Multi-robot-based large-scale equipment abnormal part cooperative detection method

Abstract

The invention belongs to the technical field of intelligent equipment fault detection, and discloses a multi-robot-based large-scale equipment abnormal component cooperative detection method. According to the large complex intelligent equipment abnormal component detection method based on the multi-robot and cloud edge collaborative architecture, clustering, an A * algorithm, shortest route distribution and a Kalman filtering fusion algorithm are utilized, and multi-robot rapid inspection and accurate abnormal component detection are achieved. Through multi-robot uniform distributed inspection, global three-dimensional coordinate system establishment, abnormal point secondary inspection, abnormal point clustering analysis and screening, task list distribution, observable path division and fitting and optimal detection scheme construction, multi-angle and multi-mode abnormal data are finally fused, and an accurate detection result is obtained. The multiple robots are evenly distributed around the large complex intelligent equipment at equal intervals for inspection, the detection efficiency is improved, the coverage range is wider, and abnormal information can be fed back in time.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent equipment fault detection, and in particular to a collaborative detection method for abnormal parts of large equipment based on multiple robots. Background Art

[0002] Large-scale intelligent equipment usually operates in extreme environments such as high temperature and high pressure that are difficult for people to access. To ensure its efficient and stable operation, industrial inspection robots play an indispensable role. Large-scale and complex industrial intelligent equipment is prone to failure and extremely harmful. To ensure its safe, reliable, continuous and stable operation, inspection robots are needed to achieve operation prediction and precise operation and maintenance of such equipment. The detection of abnormal parts of large-scale and complex intelligent equipment is an important part of promoting industrial intelligent operation and maintenance, which helps to improve factory production efficiency, product quality, and reduce factory safety risks and labor costs. However, the efficient and accurate detection of abnormal parts of large-scale and complex intelligent equipment is still a problem that needs to be solved urgently.

[0003] Cloud-edge collaboration is a close cooperation between cloud computing and edge computing. Through the collaborative work of the cloud and the edge, optimal allocation and efficient utilization of resources are achieved. The cloud-edge collaboration framework includes three main capabilities: resource collaboration, data collaboration, and service collaboration. Resource collaboration involves the sharing and scheduling of computing power, storage and other resources between the cloud and the edge; data collaboration focuses on the optimization of processes such as data collection, processing, analysis and storage; and service collaboration emphasizes the seamless connection and unified management of cloud and edge services. It is critical to perform collaborative detection of abnormal components of complex equipment based on the cloud-edge collaboration architecture.

[0004] However, at present, since large-scale and complex industrial intelligent equipment is extremely prone to failure and extremely harmful, manual inspections are time-consuming and cannot achieve the purpose of real-time monitoring. Moreover, when large-scale and complex intelligent equipment is in extreme environments, manual inspections are relatively dangerous. The coverage rate of a single robot is low, and the time for a cycle detection is long, which also cannot achieve the purpose of real-time monitoring. Summary of the invention

[0005] The purpose of the present invention is to provide a collaborative detection method for abnormal parts of large equipment based on multiple robots in order to solve the above-mentioned problems, thereby improving the existing intelligent equipment detection technology, in which a single robot cannot detect abnormalities in a timely manner and the efficiency of collaborative detection by multiple robots is low.

[0006] The present invention proposes a collaborative detection method for abnormal parts of large-scale equipment based on multiple robots. The method improves the efficiency of abnormal part detection by collaborative inspections of multiple robots around large-scale equipment. At the same time, the method is based on a cloud-edge collaborative architecture, and uploads multimodal abnormal data of a large number of abnormal parts collected at different perspectives to the cloud by multiple robots (edge ​​ends), and then performs multimodal and multi-perspective abnormal data fusion detection through the cloud, thereby improving the accuracy of abnormal detection. On the basis of the above, the present invention uses robots to find and mark the optimal observable position points of abnormal parts around large-scale equipment after the abnormality occurs, and integrates and divides all observable position points to obtain the optimal observable path. Finally, the robot is assigned tasks and path planning is performed to go to different observable paths to collect abnormal data, thereby improving the accuracy of abnormal detection and improving the system operation efficiency.

[0007] The technical solution adopted by the present invention is as follows: a collaborative detection method for abnormal parts of large equipment based on multiple robots, wherein the number of robots is set to n (n≥2) according to the size and complexity of the large equipment, and the robots are numbered from 1 to n in sequence, and the robots are evenly distributed around the large equipment for inspection; each robot is equipped with three sensors: a visible light camera, a thermal infrared camera and an audio-visual camera, and the robots scan the equipment to obtain visible light images during the inspection process. vis 、Thermal infrared image I tir Harmony Image I ac ; The system is based on a cloud-edge collaborative architecture, with the robot as the edge end collecting I vis ,I tir and I ac After the three types of image data are obtained, the image data will be uploaded to the cloud with corresponding image detection and analysis functions for processing in real time. The cloud can perform anomaly detection based on image processing on the received image data to determine whether any abnormality has occurred in the large equipment. Among them, the collaborative detection of abnormal parts by multiple robots includes two stages, the first stage is the inspection stage, and the second stage is the collaborative detection stage.

