Robot-based furnace dry ice intelligent decoking method and system

Abstract

The present invention relates to the field of combustion product removal, and in particular to a robot-based furnace dry ice intelligent decoking method and system. The method comprises the steps of: collecting a furnace image; segmenting the furnace image to obtain coke units; clustering the coke units to obtain multi-layer clustering results; using each category in each layer of clustering results as a node, constructing an undirected graph, and obtaining the shortest path; connecting the shortest paths obtained from all layers of clustering results to obtain a decoking path, so as to achieve decoking. The efficiency and accuracy of obtaining the optimal decoking path are improved.

Description

Technical Field

[0001] The present invention relates to the field of combustion product removal, and in particular to a robot-based furnace dry ice type intelligent decoking method and system. Background Art

[0002] The coke in the furnace will reduce the efficiency of the boiler and shorten its service life. In serious cases, it may even directly damage the equipment. Therefore, the furnace is decoked from time to time. Dry ice decoking technology is not only environmentally friendly and energy-saving, but also does not cause damage to the equipment. Therefore, dry ice decoking technology is often used to decoke the furnace.

[0003] When the furnace is decoking, in order to improve the decoking efficiency, a reasonable moving path will be set for the robot to decoke according to the distribution of coke in the furnace. When setting a reasonable moving path for the robot, machine vision is often used to segment the coke units, and then each coke unit is used as a position unit to obtain the shortest path. This method of using coke units as position units produces more path solutions. If they are screened one by one, the process of obtaining the shortest path takes a long time. If they are not screened one by one, it is easy to find only a local optimal path because the coke distribution cannot be grasped macroscopically. Therefore, how to quickly and accurately find the optimal decoking path becomes the research focus of the present invention.

[0004] The patent application document with publication number CN108518695A discloses an online decoking device and decoking method for a thermal storage body. The patent application document mainly discloses the process flow of removing coke, and does not involve the content of decoking path design. Therefore, the method in the patent application document cannot solve the technical problem in the present invention. Summary of the invention

[0005] In order to solve the problem of how to quickly and accurately find the optimal decoking path, the present invention provides a robot-based furnace dry ice type intelligent decoking method and system.

[0006] In a first aspect, the present invention provides a robot-based furnace dry ice intelligent decoking method, which adopts the following technical solutions:

[0007] A robot-based furnace dry ice intelligent decoking method comprises the following steps:

[0008] Collect furnace images;

[0009] Segment the furnace image to obtain coke units; cluster the coke units to obtain a first-level clustering result, and in response to the distance difference between each category in the clustering result being no greater than a preset difference threshold, cluster the coke units in each category to obtain a second-level clustering result, and in response to the distance difference between each category in the clustering result being greater than a preset difference threshold, clustering ends;

[0010] Obtain each category in the first-layer clustering result, take each category in the first-layer clustering result as a node, connect every two adjacent nodes, obtain the distance between every two adjacent nodes as the edge weight of the connecting edge, construct an undirected graph, obtain the shortest path, and record it as the shortest path of the first-layer clustering result; obtain the connected nodes of each node on the shortest path of the first-layer clustering result, obtain the category closest to the connected node in the clustering result of a node and record it as the lower endpoint of the node, and use the lower endpoint as the lower connection point of the node; take each category in the clustering result of the node as a child node, connect every two adjacent child nodes, take the distance between every two adjacent child nodes as the edge weight, construct an undirected graph of the node, obtain the shortest path with the lower connection point of the node as the starting point and end point of the path, and record it as the shortest path of the second-layer clustering result;

[0011] This process is repeated in this way until the analysis of the last layer of clustering results is completed. The shortest paths of all layers of clustering results are connected together based on the lower connected points, and the obtained path is used as the decoking path to achieve decoking processing.

[0012] The present invention uses the clustering results of hierarchical clustering processing as the node construction path. This path construction method can greatly reduce the number of generated path solutions. At the same time, this path construction method takes into account both the macro distribution of coke and the detailed distribution. Therefore, this path makes the constructed path more accurate.

[0013] Preferably, the segmentation process of the furnace image to obtain the coke unit includes:

[0014] Get a pre-trained coke segmentation network;

[0015] The furnace image is input into the pre-trained coke segmentation network to segment the coke pixels, and the connected domain composed of the coke pixels is obtained as the coke area. The coke area is evenly divided into several blocks, and each block is recorded as a coke unit.

[0016] The present invention segments coke pixels through a coke segmentation network, and this segmentation method has higher accuracy.

