is solved, achieving the technical effect of differentiated detection and priority of key areas.

[0068] S3. During the detection process, a comprehensive confidence score is calculated based on real-time detection data and a preset dynamic fusion algorithm.

[0069] In this embodiment, the real-time detection data refers to the data collected in real time by sensors mounted on the leak detection robot and / or the leak detection enclosure during the detection process. The real-time detection data includes pressure data, temperature data, acoustic data, and visual data. Pressure data is collected by a pressure sensor to characterize changes in air or hydraulic pressure at the detection point; temperature data is collected by an infrared temperature sensor to characterize the temperature field distribution at the detection point; acoustic data is collected by an ultrasonic sensor or acoustic array sensor to characterize the acoustic signal characteristics generated by the leak; and visual data is collected by an industrial camera or infrared thermal imager to characterize the image features and thermal distribution at the detection point. The pressure sensor, infrared temperature sensor, ultrasonic sensor or acoustic array sensor, industrial camera or infrared thermal imager can be mounted on the detection head of the leak detection robot or inside the leak detection enclosure.

[0070] The dynamic fusion algorithm is a multi-source data fusion algorithm preset in the control unit, which is used to fuse heterogeneous real-time detection data into a unified comprehensive confidence score to eliminate the dimensional differences and noise interference between different sensor data.

[0071] Please refer to the following: Figure 6 In one embodiment, the step of calculating the comprehensive confidence score based on real-time detection data and a preset dynamic fusion algorithm includes sub-steps S31 to S34.

[0072] S31. Normalize the real-time detection data to obtain standardized detection data.

[0073] In this embodiment, after receiving real-time detection data, the control unit preprocesses the data from various sensors. Preprocessing includes filtering and denoising, outlier removal, and timestamp alignment. Filtering and denoising employs Kalman filtering or wavelet thresholding algorithms to eliminate high-frequency noise during sensor acquisition. Outlier removal uses the statistical 3σ criterion, identifying and removing data points that deviate from the mean by more than three standard deviations. Timestamp alignment uses an interpolation algorithm based on a unified clock source to align the sampling data from different sensors to the same time reference.

[0074] After preprocessing, the real-time detection data is normalized using the Min-Max normalization method or the Z-score normalization method. The pressure data, temperature data, sound wave data, and visual data are mapped to a unified numerical range (e.g., the [0, 1] range) to obtain standardized detection data and eliminate the dimensional differences between different physical quantities.

[0075] S32. Extract time-frequency features from the standardized detection data using a preset convolutional neural network to obtain leakage time-frequency feature data.

[0076] In this embodiment, the control unit inputs standardized detection data into a preset convolutional neural network (CNN) model. The CNN model includes multiple convolutional layers, pooling layers, and fully connected layers. The convolutional layers are used to extract local feature patterns from the standardized detection data, the pooling layers are used to reduce the feature dimensionality and enhance the feature translation invariance, and the fully connected layers are used to map the extracted features to a high-dimensional feature space.

[0077] For acoustic data, the CNN model extracts the time-frequency features of the acoustic signal using a one-dimensional convolutional kernel, including frequency distribution features, energy attenuation features, and harmonic component features. For visual data, the CNN model extracts texture features, edge features, and hot spot distribution features from the image using a two-dimensional convolutional kernel. For pressure and temperature data, the CNN model extracts their temporal variation features and gradient abrupt change features using a one-dimensional convolutional kernel. The CNN model outputs a multi-dimensional feature vector as the leakage time-frequency feature data.

[0078] S33. Perform principal component analysis to reduce the dimensionality of the leakage time-frequency feature data to obtain dimensionality-reduced feature data.

[0079] In this embodiment, the control unit performs principal component analysis (PCA) dimensionality reduction on the leakage time-frequency feature data. The PCA dimensionality reduction process includes: calculating the covariance matrix of the leakage time-frequency feature data; performing eigenvalue decomposition on the covariance matrix to obtain the eigenvalues ​​and eigenvectors corresponding to each principal component; selecting the eigenvectors corresponding to the k largest eigenvalues ​​to form a projection matrix based on a preset cumulative contribution rate threshold (e.g., 95%); and projecting the leakage time-frequency feature data onto the low-dimensional subspace formed by the projection matrix to obtain the dimensionality-reduced feature data.