[0008] Furthermore, the inspection phase includes: patrol detection, in which the robot inspects the equipment in a specified inspection direction, and establishes an XYZ three-dimensional coordinate system based on the two-dimensional XY horizontal plane where the large equipment base is located; wherein, in the top-down view of the large equipment, the counterclockwise inspection direction is set as the positive direction, and the clockwise inspection direction is set as the reverse direction; abnormal data collection, when abnormal data is collected, after the cloud receives the image data collected by a single sensor of the i-th robot at time t, it finds that there is an abnormality in the component through the image detection and analysis function, then the i-th robot uploaded at the timestamp of time t is sent to the cloud. vis ,I tir and I acThe three types of image data are fused through Kalman filtering to obtain a synthetic image, and then the image detection and analysis function is used to determine whether the component is abnormal. After determining that the component is abnormal, the cloud solves and marks the observed abnormal position coordinates P in the global XYZ three-dimensional coordinate system. it (x it ,y it ,z it ), solve and mark the position coordinates R of the i-th robot in the two-dimensional XY plane coordinate system it (x i ' t ,y i ' t ), and record the one-to-one correspondence, and record the mapping relationship as f:R→P; when the image data uploaded by the i-th robot at time t is detected to be normal in the cloud, no position coordinates are solved and marked; the observable point is solved, and the reliable abnormal point position is selected through the clustering algorithm in the cloud, and the observed abnormal position coordinates P it Clustering is performed and P is transformed into it Core point P Δ , boundary points P′ and noise points P″, and several clusters are obtained and each cluster Center of gravity The core point P Δ P in the cluster it And the number of abnormal position coordinates within the spherical neighborhood ε of the point with eps as radius is greater than Pts Cor , given the radius eps and the minimum number of points Pts in the neighborhood of the core point Cor are all initial setting values; the boundary point P′ represents the P in the cluster excluding the core point it ; Noise point P″ indicates P that is not in the cluster it Different clusters Used to indicate the different abnormal components detected in this round of inspection, each center of gravity Considered as corresponding cluster The abnormal point of the component is re-labeled as P anom ; At the same time, the cloud will be different abnormal points P anom Randomly put them into the task list one by one, and in the subsequent collaborative detection stage, outliers are taken from the front to the back in the task list for detection; if there is an outlier point P anom Cluster The coordinates of the outlier points P it The number is less than the minimum number of points Pt required to meet the collaborative detection requirement. sClus , then the outlier point P anom Do not add to the task list; the cloud will then The coordinates of the abnormal position P itWith the inverse mapping f -1 :P→R, derive the coordinate system of the two-dimensional XY plane with P it The corresponding robot position coordinates R it , then the position coordinates of these robots are the outlier points P anom The observable position of these robots is re-marked as the observable position point Q k (x k ,y k ); Observable path solving, in the cloud, each abnormal point P is solved by the observable path solving algorithm anom The observable path l 1 ,l 2 ,…,l j ,…, the angle θ of a point in the two-dimensional plane is set as the angle between the point and the origin and the positive direction of the x-axis; the observable path solution algorithm first calculates each observable position point Q in the two-dimensional XY plane coordinate system k The corresponding angle θ (0≤θ<2π), and each observable position point Q k Sort by angle θ, with smaller angles in front and larger angles in the back. If the χ+1th Q k The angle θ χ+1 Subtract the xth Q k The angle θ χ If the maximum angle difference Δθ of adjacent observable points on the same path is less than or equal to the setting, then the χ+1th Q k and the xth Q k Considered as being on the same observable path l j Otherwise, the χ+1th Q k and the xth Q k In different observable paths, different observable paths are divided and fitted to obtain observable paths.

[0009] Furthermore, in the abnormal data collection step, the time when the i-th robot first discovers the abnormality is recorded as time 0, and the position coordinate of the i-th robot at time 0 is R i0 (x i ' 0 ,y i ' 0 ), with the origin as the endpoint and passing through point R i0 (x i ' 0 ,y i ' 0 ) is set as the reference position. When the i-th robot reaches the ray α times, that is, after a robot finds an abnormality for the first time, all robots continue to inspect the large equipment α times to complete the abnormal position coordinate P it (x it ,yit ,z it ) and the robot position coordinates R it (x i ' t ,y i ' t ) is marked and uploaded, the abnormal data collection is completed; after completing the α-circle detour, the robot position coordinates are initially obtained in the cloud. and the set of anomaly location coordinates And the mapping relationship f:R→P between set R and set P.

[0010] Furthermore, the collaborative detection phase includes: calculating the optimal detection solution. This step is performed in the cloud. First, the abnormal point P needs to be selected from the task list. anom Take out from the front to the back for testing, and record the current test object as P a ' nom , let the detection object P a ' nom There are m observable paths, and at the same time, there are m observable paths l j The observable points with the smallest and largest angles are set as endpoints, and the observable point with the smallest angle is recorded as The observable point with the largest angle is recorded as Through the A* algorithm, find the path l from the current position of the i-th robot j Endpoints and The shortest route distance is recorded as and The A* algorithm is used to calculate the distance cost from the robot's current position to the endpoints of each path. This allows us to obtain the distances of each robot to the observable points with the smallest and largest angles on each path. The cost matrix C is constructed using the distances of each robot to the observable points with the smallest angles on each path. 1 , construct the cost matrix C based on the distance of each robot to the observable point with the smallest angle on each path 2 ; The final cost matrix C = g(C 1 ,C 2 ), where g represents C[p][q]=min(C 1 [p][q],C 2 [p][q]), that is, the p-th row and q-th column element of the final cost matrix C is equal to C 1 and C 2The minimum value of the element in the pth row and the qth column; after obtaining the cost matrix, the optimal detection scheme of collaborative detection is solved through the shortest route allocation algorithm, so that the total path of all robots is the shortest; the optimal detection scheme is executed, and each robot is assigned to move along the route of the optimal detection scheme. After reaching the observable path, abnormal data of abnormal parts are collected, and the collected data is uploaded to the cloud in real time. The cloud uses Kalman filtering to fuse the collected multi-modal and multi-perspective abnormal data to obtain the abnormal point P a ' nom The final detection result; complete the current detection target P a ' nom After completing the detection task, check whether there are any tasks to be detected in the task list P anom , if it exists, then take out the new outlier point at the front as the target of a new round of detection; wait for the outlier point P in the list anom After all the tests are completed, the collaborative testing phase ends.

[0011] Furthermore, the shortest route allocation algorithm comprises the following steps:

[0012] S1: Subtract the minimum value in each row of the cost matrix, and then subtract the minimum value in each column from the minimum value in the column, that is, ensure that there is at least one zero in each row and column;

[0013] S2: Use as few row markers and column markers as possible to cover all zeros in the matrix. One row marker will cover the entire row elements of the matrix, and one column marker will cover the entire column elements of the matrix. If the sum of the number of row and column markers is less than min(n,m), proceed to step S3; if the number of row and column markers is equal to min(n,m), find the current optimal solution and execute step S4;

[0014] S3: Find the minimum value ξ among the elements not covered by the row and column marks, subtract ξ from all the elements not covered by the row and column marks, add ξ to the elements covered twice by the row and column marks, remove the row and column marks, and re-execute step S2;

[0015] S4: After obtaining the current optimal solution, due to detecting different abnormal points P anom When the relationship between the number of robots n and the number of observable paths m is different, the system operation scheme is also different. Compare the sizes of n and m and choose different schemes to run;

[0016] S5: The total length of the current detection route of each robot and the distance from the end point of the current detection route to the end point of the remaining path are added together as the distance cost for multiple robots to reach the remaining observable path. At the same time, the new cost matrix is ​​reconstructed in the same way as the above cost matrix construction method, and step S1 is re-executed to solve the optimal detection plan for collaborative detection.

[0017] Further, in step S4, different solutions are selected by comparing the magnitudes of n and m, which specifically include the following:

[0018] When n > m, the current optimal solution is taken as the optimal detection solution for collaborative detection, and m robots are dispatched to m paths respectively to perform collaborative detection on the current detection target P a ′ nom ; the remaining n - m robots continue to conduct patrol inspections to collect abnormal data of large equipment;

[0019] When n = m, the current optimal solution is taken as the optimal detection solution for collaborative detection, and n robots are dispatched to n paths respectively to perform collaborative detection on the current detection target P a ′ nom ;

[0020] When n < m, the current optimal solution is taken as a partial solution of the current shortest total path scheme for collaborative detection, and n robots are dispatched to n paths respectively to perform collaborative detection on the current detection target P a ′ nom ; record the current detection route of the robot as the route to the specified path plus the route for detection on the specified path. Assume that the i-th robot is dispatched to the observable path l j , and it first passes through the observable point with the smallest angle and the detection end point is the observable point with the largest angle Calculate the total length s of the current detection route of the i-th robot, that is, from the current position to the detection end point using the A* algorithm, and the shortest route distance from the current route detection end point i to the end point of the remaining path and and are respectively denoted as and ; then step S5 is executed.