[0017] Preferably, clustering the coke units to obtain a first layer of clustering results includes:

[0018] Several categories are preset, and based on each category, the coke units are clustered, the category interval is calculated, and the clustering result of the category with the largest category interval is used as the first-level clustering result.

[0019] The present invention evaluates the classification result according to the category interval, thereby achieving accurate classification.

[0020] Preferably, the method for obtaining the category interval includes:

[0021] Get the adjacent categories of each category, record the distance between the nearest pixels of each two adjacent categories as the inter-class distance, calculate the distance between each adjacent pixel in the category, take the mean of the distances between all two adjacent pixels in the category as the intra-class distance, divide the mean of the inter-class distances of all two adjacent categories by the mean of the intra-class distances of all categories, and then normalize them to get the category interval.

[0022] Preferably, the method for obtaining the distance between every two adjacent nodes includes:

[0023] Get each category in the last layer of clustering results and record them as unit categories;

[0024] Each pixel in the unit category is taken as the analysis unit, and the distance between the analysis units is taken as the comprehensive distance. Based on the comprehensive distance between the analysis units, the comprehensive distance between the categories of the last layer of clustering results is obtained; the clustering results of the previous layer are obtained, and the unit category is taken as the analysis unit. Based on the comprehensive distance between the analysis units, the comprehensive distance between the categories of the previous layer of clustering results is obtained. By analogy, the clustering results of the nodes are obtained, and each category in the clustering results of the nodes is taken as the analysis unit. Based on the comprehensive distance between the analysis units, the comprehensive distance between adjacent nodes is obtained, which is recorded as the distance between adjacent nodes.

[0025] When calculating the distance between nodes, the present invention takes into account the path design problem of the coke unit inside the node, so that the designed path can meet the macro path requirements while also taking into account the detailed path requirements.

[0026] Preferably, obtaining the comprehensive distance between categories of the last layer of clustering results based on the comprehensive distance between the analysis units includes:

[0027] Obtain each category in the last layer of clustering results as a unit category;

[0028] The outermost analysis unit in the unit category is taken as an optional point, and any two optional points in the unit category form a point pair; the analysis unit is taken as a node, and the two analysis units in the point pair are taken as the starting point and the end point of the path, and the shortest path formed by all the nodes in the unit category is obtained, which is recorded as the analysis path of the point pair;

[0029] The point pairs of two adjacent unit categories are combined into a point pair combination, the distance between each pixel in one point pair in the point pair combination and each pixel in another point pair in the point pair combination is obtained, and the minimum distance is taken as the minimum distance of the point pair combination; the average of the analysis path lengths of the two point pairs in the point pair combination is taken as the comprehensive path length of the point pair combination; with the comprehensive path length as the weight, the minimum distances of all point pair combinations are weighted and summed to obtain the comprehensive distance of the two adjacent unit categories.

[0030] The comprehensive distance calculated by the present invention includes the starting point and the ending point information of the internal path of the node, so that the path designed based on the comprehensive distance will take into account the detailed path information.

[0031] Preferably, the method for obtaining the decoking path comprises:

[0032] The lower-level connection points of each category in the last layer of clustering results are used as the starting and ending points of the path, and the shortest paths of each category of the last layer of clustering results are obtained, which are recorded as unit paths; the last layer is recorded as the first upper layer, and the upper layer of the last layer is recorded as the second upper layer. The shortest path of the clustering result of the second upper layer is used as the reference path, and the connection points on the reference path are obtained. The unit paths corresponding to the connection points are connected together using the lower-level connection points to obtain the comprehensive path of the second upper layer; the upper layer of the second upper layer is used as the third upper layer, and the shortest path of the clustering result of the third upper layer is obtained as the reference path. The connection points on the reference path are obtained, and the comprehensive paths corresponding to the connection points are connected together using the lower-level connection points to obtain the comprehensive path of the third upper layer; and so on, until the connection processing of the first layer of clustering results is completed, and the obtained path is used as the defocused path.

[0033] Preferably, the decoking process comprises:

[0034] The dry ice decoking robot moves along the decoking path and uses dry ice to perform decoking.

[0035] In a second aspect, the present invention provides a robot-based furnace dry ice intelligent decoking system, which adopts the following technical solutions:

[0036] A robot-based furnace dry ice type intelligent decoking system comprises: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the robot-based furnace dry ice type intelligent decoking method is implemented.

[0037] By adopting the above technical solution, the above robot-based furnace dry ice intelligent decoking method is generated into a computer program and stored in a memory to be loaded and executed by a processor, so that a terminal device is manufactured based on the memory and the processor for easy use.