[0080] By using PCA dimensionality reduction, the complexity of subsequent fusion calculations is reduced while preserving the main leaked feature information, and the interference of redundant features on the fusion results is minimized.

[0081] S34. Calculate the dimensionality reduction feature data according to the preset weighted fusion function to obtain the comprehensive confidence score.

[0082] In this embodiment, the control unit performs a weighted summation of the dimensionality-reduced feature data using a preset weighted fusion function to calculate the comprehensive confidence score. The expression for the weighted fusion function is: F = w1 × f1 + w2 × f2 + ... + wn × fn.

[0083] Where F is the overall confidence score, wi is the feature weight corresponding to the i-th feature dimension, fi is the normalized feature value of the i-th feature dimension in the dimensionality-reduced feature data, and n is the number of dimensions of the dimensionality-reduced feature data.

[0084] The feature weights wi are dynamically adjusted through a pre-defined reinforcement learning mechanism. This mechanism employs the Q-learning algorithm, which dynamically updates the weight values ​​of each feature dimension based on feedback from each detection result (such as false detection, missed detection, or correct detection). This allows the overall confidence score to adapt to changes in leakage characteristics under different operating conditions. For example, when a certain type of leak is detected to be more likely to occur under specific temperature conditions, the system automatically increases the weight value of the feature dimension corresponding to the temperature data.

[0085] The overall confidence score F characterizes the leakage risk level of the current detection point or area; a higher score indicates a greater leakage risk. As the core decision-making basis for subsequent dynamic adjustment of the detection path, the overall confidence score does not directly control the motor movement of the leak detection robot, but rather indirectly influences its behavior by triggering feedback signals and guiding actions.

[0086] In this way, the deep fusion and unified quantitative evaluation of multi-source heterogeneous detection data are achieved, solving the technical problems of severe data silos and difficulty in effectively fusion and analysis of multi-source data in traditional leak detection systems, and improving the adaptability of detection.

[0087] S4. The initial optimal detection path is dynamically adjusted according to the score and the preset dynamic adjustment strategy.

[0088] In this embodiment, the dynamic adjustment strategy is a set of preset path adjustment rules in the control unit, which is used to reconstruct the initial optimal detection path in real time based on the comprehensive confidence score calculated in real time during the detection process, in order to cope with complex working conditions such as sudden leakage, environmental interference and individual product differences.

[0089] Dynamic adjustments include real-time reconstruction of the detection route, dynamic increase or decrease of the detection frequency, and real-time adjustment of the detection time. During the leak detection robot's detection task, the control unit calculates the comprehensive confidence score of the current detection point in real time at a preset sampling period (e.g., every 0.5 seconds), matches the score with the preset dynamic adjustment strategy, and generates corresponding path adjustment instructions to send to the leak detection robot.

[0090] Please refer to the following: Figure 7 In one embodiment, the step of dynamically adjusting the optimal detection path according to the score and a preset dynamic adjustment strategy includes sub-steps S41 to S45.

[0091] S41. The comprehensive confidence score is compared with a preset first threshold and a second threshold, wherein the first threshold is greater than the second threshold.

[0092] In this embodiment, after the control unit obtains the comprehensive confidence score F of the current detection point, it compares F with preset first threshold Th1 and second threshold Th2. The first threshold Th1 is the confirmed leakage risk threshold, used to determine whether there is a highly certain leakage risk; the second threshold Th2 is the suspected leakage risk threshold, used to determine whether there is a potential leakage risk. The first threshold Th1 is greater than the second threshold Th2, for example, Th1 = 0.85 and Th2 = 0.45. The thresholds are set based on historical data statistics and engineering experience and can be adjusted in practical applications.

[0093] S42. If the score is greater than the first threshold and the detection area corresponding to the score does not belong to the planned detection area in the current detection route, a priority detection path is generated, and the leak detection robot is controlled to interrupt the current detection route and perform detection according to the priority detection path.