[0021] Further, the distances of each robot to the observable points with the smallest and largest angles of each path are used to construct a cost matrix C 1 with the distances of each robot to the observable points with the smallest angles of each path, and a cost matrix C 2 is constructed with the distances of each robot to the observable points with the largest angles of each path:

[0022]

[0023]

[0024] wherein, is the distance from the current position of the i-th robot to the end point j of path l The shortest route distance, The path from the current position of the i-th robot to the path l j Endpoints The shortest route distance.

[0025] Furthermore, when the abnormal data is collected, when the robot detects multiple abnormal points at time t, the robot position and the abnormal point position are marked respectively; if the i-th robot detects three abnormal points at time t, the abnormal position coordinates are marked as P it 1 (x it ,y it ,z it ), P it 2 (x it ,y it ,z it ) and P it 3 (x it ,y it ,z it ), and the robot position is R it 1 (x i ' t ,y i ' t ), R it 2 (x i ' t ,y i ' t ) and R it 3 (x i ' t ,y i ' t ) and record the one-to-one correspondence.

[0026] Furthermore, in the observable point calculation, the core point P whose distance is less than eps is Δ and the boundary point P′ are clustered in the same initial cluster According to the core point P Δ and the weights of the boundary points P′ to calculate the initial cluster Center of gravity Core point P Δ The weight is set to τ times of the boundary point P′; get the initial cluster center Afterwards, filter the clusters Center and center of gravity The core points P of the first v% with the closest Euclidean distance Δ and boundary point P′, the filtered core point P Δand the boundary point P′ form a cluster representing abnormal components And calculate the clusters Center of gravity

[0027] Furthermore, the current optimal solution in step S2 is an adjusted cost matrix that satisfies the number of row and column markers equal to min(n,m). The specific solution includes the following:

[0028] If n≤m, then each row of the cost matrix has at least one zero element, and at least one zero element can be selected from each row to ensure that the selected zero elements are in different columns. The row and column where the selected zero element is located indicates that the robot corresponding to the row should go to the observable path endpoint corresponding to the column.

[0029] If n>m, then each column of the cost matrix has at least one zero element, and at least one zero element can be selected from each column to ensure that the selected zero elements are in different rows. At this time, the row and column where the selected zero element is located indicates that the robot corresponding to the row should go to the observable path endpoint corresponding to the column;

[0030] Furthermore, the mapping relationship between the elements in the set R and the set P is a bijection, and the mapping relationship is denoted as f:R→P, and there is an inverse mapping f -1 :P→R.

[0031] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0032] 1. Distribute multiple robots evenly around large equipment at equal intervals, and conduct collaborative inspections on the equipment in two stages: patrol inspection and collaborative inspection. This avoids the problem that a single robot may not be able to detect anomalies in time, improves inspection efficiency, and can cover a wider range. At the same time, it improves the response efficiency of the system and enables collaborative inspections by multiple robots.

[0033] 2. Determine the number of robots based on a comprehensive judgment of the size and complexity of the on-site equipment. Each robot is equipped with three sensors: a visible light camera, a thermal infrared camera, and an acoustic imager, which respectively obtain visible light images, thermal infrared images, and acoustic images. At the same time, the system is based on a cloud-edge collaborative architecture. After the robot collects three types of image data as the edge end, it uploads the image data to the cloud with corresponding image detection and analysis functions for processing in real time. The cloud can perform anomaly detection based on image processing on the received image data to determine whether an abnormality has occurred in large equipment. The number of robots is set according to the equipment situation so that the fault condition can be judged in a shorter time, the optimal observable path can be obtained, and improvement measures can be made. While improving the accuracy of anomaly detection, it can also improve the detection efficiency of the system.

[0034] 3. By clustering the abnormal position coordinates, the collected abnormal position coordinates are screened to obtain more reliable abnormal position coordinates, and more reliable observable position points are inferred from them; and the angle of the observable position point obtained above on the two-dimensional horizontal plane is calculated through the observable path solution algorithm, and then the difference in the angle values ​​of adjacent points is compared with the maximum difference in the angles of adjacent observable points on the same path set, and divided into different observable paths, that is, the observable path of each abnormal component is obtained, which improves the accuracy of detection.

[0035] 4. The optimal detection scheme of the system is calculated through the A* algorithm and the shortest route allocation algorithm. According to the optimal detection scheme, each robot is assigned to its corresponding observable path for collaborative detection. The scheme also satisfies the shortest overall running distance of multiple robots, further improving the overall operation efficiency of the system.

[0036] 5. By having multiple robots go to their respective observable paths for collaborative detection, multi-view and multi-modal data are collected, and these multi-view and multi-modal abnormal image data are fused and detected in the cloud through Kalman filtering, which improves the accuracy of the system in detecting equipment abnormalities. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a flow chart of the overall method of the present invention;

[0038] Figure 2 It is a schematic diagram of the cloud-edge collaborative architecture of the present invention;

[0039] Figure 3 is a schematic diagram of a multi-robot inspection of the present invention;

[0040] Figure 4 is a flow chart of the inspection phase method of the present invention;

[0041] Figure 5 It is a global XYZ three-dimensional coordinate system diagram of the present invention;

[0042] Figure 6 This is a top view distribution diagram of the robot inspection of the present invention;

[0043] Figure 7 It is a flow chart of the clustering algorithm of the present invention;

[0044] Figure 8 is a flow chart of the observable path solving algorithm of the present invention;

[0045] Fig. 9 is a flow chart of the shortest route allocation algorithm of the present invention;

[0046] Fig.10 is a flow chart of the collaborative detection phase of the present invention;

[0047] Fig.11 is a mapping relationship diagram between abnormal position coordinates and robot position coordinates of the present invention;

[0048] Fig.12 is a schematic diagram of the observable path solving algorithm of the present invention;

[0049] Fig.13 It is a schematic diagram of the path endpoints and the shortest route of the present invention;

[0050] Fig.14 is a distance cost table from each robot to each path endpoint of the present invention;

[0051] Fig.15 It is a schematic diagram of two types of distances for the robot of the present invention to reach the remaining path;

[0052] Fig.16 It is a cost table of the distances from each robot to the remaining path breakpoints in the new detection route of the present invention;

[0053] Fig.17 This is an example of an embodiment of the present invention. Figure 1 ;

[0054] Fig.18 This is an example of an embodiment of the present invention. Figure 2 ;

[0055] Fig.19 This is an example of an embodiment of the present invention. Figure 3 ;

[0056] Fig. 20 This is an example of an embodiment of the present invention. Figure 4 . DETAILED DESCRIPTION

[0057] The present invention will be described in detail below in conjunction with the accompanying drawings.