[0038] The present invention has the following technical effects:

[0039] The present invention uses the clustering results of hierarchical clustering processing as the node construction path. This path construction method can greatly reduce the number of generated path solutions. At the same time, this path construction method takes into account both the macro distribution of coke and the detailed distribution. Therefore, this path makes the constructed path more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] By reading the detailed description below with reference to the accompanying drawings, the above and other purposes, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-restrictive manner, and the same or corresponding reference numerals represent the same or corresponding parts.

[0041] Figure 1 It is a flow chart of a method in a robot-based furnace dry ice intelligent decoking method according to an embodiment of the present invention;

[0042] Figure 2 A schematic diagram of the decoking path connection provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0044] It should be understood that when the terms "first", "second", etc. are used in the claims, descriptions, and drawings of the present invention, they are only used to distinguish different objects, rather than to describe a specific order. The terms "include" and "comprise" used in the description and claims of the present invention indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their collections.

[0045] The embodiment of the present invention discloses a robot-based furnace dry ice intelligent decoking method, referring to Figure 1 , comprising steps S1 to S4:

[0046] S1: Collect furnace images.

[0047] Specifically, a camera is used to collect images inside the furnace, which are recorded as furnace images.

[0048] S2: Segment the furnace image to obtain coke units; cluster the coke units to obtain a first-level clustering result. In response to the distance difference between each category in the clustering result being no greater than a preset difference threshold, cluster the coke units in each category to obtain a second-level clustering result. In response to the distance difference between each category in the clustering result being greater than a preset difference threshold, clustering ends.

[0049] S20: Segment the furnace image to obtain coke units.

[0050] It should be noted that in order to perform decoking, the location of the coke in the furnace must be obtained first, so the furnace image must be segmented to obtain the coke area. However, dry ice decoking does not remove coke at one pixel position at a time, and there will be a certain radiation width, so the coke area needs to be divided into blocks according to the radiation width.

[0051] Preferably, as an example, segmenting the furnace image to obtain the coke unit includes:

[0052] Get a pre-trained char segmentation network.

[0053] This embodiment uses Unet as the coke segmentation network, and other embodiments may use other networks, which are not specifically limited in this embodiment. The labeled historically collected furnace images are used to form a data set, and the coke segmentation network is trained using the data set to obtain a pre-trained coke segmentation network. The labeled historically collected furnace images are processed by manually labeling the historically collected furnace images.

[0054] The furnace image is input into the pre-trained coke segmentation network to segment the coke pixels, and the connected domain formed by the coke pixels is obtained as the coke area. The coke area is evenly divided into several blocks of preset sizes, and each block is recorded as a coke unit. In this embodiment, the preset size is determined according to the radiation range of dry ice decoking. For example, dry ice decoking can remove The coke within the pixel range, the preset size is .

[0055] In particular, when segmenting the coke area, pixels at the edge of the coke area may not be segmented into blocks that meet the size requirements. In this case, it is only necessary to add the surrounding pixels into the blocks.

[0056] S21: Clustering the coke units to obtain a first layer of clustering results. In response to the distance difference between each category in the clustering results being no greater than a preset difference threshold, clustering is performed on the coke units in each category to obtain a second layer of clustering results. In response to the distance difference between each category in the clustering results being greater than a preset difference threshold, clustering ends.

[0057] Optionally, as an example, clustering the coke units to obtain a first layer of clustering results, in response to the distance difference between each category in the clustering results being greater than a preset difference threshold, clustering the coke units in each category to obtain a second layer of clustering results, in response to the distance difference between each category in the clustering results being not greater than the preset difference threshold, clustering ends, including:

[0058] The K-means clustering algorithm is used to cluster the coke units into a preset number of categories, and each category is used as the first-level clustering result. This embodiment is described by taking the preset number of 5 as an example, and other embodiments may take other values, which are not specifically limited in this embodiment.

[0059] In response to the distance difference between each category in the clustering result being no greater than a preset difference threshold, clustering is performed on the coke units in each category according to the method for obtaining the first layer clustering result to obtain a second layer clustering result. In response to the distance difference between each category in the clustering result being greater than a preset difference threshold, clustering ends.

[0060] The distance difference calculation method includes:

[0061] The distance between each coke unit and the adjacent coke unit in each category is obtained, the mean of the distance between each coke unit and the adjacent coke unit is used as the adjacent distance of each coke unit, and the normalized value of the variance of the adjacent distance of all coke units is used as the distance difference.