[0094] In this embodiment, if the comprehensive confidence score F is greater than the first threshold Th1, and the detection area that generated the score is not in the current or subsequent planned detection routes of the initial optimal detection path, it indicates that a sudden high-risk leakage signal has occurred. At this time, the control unit generates a priority detection path, which is the shortest path from the current position of the leak detection robot to the sudden detection area.

[0095] The control unit sends a path interruption command and a priority detection path command to the leak detection robot, causing the robot to immediately stop its current detection action and move to the sudden detection area for priority inspection. After completing the priority inspection, the leak detection robot returns to the interrupted position in its original detection route and continues to perform the remaining detection tasks. The priority detection path can be generated using shortest path planning based on Dijkstra's algorithm, with the goal of minimizing travel time.

[0096] If the score is greater than the first threshold, but the detection area corresponding to the score belongs to the planned detection area in the current detection route, the leak detection robot will continue to perform detection according to the initial optimal detection path without triggering path interruption.

[0097] S43. If the score is less than the second threshold multiple times in a row and matches the historical no-leakage feature of the corresponding detection point in the leakage knowledge graph, a path skipping instruction or a duration shortening instruction is generated to control the leak detection robot to skip secondary detection points near the detection point or reduce the dwell time at the detection point.

[0098] In this embodiment, if the comprehensive confidence score F is less than the second threshold Th2 for N consecutive times (e.g., N=5), and the control unit compares the real-time detection data features of the current detection point with the historical no-leakage features corresponding to the detection point in the leakage knowledge graph, and the matching degree between the two is greater than the preset matching threshold, it indicates that the current detection point and the adjacent area are in a safe state.

[0099] At this point, the control unit generates either a path skipping command or a time reduction command. The path skipping command controls the leak detection robot to skip secondary detection points (i.e., third-priority detection points) near the current detection point and proceed directly to the next critical detection node. The time reduction command controls the leak detection robot to reduce its dwell time at the detection point, for example, reducing the planned 30-second dwell time to 10 seconds, and correspondingly reducing the detection frequency. By skipping or accelerating through the safe zone, the detection cycle time for a single product is shortened, improving overall detection efficiency.

[0100] S44. If the score is greater than the second threshold and less than the first threshold, and the coordinate accuracy of the corresponding detection point meets the preset accuracy requirement, then an instruction to increase the collection frequency is generated to control the leak detection robot to increase the number of data collections, and to perform statistical averaging on the data collected multiple times.

[0101] In this embodiment, specifically, if the overall confidence score F is greater than the second threshold Th2 and less than the first threshold Th1, it indicates that there is a potential leakage risk at the current detection point, but the risk level has not yet reached the confirmation standard. At this time, if the deviation between the current positioning coordinates of the leak detection robot and the preset coordinates of the detection point is less than the preset accuracy requirement (e.g., the deviation is less than ±1mm), then the coordinate accuracy is determined to meet the requirements.

[0102] The control unit generates a command to increase the data acquisition frequency, controlling the leak detection robot to remain stationary or fine-tune its posture at the detection point, increasing the data acquisition frequency from the default once per second to twice per second, or continuously acquiring data for M seconds (e.g., M=10 seconds), obtaining multiple real-time detection data. The control unit performs statistical averaging on the multiple acquired data, such as calculating the arithmetic mean of multiple pressure data and the root mean square value of multiple sound wave data. Statistical averaging eliminates random noise interference and improves the confidence level of the judgment.

[0103] S45. If the score is greater than the second threshold and the coordinate accuracy of the corresponding detection point is insufficient, a grid-like scanning path centered on the detection point is generated, and the leak detection robot is controlled to perform high-density repeated scanning according to the grid-like scanning path.

[0104] In this embodiment, if the overall confidence score F is greater than the second threshold Th2, and the deviation between the current positioning coordinates of the leak detection robot and the preset coordinates of the detection point is greater than or equal to the preset accuracy requirement, or the uncertainty of the leak source location corresponding to the score is greater than the preset uncertainty threshold, then the coordinate accuracy is determined to be insufficient.