[0058] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0059] like Figure 1-3 As shown, a collaborative detection method for abnormal parts of large equipment based on multiple robots is shown. The system is based on a cloud-edge collaborative architecture. Through the comprehensive application of clustering algorithm, observable path solving algorithm, A* algorithm, shortest route allocation algorithm and Kalman filter fusion algorithm, multiple robots can quickly run and complete the detection of abnormal parts when abnormalities are found during the inspection process, and the detection results are accurate and reliable. The technical solution is as follows:

[0060] First, multiple robots are evenly distributed around large equipment for inspection, and the collected image data is uploaded to the cloud in real time for image detection and analysis, which improves the system's ability to detect abnormalities.

[0061] Secondly, the global XYZ three-dimensional coordinate system is established based on the two-dimensional XY horizontal plane where the large equipment base is located. The robot position coordinates are marked on the two-dimensional XY horizontal plane to solve and fit the observable path. The observed abnormal position coordinates are marked on the global XYZ three-dimensional coordinate system to solve the abnormal parts and abnormal point positions. And the robot position coordinates correspond to the abnormal position coordinates one by one;

[0062] Then, in order to more accurately locate the abnormal point and determine the observable path, after the abnormality is first discovered in the cloud, the robot continues to patrol the large equipment for α circles to collect more abnormal position coordinates and robot position coordinates;

[0063] Subsequently, the cloud uses a clustering algorithm to group the collected and marked abnormal position coordinates into multiple clusters, remove noise points, and classify the points in the cluster into core points and boundary points for calculating the cluster centroid. The reliable abnormal position coordinates in the cluster are then further screened. The screened clusters represent abnormal parts, and the screened cluster centroids represent abnormal points. In the end, abnormal parts and abnormal points are obtained.

[0064] At the same time, the cloud puts different abnormal points into the task list in random order, and takes out the task points for detection in the collaborative detection stage, and when the number of abnormal position coordinates contained in the cluster where the abnormal point is located is less than the set value Pts Clus , not added to the task list.

[0065] After initially obtaining the abnormal parts and abnormal points, the cloud uses the one-to-one correspondence between the abnormal position coordinates and the robot position coordinates to find the observable position points of each abnormal part, and then calculates the angle difference between each adjacent observable position point on the two-dimensional XY horizontal plane, and compares it with the set maximum angle difference Δθ of adjacent observable points on the same path, dividing and fitting different observable paths;

[0066] After obtaining the observable paths of each abnormal component, the cloud uses the A* algorithm to calculate the distance cost from the robot's current position to each path endpoint, and takes the minimum distance cost of each robot to reach each path to construct a cost matrix. The cost matrix is ​​then solved by the shortest path allocation algorithm to obtain the optimal detection solution for the system.

[0067] Finally, each robot goes to the designated path to collect abnormal data of abnormal parts. The cloud uses Kalman filtering to fuse the multi-angle and multi-modal abnormal data collected by each robot to obtain a more accurate final detection result.

[0068] like Figure 4 As shown, the patrol phase includes four steps: patrol detection, abnormal data collection, observable point solution and observable path solution.

[0069] Cruise detection: Take the two-dimensional XY horizontal plane where the large equipment base is located as the reference, establish Figure 5 The global XYZ three-dimensional coordinate system shown in the figure, where the cylinder in the figure represents the circumscribed cylinder of the large equipment.

[0070] like Figure 6 As shown, patrol inspection is performed by the robot in a specified patrol direction, with the counterclockwise patrol direction being the positive direction and the clockwise patrol direction being the negative direction, patrolling around the equipment.

[0071] Abnormal data collection: When the cloud receives the image data collected by a single sensor of the i-th robot at time t, it finds that there is an abnormality in the component through the image detection and analysis function, and then sends the i-th robot uploaded at the timestamp of time t to the cloud. vis ,I tir and I ac The three types of image data are fused through Kalman filtering to obtain a synthetic image, and then the image detection and analysis function is used to determine whether the component is abnormal. After determining that the component is abnormal, the cloud solves and marks the observed abnormal position coordinates P in the global XYZ three-dimensional coordinate system. it (x it ,y it ,z it ), solve and mark the position coordinates R of the i-th robot in the two-dimensional XY plane coordinate system it (x i ' t ,y i ' t ) and record the one-to-one correspondence.

[0072] When the cloud detects multiple abnormal points in the image data of the i-th robot at time t through the image detection and analysis function, the abnormal position coordinates are also marked in the global XYZ three-dimensional coordinate system and the position coordinates of the i-th robot are marked in the two-dimensional XY plane coordinate system, and the one-to-one correspondence is recorded. For example, when the i-th robot detects three abnormal points at time t, they are marked as P it 1 (x it ,y it ,z it ) and R it 1 (x i ' t ,y i ' t ), P it 2 (x it,y it ,z it ) and R it 2 (x i ' t ,y i ' t ), P it 3 (x it ,y it ,z it ) and R it 3 (x i ' t ,y i ' t ).

[0073] When the image data uploaded by the i-th robot at time t is detected to have no abnormalities in the cloud, no position coordinates are solved and marked.

[0074] After all the data uploaded by the robot are processed by the cloud through image detection analysis and position recording, the recorded abnormal position coordinates, robot position coordinates and corresponding relationships are as follows: Fig.11 shown.

[0075] Fig.11 Indicates that time 0 is the time when the system first discovers the abnormality of large equipment, and the first abnormality is discovered, solved and marked by the i-th robot; t 1 At time t, robot 1 detects three abnormal points; 2 At time 0, the position coordinates of robot No. i at time 0 are R i0 (x i ' 0 ,y i ' 0 ), with the origin as the endpoint and passing through point R i0 (x i ' 0 ,y i ' 0 ) is set as the reference position. When the i-th robot reaches the ray α times, that is, after a robot finds an abnormality for the first time, all robots continue to inspect the large equipment α times to complete the abnormal position coordinate P it (x it ,y it ,z it ) and the robot position coordinates R it (x i ' t ,y i ' t ) is marked and uploaded, the abnormal data collection is completed.

[0076] After completing the α-circle orbit, now in the cloud, Fig.11 The robot position coordinates have been preliminarily obtained. and the set of anomaly location coordinates The mapping relationship between the elements in the set R and the set P is a bijection, and the mapping relationship is f:R→P, and there is an inverse mapping f -1 :P→R.

[0077] Observable point solution: Observable point solution is to select reliable abnormal point locations through clustering algorithm in the cloud, and calculate the observed abnormal location coordinates P it Clustering is performed and P is transformed into it Core point P □ , boundary points P′ and noise points P″, and several clusters are obtained and each cluster Center of gravity The core point P □ P in the cluster it And the number of abnormal position coordinates within the spherical neighborhood ε of the point with eps as radius is greater than Pts Cor , given the radius eps and the minimum number of points Pts in the neighborhood of the core point Cor are all initial setting values; the boundary point P′ represents the P in the cluster excluding the core point it ; Noise point P″ indicates P that is not in the cluster it Different clusters Used to indicate the different abnormal components detected in this round of inspection, each center of gravity Considered as corresponding cluster The abnormal point of the component is re-labeled as P anom .in, Figure 7 This is the flow chart of clustering algorithm.

[0078] At the same time, the cloud will be different abnormal points P anom Randomly put them into the task list one by one, and then take outliers from the task list from the beginning to the end for detection in the subsequent collaborative detection. anom Cluster The coordinates of the outlier points P it The number is less than the minimum number of points Pts required to meet collaborative detection Clus , then the outlier point P anom Not added to the to-do list.

[0079] The cloud then The coordinates of the abnormal position P it With the inverse mapping f -1 :P→R, derive the coordinate system of the two-dimensional XY plane with P itThe corresponding robot position coordinates R it , then the position coordinates of these robots are the outlier points P anom The observable position of these robots is re-marked as the observable position point Q k (x k ,y k ).

[0080] like Figure 8 As shown, observable path solution: In the cloud, each abnormal point P is solved by the observable path solution algorithm. anom The observable path l 1 ,l 2 ,…,l j ,…. It is agreed that the angle θ of a point in a two-dimensional plane is the angle between the point and the origin and the positive direction of the x-axis. The observable path solution algorithm first calculates each observable position point Q in the two-dimensional XY plane coordinate system. k The corresponding angle θ (0≤θ<2π), and each observable position point Q k Sort by angle θ, with smaller angles in front and larger angles in the back. If the χ+1th Q k The angle θ χ+1 Subtract the xth Q k The angle θ χ If the maximum angle difference Δθ of adjacent observable points on the same path is less than or equal to the setting, then the χ+1th Q k and the xth Q k Considered as being on the same observable path l j Otherwise, the χ+1th Q k and the xth Q k In different observable paths.

[0081] The observable path is finally obtained through the observable path solving algorithm, such as Fig.12 As shown. It only shows the solved l 1 and l j Two observable paths, the remaining paths are omitted in the figure; They are the observable positions ranked at the χ-1th, χth, and χ+1th according to the angle θ, as shown in the figure. and The corresponding angle θ χ With θ χ-1 The difference is greater than Δθ, and The corresponding angle θ χ+1 With θ χ The difference is less than Δθ; the shaded part represents the blocked area, that is, the corresponding path is an unobservable path.

[0082] like Fig.10 As shown, the collaborative detection phase includes two steps: calculating the optimal detection plan and executing the optimal detection plan.

[0083] The calculation of the optimal detection solution is performed in the cloud. First, the outlier point P needs to be added to the task list. anom Take out the objects from the front to the back for testing. The object currently taken out is called P. a ' nom , let the detection object P a ' nom There are m observable paths. j The observable points with the smallest and largest upper angles are agreed to be endpoints, denoted as and

[0084] The A* algorithm is a heuristic search algorithm used to find the shortest path from the start point to the end point in a graph. It selects the optimal path by combining the actual cost of the path with the heuristic estimate. In the collaborative detection task, the constraints of the A* algorithm include: the path cannot enter the area of ​​large equipment; the standard heuristic function Manhattan distance is used. With these constraints, the A* algorithm can efficiently calculate the shortest route distance from the robot position to the path endpoint.

[0085] The A* algorithm is used to find the path l from the current position of the i-th robot j Endpoints and The shortest route distance is recorded as and like Fig.13 shown.

[0086] The distance cost from the robot's current position to each path endpoint is calculated using the A* algorithm, such as Fig.14 shown.

[0087] In this way, we can obtain the distances of each robot to the observable points with the smallest and largest angles on each path, and construct the cost matrix C based on the distances of each robot to the observable points with the smallest angles on each path. 1 , construct the cost matrix C based on the distance of each robot to the observable point with the largest angle on each path 2 :

[0088]

[0089]

[0090] The final cost matrix C = g(C 1 ,C 2 ), where g represents C[p][q]=min(C 1 [p][q],C2 [p][q]), that is, the p-th row and q-th column element of the final cost matrix C is equal to C 1 and C 2 Minimum value of the element in row p and column q.

[0091] After obtaining the cost matrix, the optimal detection plan for collaborative detection is solved through the shortest path allocation algorithm, so that the total path of all robots is the shortest.

[0092] like Fig. 9 As shown in Figure 1, the shortest path allocation algorithm has the following five steps:

[0093] The first step is to subtract the minimum value in each row of the cost matrix, and then subtract the minimum value in each column, that is, to ensure that there is at least one zero in each row and column;

[0094] The second step is to cover all zeros in the matrix with as few row and column markers as possible. One row marker will cover the entire row of the matrix, and one column marker will cover the entire column of the matrix. If the sum of the number of row and column markers is less than min(n,m), proceed to the third step; if the number of row and column markers is equal to min(n,m), find the current optimal solution (the current optimal solution means: take out min(n,m) zeros from the matrix, and these zeros occupy different rows and columns of the matrix. The solution is that the robot represented by the row where these zeros are located goes to the path endpoint represented by the column where they are located, and then goes to the other endpoint of the same path after arriving), and proceed to the fourth step;

[0095] The third step is to find the minimum value ξ among the elements not covered by the row and column markers, subtract ξ from all the elements not covered by the row and column markers, add ξ to the elements covered twice by the row and column markers, remove the row and column markers, and re-execute the second step;

[0096] Step 4: After obtaining the current optimal solution, since the relationship between the number of robots n and the number of observable paths m is different when detecting different abnormal points, the system operation scheme is also different, so the following discussion is carried out:

[0097] ① When n>m, the current optimal solution is used as the optimal detection solution for collaborative detection, and m robots are dispatched to m paths to detect the current detection target P. a ' nom Conduct collaborative detection, and the remaining nms continue to inspect and collect abnormal data of large equipment;

[0098] ② When n = m, the current optimal solution is used as the optimal detection solution for collaborative detection, and n robots are dispatched to n paths to detect the current detection target P. a ' nom Conduct collaborative testing;