[0105] The control unit generates a grid-like scanning path, which is a dense grid trajectory planned within a square area with a preset side length (e.g., 50mm) centered on the current detection point. The step size of the grid trajectory is smaller than that of a conventional detection path (e.g., a conventional step size of 20mm, while the grid step size is 5mm). The control unit controls the leak detection robot to perform high-density, small-step repeated scans within this area. By increasing the spatial sampling density, the center coordinates of the leak source are accurately located, reducing the uncertainty range of the leak location.

[0106] In this way, real-time feedback and strategy adjustment can be achieved during the detection process, solving the technical problems of fixed and rigid detection paths and lack of real-time feedback and strategy adjustment capabilities in traditional leak detection systems, thus achieving the technical effect of improving detection efficiency and intelligence.

[0107] S5. After the test is completed, the test results are compared with the actual maintenance data, and the parameters of the model and the leakage knowledge graph are updated through a self-learning mechanism. The model is the model corresponding to the dynamic fusion algorithm.

[0108] In this embodiment, the detection results include the comprehensive confidence score of each detection point, the leakage status determination result, and the coordinate information of the leakage point. The actual maintenance data is the maintenance record data that has been manually reviewed or confirmed by high-precision verification after the detection is completed. The actual maintenance data includes the actual leakage location, the actual leakage cause, the actual leakage degree, and the maintenance measures.

[0109] Self-learning mechanisms include incremental learning mechanisms, which are used to continuously optimize the model by adding new samples without affecting the performance of the existing model.

[0110] Please refer to the following: Figure 8 In one embodiment, updating the model parameters and the leaked knowledge graph through a self-learning mechanism includes sub-steps S51 to S53.

[0111] S51. Compare the test results with the actual maintenance data, and mark the falsely detected samples and the missed samples.

[0112] In this embodiment, the control unit acquires the leakage status determination result and leakage point coordinate information from the detection results, and compares them one by one with the actual leakage location and actual leakage cause in the actual maintenance data. If the detection result determines that a certain detection point is in a leakage state, but the actual maintenance data shows that there is no leakage at that location, then the real-time detection data and comprehensive confidence score corresponding to that detection point are marked as false detection samples; if the detection result determines that a certain detection point is in a normal state, but the actual maintenance data shows that there is a leakage at that location, then the real-time detection data and comprehensive confidence score corresponding to that detection point are marked as missed detection samples. Samples that are correctly determined are marked as correct detection samples.

[0113] The false detection samples, missed detection samples, and correct detection samples together constitute the incremental training sample set. Each sample in the sample set includes input features (i.e., standardized detection data of real-time detection data) and labels (i.e., the actual leakage status after verification by actual maintenance data).

[0114] S52. Based on the false detection samples and false negative samples, the model is retrained through incremental learning to update the parameters of the model, and the updated model is automatically deployed to the control unit.

[0115] In this embodiment, the control unit inputs the incremental training sample set into the weight allocator of the CNN model and the weighted fusion function, and retrains it using an incremental learning algorithm. The incremental learning algorithm is an incremental learning based on an experience playback mechanism. During the training process, batch samples are randomly selected from the incremental training sample set, and an equal number of historical samples are selected from the historical sample library. These samples are mixed to form a training batch to prevent new samples from forgetting historical knowledge.

[0116] The convolutional kernel parameters and fully connected layer weights of the CNN model are updated using the backpropagation algorithm, while the feature weights wi in the weighted fusion function are updated using the reward feedback mechanism of the Q-learning algorithm. After training, the updated model parameters are obtained, and the control unit automatically deploys the updated model to the local AI processing unit, replacing the old model, so that subsequent detection tasks can directly use the optimized model.

[0117] S53. Based on the comparison results, update the association between product model, environmental conditions and leakage mode in the leakage knowledge graph.