[0099] ③ When n < m, the current optimal solution is used as a partial solution for the overall shortest path plan of the current collaborative detection. Then, n robots are dispatched to n paths respectively to perform collaborative detection on the current detection target P a ′ nom The current detection route of the robot is recorded as the route to the specified path plus the route for detection on the specified path. Assume that the i-th robot is dispatched to the observable path l j , and it first passes through The detection end point is Calculate the total length s of the current detection route of the i-th robot, that is, from the current position to the detection end point using the A* algorithm i and the shortest route distances from the detection end point of the current route to the endpoints of the remaining paths are recorded as and respectively, as shown in and the distances, and then go to the fifth step Fig.15 as shown

[0100] Step 5: Add the sum of the total length of the current detection route and the distance from the detection end point of the current detection route to the endpoints of the remaining paths as the distance cost of the new detection route for the multi-robot. As shown in Fig.16 at the same time, similar to the method of constructing the cost matrix above, reconstruct a new cost matrix and re-execute the first step to solve the optimal detection plan for collaborative detection

[0101] The current optimal solution in the second step described above, which is the adjusted cost matrix, satisfies that the number of row and column labels is equal to min(n, m), and specifically includes the following

[0102] If n ≤ m, at this time, each row of the cost matrix has at least one zero element, and at least one zero element is selected from each row to ensure that the selected zero elements are in different columns. At this time, the rows and columns where the selected zero elements are located indicate that the robots corresponding to the rows should go to the endpoints of the observable paths corresponding to the columns

[0103] If n > m, at this time, each column of the cost matrix has at least one zero element, and at least one zero element is selected from each column to ensure that the selected zero elements are in different rows. At this time, the rows and columns where the selected zero elements are located indicate that the robots corresponding to the rows should go to the endpoints of the observable paths corresponding to the columns

[0104] Through the above shortest route allocation algorithm, the optimal detection plan for system collaborative detection is obtained

[0105] When executing the optimal detection plan, each robot is assigned to move along the route of the optimal detection plan. After reaching the observable path, abnormal data of abnormal components are collected and uploaded to the cloud in real time. The cloud uses Kalman filtering to fuse multimodal abnormal data from different angles to obtain more accurate detection results.

[0106] The Kalman filter algorithm is a recursive estimation algorithm used to provide an optimal estimate of the system state in a dynamic system by combining the system model and measurement data. It is implemented in two main steps: the prediction step predicts the state of the next moment based on the system model, and the update step corrects the prediction result based on the actual measurement value, thereby achieving an accurate estimate of the system state in the presence of noise.

[0107] Complete the current detection target P a ' nom After completing the detection task, check whether there are any tasks to be detected in the task list P anom , if it exists, the new outlier point at the front is taken out as the target of a new round of detection.

[0108] After all the abnormal points in the list have been detected, the collaborative detection phase ends.

[0109] In this embodiment, the number of robots is set to n=4, the number of inspection circles of all robots around large equipment after the first abnormality is found is α=3, the given radius eps=15cm, and the minimum number of points Pts in the neighborhood of the core point Cor =10, minimum number of points required for collaborative testing Pts Clus =5000, the ratio of the core point weight to the boundary point weight is τ=2, the cluster screening ratio is v=95%, and the maximum difference in angles of adjacent observable points on the same path is Δθ=1°.

[0110] Reference Fig.17 The application scenario of the present invention is the large equipment shown in Figure 2. In the figure, the numbers 2-5 are four inspection robots placed around the large equipment, which are evenly spaced around the large equipment and patrol along the directions indicated by arrows 6-9 respectively;

[0111] During the multi-robot inspection process, the collected I vis ,I tir and I ac The three types of image data are uploaded to the cloud. When the cloud finds that a robot has an abnormality at a certain moment, all robots continue to patrol the large equipment three times. During the patrol process, the robot position coordinates, the abnormal position coordinates, and the corresponding relationship between the two coordinates are marked at the position where the abnormal parts can be observed.

[0112] After the three circles are completed, the cloud uses a clustering algorithm based on the given radius eps = 15 and the minimum number of points Pts in the neighborhood of the core point. Cor =10 classifies the abnormal position coordinates into core points, boundary points, and noise points. At the same time, the reliable abnormal point positions are screened according to the ratio of the core point weight to the boundary point weight τ = 2 and the cluster screening ratio v = 95%, and the abnormal component position and the abnormal point position are preliminarily obtained. Based on this, the corresponding relationship between the robot position coordinates and the abnormal position coordinates is reversed to obtain more reliable observable position points;

[0113] Reference Fig.18 , using the above observable position points, calculate the angle of each observable position point on the two-dimensional XY horizontal plane in the cloud (same as the above agreement, the angle θ of a point in the two-dimensional plane is the angle between the point and the origin and the positive direction of the x-axis). The angle is θ χ =90°, the position point can be observed at the same time Its adjacent observable position points The difference in the angle values ​​of is greater than the maximum difference in the angles of adjacent observable points on the same path, Δθ = 1°, so it is divided into different paths, and finally five observable paths l are obtained. 1 , l 2 , l 3 , l 4 , l 5 ;

[0114] Reference Fig.19 , using the A* algorithm and the shortest path allocation algorithm, in the cloud, the system collaboratively detects the current optimal solution through the first three steps of the shortest path allocation algorithm and assigns the robot to the corresponding path. The current optimal solution is: robot No. 1 goes to the observable path l 2 , robot No. 2 goes to l 3 , robot No. 3 goes to l 5 , robot No. 4 goes to l 1 ;

[0115] Reference Fig. 20 In the fourth step of the shortest path allocation algorithm, since the number of observable paths is greater than the number of robots, the shortest path allocation algorithm also needs to calculate the distance cost of the remaining paths, and then calculate the new cost matrix from the fifth step, and then return to the first step of the shortest path allocation algorithm for iterative calculation to solve a new round of detection solutions until the optimal detection solution of the system is obtained and the assignment information of the robots is updated. In this example, the optimal detection solution is as follows Fig. 20 Shown: Robot No. 1 goes to the observable path l 2 , robot No. 2 goes to l 3 Then go to l 4, robot No. 3 goes to l 5 , robot No. 4 goes to l 1 ;

[0116] Each robot goes to the corresponding path to collect information about abnormal components and uploads it to the cloud in real time. The cloud uses Kalman filtering to fuse multi-modal and multi-perspective abnormal data to obtain the final detection results.

[0117] The present invention provides a fast and accurate anomaly detection method for collaborative detection of abnormal parts of large equipment by comprehensively applying a clustering algorithm, an observable path solving algorithm, an A* algorithm, a shortest route allocation algorithm and a Kalman filter fusion algorithm; at the same time, the corresponding relationship between the robot position and the simultaneously observed fault point position is used to solve a better observable position point, and the observable position point is converted into an observable path through the observable path solving algorithm; the A* algorithm and the shortest route allocation algorithm are combined to solve the problem of how to allocate observable paths when the number of robots and the number of observable paths are in different size relationships.