[0118] In this embodiment, the control unit extracts new leakage pattern features based on the actual leakage cause and location in the actual maintenance data, combined with the environmental parameters (temperature, humidity, air pressure) during the current detection. The new leakage pattern features are then correlated with existing nodes in the leakage knowledge graph. If a correlation exists between the new leakage pattern feature and an existing node, the weight value of the corresponding correlation edge is updated; otherwise, a new node and correlation edge are added to the leakage knowledge graph.

[0119] For example, if actual maintenance data shows that a certain product model leaked at weld X in a high-temperature and high-humidity environment, but the leakage knowledge graph does not record this environment-location association for that product model, then a new association edge is added to the leakage knowledge graph between the product model node, the high-temperature and high-humidity environment condition node, and the weld X location node, and the confidence level of this association is recorded. Through continuous updates, the leakage knowledge graph continuously accumulates the association knowledge between different product models, environmental conditions, and leakage modes, which is used to assist subsequent detection decisions, allowing the system's detection experience to continuously grow over time.

[0120] In this way, the detection system can achieve continuous self-learning and knowledge evolution, solving the technical problem of fixed model parameters and inability to continuously optimize based on actual detection feedback in traditional leak detection systems, thus achieving the technical effect of improving detection accuracy and system adaptability.

[0121] In summary, the AI ​​and robot collaborative intelligent leak detection method provided by the embodiments of the present invention solves the technical problems of low detection accuracy, poor adaptability, low efficiency, and difficulty in dynamically optimizing detection strategies in traditional leak detection systems.

[0122] It has at least the following beneficial effects: Through the intelligent flexible leak detection enclosure and self-closing leak detection hole structure, the system can automatically match the optimal detection hole position according to different product types and shapes, improving the adaptability and flexibility of detection, and achieving the technical effect of rapid switching and efficient detection.

[0123] By incorporating a built-in laser rangefinder sensor array into the intelligent flexible leak detection enclosure, the system can scan the product's outline in real time, enabling it to perceive the product's structure and dynamically adjust detection parameters to improve detection matching and accuracy.

[0124] By using AI-powered intelligent processing and dynamic fusion algorithms, the system normalizes multi-source data such as historical fault data, real-time detection data, and environmental parameters. Combined with reinforcement learning mechanisms, it automatically adjusts the fusion weights, enabling the system to autonomously optimize model parameters according to different detection scenarios, identify high-risk leakage points, and achieve the technical effect of improving detection robustness and adaptability.

[0125] By acquiring precise coordinates of the leak point through a robot positioning system and collaborating with an AI processing unit to generate the optimal detection path, the robot can efficiently execute detection tasks and provide real-time data feedback, thereby improving detection efficiency and intelligence.

[0126] By implementing unified scheduling and status monitoring through the control unit, the system ensures that all modules work together, improves system stability and reliability, and achieves the technical effect of efficient collaboration and controllable status.

[0127] This invention also provides an AI-robot collaborative intelligent leak detection device, which can be configured in a control unit and is used to execute any embodiment of the aforementioned AI-robot collaborative intelligent leak detection method. Specifically, please refer to... Figure 9 , Figure 9 This is a schematic block diagram of an AI and robot collaborative intelligent leak detection device provided in an embodiment of the present invention.

[0128] like Figure 9As shown, the AI ​​and robot collaborative intelligent leak detection device 100 includes: The hole layout module 110 is used to receive the shape data of the product to be inspected collected by the laser sensor, compare the shape data with the product model database, generate the optimal hole layout scheme, and send an opening command to the detection hole assembly according to the optimal hole layout scheme. The path generation module 120 is used to generate an initial optimal detection path based on the pre-stored leakage knowledge graph, the shape data of the product to be inspected, and historical leakage data, and to send a detection command to the leak detection robot based on the optimal detection path. The scoring calculation module 130 is used to calculate the comprehensive confidence score based on real-time detection data and a preset dynamic fusion algorithm during the detection process. The path adjustment module 140 is used to dynamically adjust the initial optimal detection path according to the score and the preset dynamic adjustment strategy. The model update module 150 is used to compare the detection results with the actual maintenance data after the detection is completed, and update the parameters of the model and the leakage knowledge graph through a self-learning mechanism. The model is the model corresponding to the dynamic fusion algorithm.

[0129] In one embodiment, the hole layout module 110 is specifically used for: The target product model of the product to be inspected is determined based on the comparison results. Obtain the standard shape parameters and required inspection locations corresponding to the target product model from the product model database; Based on the standard external shape parameters and the external shape data, calculate the distribution of the detection holes that need to be opened on the leak detection cover box; A detection hole status list is generated based on the location distribution. The detection hole status list is used to indicate the open or closed state of each detection hole.

[0130] In one embodiment, the path generation module 120 is specifically used for: Based on the shape data of the product to be inspected, determine the target product model corresponding to the product to be inspected from the product model database; Based on the target product model, retrieve historical leakage pattern data associated with the target product model from the leakage knowledge graph; Statistical processing of the historical leakage data yields the historical high-incidence areas of leakage for the product under inspection. Based on the historical leakage pattern data and the historical high-incidence areas of leakage, the detection priority of each detection point is determined; Based on the detection priority and the coordinate information of each detection point, an initial optimal detection path is generated, which includes the detection route, detection frequency, and detection time.

[0131] In one embodiment, the scoring calculation module 130 is specifically used for: The real-time detection data is normalized to obtain standardized detection data; The standardized detection data is subjected to time-frequency feature extraction using a pre-defined convolutional neural network to obtain leakage time-frequency feature data; Principal component analysis was performed on the leakage time-frequency characteristic data to reduce its dimensionality, resulting in dimensionality-reduced characteristic data. The dimensionality-reduced feature data are calculated based on a preset weighted fusion function to obtain the comprehensive confidence score.

[0132] In one embodiment, the path adjustment module 140 is specifically used for: The overall confidence score is compared with a preset first threshold and a second threshold, wherein the first threshold is greater than the second threshold; If the score is greater than the first threshold and the detection area corresponding to the score does not belong to the planned detection area in the current detection route, a priority detection path is generated, and the leak detection robot is controlled to interrupt the current detection route and perform detection according to the priority detection path. If the score is less than the second threshold multiple times in a row and matches the historical no-leakage feature of the corresponding detection point in the leakage knowledge graph, a path skipping instruction or a duration shortening instruction is generated to control the leak detection robot to skip secondary detection points near the detection point or reduce the dwell time at the detection point. If the score is greater than the second threshold and less than the first threshold, and the coordinate accuracy of the corresponding detection point meets the preset accuracy requirement, then an instruction to increase the collection frequency is generated to control the leak detection robot to increase the number of data collections and to perform statistical averaging on the data collected multiple times. If the score is greater than the second threshold and the coordinate accuracy of the corresponding detection point is insufficient, a grid-like scanning path centered on the detection point is generated, and the leak detection robot is controlled to perform high-density repeated scanning according to the grid-like scanning path.

[0133] In one embodiment, the model update module 150 is specifically used for: The test results are compared with the actual maintenance data, and false positives and false negatives are marked. Based on the false positive and false negative samples, the model is retrained through incremental learning, the parameters of the model are updated, and the updated model is automatically deployed to the control unit. Based on the comparison results, update the association between product model, environmental conditions and leakage mode in the leakage knowledge graph.

[0134] The aforementioned AI- and robot-assisted intelligent leak detection device can be implemented as a computer program, which can, for example... Figure 10 It runs on the computer device shown.

[0135] Please see Figure 10 , Figure 10 This is a schematic block diagram of a computer device provided in an embodiment of the present invention.

[0136] The computer device 500 includes a processor 502, a memory, and a network interface 505 connected via a communication bus 501. The memory may include a storage medium 503 and internal memory 504.

[0137] The storage medium 503 may store an operating system 5031 and a computer program 5032. When the computer program 5032 is executed, it enables the processor 502 to execute an intelligent leak detection method that integrates AI and robotics. The storage medium 503 may be a volatile storage medium or a non-volatile storage medium.

[0138] The processor 502 provides computing and control capabilities to support the operation of the entire computer device 500.

[0139] The internal memory 504 provides an environment for the operation of the computer program 5032 in the storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute an intelligent leak detection method that combines AI and robots.