[0118] By clustering the abnormal position coordinates, the collected abnormal position coordinates are screened to obtain more reliable abnormal position coordinates, and then more reliable observable position points are inferred;

[0119] The observable path solving algorithm is used to calculate the angle of the observable position point on the two-dimensional XY horizontal plane, and then the difference in the angle values ​​of adjacent points is compared with the set maximum angle difference Δθ of adjacent observable points on the same path, and the points are divided into different observable paths, that is, the observable path of each abnormal component is obtained;

[0120] Through the A* algorithm and the shortest route allocation algorithm, the optimal detection plan of the system is calculated, and each robot should go to its corresponding observable path for collaborative detection. The plan also satisfies the shortest overall running distance of multiple robots, further improving the overall operation efficiency of the system.

[0121] Finally, multiple robots go to their respective observable paths for collaborative detection, collect multi-view and multi-modal data, and use Kalman filtering to fuse these multi-view and multi-modal abnormal image data in the cloud, thereby improving the accuracy of the system in detecting equipment abnormalities.

[0122] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A collaborative detection method for abnormal parts of large equipment based on multiple robots, characterized in that: The method sets the number of robots to n (n≥2) according to the size and complexity of the large equipment, and numbers the robots from 1 to n in sequence, and evenly distributes the robots around the large equipment for inspection; Each robot is equipped with three sensors: a visible light camera, a thermal infrared camera, and an audio-visual camera. During the inspection process, the robot scans the equipment to obtain visible light images. vis 、Thermal infrared image I tir Harmony Image I ac ; The system is based on a cloud-edge collaborative architecture, with the robot as the edge end collecting I vis ,I tir and I ac After the three types of image data are collected, the image data is uploaded to the cloud with corresponding image detection and analysis functions for processing in real time. The cloud can perform anomaly detection based on image processing on the received image data to determine whether there is an abnormality in the large equipment; Among them, the collaborative detection of abnormal parts by multiple robots includes two stages. The first stage is the inspection stage, and the second stage is the collaborative detection stage.

2. The method for collaboratively detecting abnormal parts of large equipment based on multiple robots according to claim 1 is characterized in that: The inspection phase includes: Patrol inspection: the robot patrols around the equipment in the specified patrol direction and establishes an XYZ three-dimensional coordinate system based on the two-dimensional XY horizontal plane where the large equipment base is located; Among them, the counterclockwise inspection direction is set as the positive direction, and the clockwise inspection direction is set as the negative direction; Abnormal data collection: When the cloud receives the image data collected by a single sensor of the i-th robot at time t, it finds that there is an abnormality in the component through the image detection and analysis function, and then sends the i-th robot uploaded at the timestamp of time t to the cloud. vis ,I tir and I ac The three types of image data are fused through Kalman filtering to obtain a synthetic image, and then the image detection and analysis function is used to determine whether the component is abnormal. After determining that the component is abnormal, the cloud solves and marks the observed abnormal position coordinates P in the global XYZ three-dimensional coordinate system. it (x it ,y it ,z it ), solve and mark the position coordinates R of the i-th robot in the two-dimensional XY plane coordinate system it (x i ' t ,y i ' t ), and record the one-to-one correspondence, and record the mapping relationship as f:R→P; and when the image data uploaded by the i-th robot at time t is detected to be normal in the cloud, no position coordinates are solved and marked; Observable point calculation, in the cloud through the clustering algorithm to select the reliable abnormal point location, the observed abnormal location coordinates P it Clustering is performed and P is transformed into it Core point P Δ , boundary points P′ and noise points P″, and several clusters are obtained and each cluster Center of gravity The core point P Δ P in the cluster it And the number of abnormal position coordinates within the spherical neighborhood ε of the point with eps as radius is greater than Pts Cor , given the radius eps and the minimum number of points Pts in the neighborhood of the core point Cor are all initial setting values; the boundary point P′ represents the P in the cluster excluding the core point it ; Noise point P″ indicates P that is not in the cluster it Different clusters Used to indicate the different abnormal components detected in this round of inspection, each center of gravity Considered as corresponding cluster The abnormal point of the component is re-labeled as P anom ; At the same time, the cloud will be different abnormal points P anom Randomly put them into the task list one by one, and in the subsequent collaborative detection stage, outliers are taken from the front to the back in the task list for detection; if there is an outlier point P anom Cluster The coordinates of the outlier points P it The number is less than the minimum number of points Pts required to meet collaborative detection Clus , then the outlier point P anom Don’t add to your to-do list; The cloud then The coordinates of the abnormal position P it With the inverse mapping f -1 :P→R, derive the coordinate system of the two-dimensional XY plane with P it The corresponding robot position coordinates R it , then the position coordinates of these robots are the outlier points P anom The observable position of these robots is re-marked as the observable position point Q k (x k ,y k ); Observable path solving: each abnormal point P is solved in the cloud through the observable path solving algorithm anom The observable paths l1,l2,…,l j ,…, the angle θ of a point in the two-dimensional plane is set as the angle between the point and the origin and the positive direction of the x-axis; the observable path solution algorithm first calculates each observable position point Q in the two-dimensional XY plane coordinate system k The corresponding angle θ (0≤θ<2π), and each observable position point Q k Sort by angle θ, with smaller angles in front and larger angles in the back. If the χ+1th Q k The angle θ χ+1 Subtract the xth Q k The angle θ χ If the maximum angle difference Δθ of adjacent observable points on the same path is less than or equal to the setting, then the χ+1th Q k and the xth Q k Considered to be on the same observable path l j Otherwise, the χ+1th Q k and the xth Q k In different observable paths, different observable paths are divided and fitted to obtain observable paths.

3. The method for collaboratively detecting abnormal parts of large equipment based on multiple robots according to claim 2 is characterized in that: In the abnormal data collection step, the time when the i-th robot first discovers the abnormality is recorded as time 0, and the position coordinate at time 0 is R i0 (x′ i0 ,y′ i0 ), with the origin as the endpoint and passing through point R i0 (x′ i0 ,y′ i0 ) is set as the reference position. When the i-th robot reaches the ray α times, that is, after a robot finds an abnormality for the first time, all robots continue to inspect the large equipment α times to complete the abnormal position coordinate P it (x it ,y it ,z it ) and the robot position coordinates R it (x′ it ,y′ it ) is marked and uploaded, then the abnormal data collection is completed; After completing the α-circle orbit, the robot position coordinates are initially obtained in the cloud. and the set of anomaly location coordinates And the mapping relationship f:R→P between set R and set P.

4. A method for collaboratively detecting abnormal parts of large equipment based on multiple robots according to any one of claims 1-2, characterized in that: The collaborative detection stage includes: Calculate the optimal detection solution. This step is performed in the cloud. First, the outlier point P needs to be added to the task list. anom Take out from the front to the back for testing, and record the current test object as P a ' nom , let the detection object P a ' nom There are m observable paths, and at the same time, there are m observable paths l j The observable points with the smallest and largest angles are set as endpoints, and the observable point with the smallest angle is recorded as The observable point with the largest angle is recorded as Through the A* algorithm, find the path l from the current position of the i-th robot j Endpoints and The shortest route distance is recorded as and The distance cost from the current position of the robot to the endpoint of each path is calculated by the A* algorithm, thereby obtaining the distances of each robot to the observable points with the smallest and largest angles on each path, and constructing the cost matrix C1 with the distances of each robot to the observable point with the smallest angle on each path, and constructing the cost matrix C2 with the distances of each robot to the observable point with the largest angle on each path; The final cost matrix C = g(C1, C2), where g represents C[p][q] = min(C1[p][q], C2[p][q]), that is, the p-th row and q-th column element of the final cost matrix C is equal to the minimum value of the p-th row and q-th column elements of C1 and C2; After obtaining the cost matrix, the optimal detection scheme for collaborative detection is solved through the shortest path allocation algorithm, so that the total path of all robots is the shortest; Execute the optimal detection plan, assign each robot to move along the route of the optimal detection plan, collect abnormal data of abnormal parts after reaching the observable path, and upload the collected data to the cloud in real time. The cloud uses Kalman filtering to fuse multi-modal and multi-perspective abnormal data to obtain the final detection result of the abnormal point; Complete the current detection target P a ' nom After completing the detection task, check whether there are any tasks to be detected in the task list P anom , if it exists, take out the new anomaly point at the front as the target of a new round of detection; The outlier point P in the waiting list anom After all the tests are completed, the collaborative testing phase ends.

5. The method for collaboratively detecting abnormal parts of large equipment based on multiple robots according to claim 4 is characterized in that: The shortest route allocation algorithm comprises the following steps: S1: Subtract the minimum value in each row of the cost matrix, and then subtract the minimum value in each column from the minimum value in the column, that is, ensure that there is at least one zero in each row and column; S2: Use as few row markers and column markers as possible to cover all zeros in the matrix. One row marker will cover the entire row elements of the matrix, and one column marker will cover the entire column elements of the matrix. If the sum of the number of row and column markers is less than min(n,m), proceed to step S3; if the number of row and column markers is equal to min(n,m), find the current optimal solution and execute step S4; S3: Find the minimum value ξ among the elements not covered by the row and column marks, subtract ξ from all the elements not covered by the row and column marks, add ξ to the elements covered twice by the row and column marks, remove the row and column marks, and re-execute step S2; S4: After obtaining the current optimal solution, due to detecting different abnormal points P anom When the relationship between the number of robots n and the number of observable paths m is different, the system operation scheme is also different. Compare the sizes of n and m and choose different schemes to run; S5: The total length of the current detection route of each robot and the distance from the end point of the current detection route to the end point of the remaining path are added together as the distance cost for multiple robots to reach the remaining observable path. At the same time, the new cost matrix is ​​reconstructed in the same way as the above cost matrix construction method, and step S1 is re-executed to solve the optimal detection plan for collaborative detection.

6. The method for collaboratively detecting abnormal parts of large equipment based on multiple robots according to claim 5 is characterized in that: The step S4 compares the sizes of n and m and selects different solutions, which specifically include the following: When n>m, the current optimal solution is used as the optimal detection solution for collaborative detection, and m robots are dispatched to m paths to detect the current detection target P. a ' nom Conduct collaborative detection, and the remaining nm robots continue to patrol and collect abnormal data of large equipment; When n=m, the current optimal solution is used as the optimal detection solution for collaborative detection, and n robots are dispatched to n paths to detect the current detection target P. a ' nom Conduct collaborative testing; When n < m, the current optimal solution is used as a partial solution for the overall shortest path plan of the current collaborative detection, and n robots are dispatched to n paths respectively to detect the current target P a ′ nom for collaborative detection. Denote the current detection route of the robot as the route to the specified path plus the route for detection on the specified path. Assume that the i-th robot is dispatched to the observable path l j , and first passes through the observable point with the smallest angle and the detection end point is the observable point with the largest angle Calculate the total length s of the current detection route of the i-th robot, that is, from the current position to the detection end point using the A* algorithm i and the shortest route distances from the detection end point of the current route to the endpoints of the remaining paths and are denoted as and respectively, and step S5 is executed.

7. The method for collaboratively detecting abnormal parts of large equipment based on multiple robots according to claim 4 is characterized in that: The initial cost matrices C1 and C2 of the distances of each robot to the observable points with the smallest and largest angles on each path are: in, The path from the current position of the i-th robot to the path l j Endpoints The shortest route distance, The path from the current position of the i-th robot to the path l j Endpoints The shortest route distance.

8. The method for collaboratively detecting abnormal parts of large equipment based on multiple robots according to claim 3 is characterized in that: When the abnormal data is collected, when the robot detects multiple abnormal position coordinates P at time t it When the robot position and abnormal position coordinates P are marked respectively it ; If the i-th robot detects three abnormal position coordinates at time t, the abnormal position coordinates are marked as P respectively. it 1 (x it ,y it ,z it ), P it 2 (x it ,y it ,z it ) and P it 3 (x it ,y it ,z it ), and the robot position is R it 1 (x i ' t ,y i ' t ), R it 2 (x i ' t ,y i ' t ) and R it 3 (x i ' t ,y i ' t ) and record the one-to-one correspondence.

9. The method for collaboratively detecting abnormal parts of large equipment based on multiple robots according to claim 2 is characterized in that: In the observable point solution, the core point P whose distance is less than eps is Δ and the boundary point P′ are clustered in the same initial cluster According to the core point P Δ The weights of the boundary points P′ are used to calculate the initial cluster Center of gravity Core point P Δ The weight is set to τ times of the boundary point P′; get the initial cluster center Then, filter the clusters Center and center of gravity The core points P of the first v% with the closest Euclidean distance Δ and boundary point P′, the filtered core point P Δ and the boundary point P′ form a cluster representing abnormal components And calculate the clusters Center of gravity 10. The method for collaboratively detecting abnormal parts of large equipment based on multiple robots according to claim 5, characterized in that: The current optimal solution in step S2 is the adjusted cost matrix, which satisfies that the number of row and column markers is equal to min(n,m), and specifically includes the following: If n≤m, then each row of the cost matrix has at least one zero element, and at least one zero element is selected from each row to ensure that the selected zero elements are in different columns. The row and column where the selected zero element is located indicates that the robot corresponding to the row should go to the observable path endpoint corresponding to the column. If n>m, then each column of the cost matrix has at least one zero element, and at least one zero element is selected from each column to ensure that the selected zero elements are in different rows. At this time, the row and column where the selected zero element is located indicate that the robot corresponding to the row should go to the observable path endpoint corresponding to the column.

Images