[0140] This network interface 505 is used for network communication, such as providing data transmission. Those skilled in the art will understand that... Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device 500 to which the present invention is applied. The specific computer device 500 may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0141] The processor 502 is used to run the computer program 5032 stored in the memory to realize the corresponding functions in the AI ​​and robot collaborative intelligent leak detection method described above.

[0142] Those skilled in the art will understand that Figure 10 The embodiments of the computer device shown do not constitute a limitation on the specific configuration of the computer device. In other embodiments, the computer device may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. For example, in some embodiments, the computer device may include only memory and a processor. In such embodiments, the structure and function of the memory and processor are different from those shown. Figure 10 The embodiments shown are consistent and will not be repeated here.

[0143] It should be understood that, in this embodiment of the invention, the processor 502 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0144] In another embodiment of the invention, a computer-readable storage medium is provided. This computer-readable storage medium may be volatile or non-volatile. The computer-readable storage medium stores a computer program, wherein when executed by a processor, the computer program implements the steps included in the above-described AI-robot collaborative intelligent leak detection method.

[0145] Those skilled in the art will readily understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention.

[0146] In the embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Units with the same function may be grouped into one unit. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, or it may be an electrical, mechanical, or other form of connection.

[0147] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of the present invention, depending on actual needs.

[0148] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0149] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a computer-readable storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned computer-readable storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), magnetic disks, or optical disks.

[0150] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An intelligent leak detection method that integrates AI and robots, characterized in that, An intelligent leak detection system is applied, comprising a leak detection chamber, a leak detection robot, and a control unit. The leak detection chamber has several detection holes and detection hole assemblies for closing or opening the detection holes. The product to be tested is placed inside the leak detection chamber. A laser sensor for scanning the product's shape is installed inside the leak detection chamber. The laser sensor, the detection hole assemblies, and the leak detection robot are communicatively connected to the control unit. The leak detection method includes: The system receives the shape data of the product to be inspected collected by the laser sensor, compares the shape data with the product model database, generates an optimal hole layout scheme, and sends an opening command to the detection hole assembly according to the optimal hole layout scheme. Based on the pre-stored leakage knowledge graph, the shape data of the product to be inspected, and historical leakage data, an initial optimal detection path is generated, and a detection command is sent to the leak detection robot based on the optimal detection path. During the detection process, a comprehensive confidence score is calculated based on real-time detection data and a preset dynamic fusion algorithm; The initial optimal detection path is dynamically adjusted based on the scoring and the preset dynamic adjustment strategy. After the test is completed, the test results are compared with the actual maintenance data, and the parameters of the model and the leakage knowledge graph are updated through a self-learning mechanism. The model is the model corresponding to the dynamic fusion algorithm.

2. The intelligent leak detection method based on AI and robot collaboration according to claim 1, characterized in that, The real-time detection data includes pressure data, temperature data, acoustic data, and visual data, which are collected by sensors mounted on the leak detection robot and / or the leak detection enclosure.

3. The intelligent leak detection method based on AI and robot collaboration according to claim 1, characterized in that, The process of generating an initial optimal detection path based on a pre-stored leakage knowledge graph, the shape data of the product to be inspected, and historical leakage data includes: Based on the shape data of the product to be inspected, determine the target product model corresponding to the product to be inspected from the product model database; Based on the target product model, retrieve historical leakage pattern data associated with the target product model from the leakage knowledge graph; Statistical processing of the historical leakage data yields the historical high-incidence areas of leakage for the product under inspection. Based on the historical leakage pattern data and the historical high-incidence areas of leakage, the detection priority of each detection point is determined; Based on the detection priority and the coordinate information of each detection point, an initial optimal detection path is generated, which includes the detection route, detection frequency, and detection time.

4. The intelligent leak detection method based on AI and robot collaboration according to claim 1, characterized in that, The calculation of the comprehensive confidence score based on real-time detection data and a preset dynamic fusion algorithm includes: The real-time detection data is normalized to obtain standardized detection data; The standardized detection data is subjected to time-frequency feature extraction using a pre-defined convolutional neural network to obtain leakage time-frequency feature data; Principal component analysis was performed on the leakage time-frequency characteristic data to reduce its dimensionality, resulting in dimensionality-reduced characteristic data. The dimensionality-reduced feature data are calculated based on a preset weighted fusion function to obtain the comprehensive confidence score.

5. The intelligent leak detection method based on AI and robot collaboration according to claim 1, characterized in that, The step of dynamically adjusting the optimal detection path based on the score and a preset dynamic adjustment strategy includes: The overall confidence score is compared with a preset first threshold and a second threshold, wherein the first threshold is greater than the second threshold; If the score is greater than the first threshold and the detection area corresponding to the score does not belong to the planned detection area in the current detection route, a priority detection path is generated, and the leak detection robot is controlled to interrupt the current detection route and perform detection according to the priority detection path. If the score is less than the second threshold multiple times in a row and matches the historical no-leakage feature of the corresponding detection point in the leakage knowledge graph, a path skipping instruction or a duration shortening instruction is generated to control the leak detection robot to skip secondary detection points near the detection point or reduce the dwell time at the detection point. If the score is greater than the second threshold and less than the first threshold, and the coordinate accuracy of the corresponding detection point meets the preset accuracy requirement, then an instruction to increase the collection frequency is generated to control the leak detection robot to increase the number of data collections and to perform statistical averaging on the data collected multiple times. If the score is greater than the second threshold and the coordinate accuracy of the corresponding detection point is insufficient, a grid-like scanning path centered on the detection point is generated, and the leak detection robot is controlled to perform high-density repeated scanning according to the grid-like scanning path.

6. The intelligent leak detection method based on AI and robot collaboration according to claim 1, characterized in that, The process of receiving the shape data of the product to be inspected collected by the laser sensor, comparing the shape data with the product model database, and generating an optimal hole layout scheme includes: The target product model of the product to be inspected is determined based on the comparison results. Obtain the standard shape parameters and required inspection locations corresponding to the target product model from the product model database; Based on the standard external shape parameters and the external shape data, calculate the distribution of the detection holes that need to be opened on the leak detection cover box; A detection hole status list is generated based on the location distribution. The detection hole status list is used to indicate the open or closed state of each detection hole.

7. The intelligent leak detection method based on AI and robot collaboration according to claim 1, characterized in that, The updating of model parameters and the leaked knowledge graph through a self-learning mechanism includes: The test results are compared with the actual maintenance data, and false positives and false negatives are marked. Based on the false positive and false negative samples, the model is retrained through incremental learning, the parameters of the model are updated, and the updated model is automatically deployed to the control unit. Based on the comparison results, update the association between product model, environmental conditions and leakage mode in the leakage knowledge graph.

8. An intelligent leak detection device, characterized in that, The step of performing the AI ​​and robot collaborative intelligent leak detection method according to any one of claims 1-7, wherein the intelligent leak detection device comprises: The hole layout module is used to receive the shape data of the product to be inspected collected by the laser sensor, compare the shape data with the product model database, generate the optimal hole layout scheme, and send an opening command to the detection hole assembly according to the optimal hole layout scheme. The path generation module is used to generate an initial optimal detection path based on the pre-stored leakage knowledge graph, the shape data of the product to be inspected, and historical leakage data, and to send detection instructions to the leak detection robot based on the optimal detection path. The scoring calculation module is used to calculate the comprehensive confidence score based on real-time detection data and a preset dynamic fusion algorithm during the detection process. The path adjustment module is used to dynamically adjust the initial optimal detection path according to the score and the preset dynamic adjustment strategy. The model update module is used to compare the detection results with the actual maintenance data after the detection is completed, and update the parameters of the model and the leakage knowledge graph through a self-learning mechanism. The model is the model corresponding to the dynamic fusion algorithm.

9. A computer device, characterized in that, The device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in memory, it implements the steps of the AI ​​and robot collaborative intelligent leak detection method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the AI ​​and robot collaborative intelligent leak detection method as described in any one of claims 1-7.

Images