[0062] It should be added that the method for obtaining adjacent coke units includes: connecting the geometric center of a coke unit with the geometric center of another coke unit, and if the connecting line does not intersect with the other coke unit, taking the other coke unit as the reference coke unit of the coke unit, and taking the reference coke unit within the preset radius of the coke unit as the adjacent coke unit. This embodiment is described by taking the preset radius as 100 as an example, and other embodiments may take other values, which are not specifically limited in this embodiment.

[0063] It should be noted that directly clustering the coke units into a preset number of categories may easily lead to poor clustering effect due to unreasonable setting of the preset number.

[0064] Preferably, as an example, clustering the coke units to obtain a first layer of clustering results, in response to the distance difference of the clustering results being no greater than a preset difference threshold, clustering the coke units in each category to obtain a second layer of clustering results, in response to the distance difference of the clustering results being greater than the preset difference threshold, clustering ends, including:

[0065] Several class numbers are preset, and based on each class number, the coke units are clustered using the K-means distance algorithm, the class interval is calculated, and the clustering result of the class number with the largest class interval is used as the first-level clustering result. This embodiment is described by taking the preset class number as each integer between 3 and 10 as an example, and other embodiments may take other values, which are not specifically limited in this embodiment.

[0066] The calculation method of the category interval includes:

[0067] The adjacent categories of each category are obtained, and the distance between each category and the nearest coke unit of the adjacent category is taken as the inter-class distance. The distance between every two adjacent coke units in the category is calculated, and the mean of the distances between all two adjacent coke units in the category is taken as the intra-class distance. The mean of the inter-class distances of all two adjacent categories is divided by the mean of the intra-class distances of all categories and then normalized to obtain the class interval.

[0068] It should be added that the methods for obtaining adjacent categories include:

[0069] Connect the geometric center of one category with the geometric center of another category. If the connection line does not pass through the other category, the other category is used as the reference category of the category, and the reference category within the preset radius of the category is used as the adjacent category. This embodiment is described by taking the preset radius as 100 as an example, and other embodiments may take other values, which are not specifically limited in this embodiment.

[0070] In response to the distance difference of the clustering result being no greater than the preset difference threshold, clustering is performed on the coke units in each category according to the method for obtaining the first layer of clustering results to obtain the second layer of clustering results. In response to the distance difference of the clustering result being greater than the preset difference threshold, clustering ends. This embodiment is described by taking the difference threshold of 0.6 as an example, and other embodiments may take other values, which are not specifically limited in this embodiment.

[0071] It should be added that the distance difference calculation method is the same as the distance difference calculation method in the optional method.

[0072] S3: Obtain each category in the first-layer clustering result, use each category in the first-layer clustering result as a node, connect every two adjacent nodes, obtain the distance between every two adjacent nodes as the edge weight of the connecting edge, construct an undirected graph, obtain the shortest path, and record it as the shortest path of the first-layer clustering result; obtain the connected nodes of each node on the shortest path of the first-layer clustering result, obtain the category closest to the connected node in the clustering result of a node and record it as the lower endpoint of the node, and use the lower endpoint as the lower connection point between nodes on the shortest path; use each category in the clustering result of the node as a child node, connect every two adjacent child nodes, use the distance between every two adjacent child nodes as the edge weight, construct an undirected graph of the node, and obtain the shortest path with the endpoints of the node as the starting point and end point of the path, which is recorded as the shortest path of the second-layer clustering result.

[0073] It should be noted that since there are many coke units, taking each coke unit as a path node will generate a large number of path solutions, and it takes a long time to select the optimal path from a large number of paths. For example, if there are 8 coke units distributed around each coke unit, there will be 8 possible paths for each coke unit. If there are 1,000 coke units in the graph, There are a lot of path plans, which is a large amount of data. If we use the method of traversing one by one to screen the optimal path, it will take a lot of time. However, if we do not use the method of traversing one by one to screen the optimal path, it is easy to screen out a local optimal solution.

[0074] It should be further explained that the internal space of the furnace is large, and the distribution space of the coke area is large. If each coke unit is used as a path node, too much attention will be paid to the distribution details of each coke area, making it difficult to grasp the distribution position at a macro level, and thus it is difficult to quickly design the optimal path.

[0075] S30: Obtain each category in the first layer clustering result, use each category in the first layer clustering result as a node, connect every two adjacent nodes, obtain the distance between every two adjacent nodes as the edge weight of the connecting edge, construct an undirected graph, obtain the shortest path, and record it as the shortest path of the first layer clustering result.

[0076] It should be noted that in order to prevent excessive attention to distribution details when designing the path, it is necessary to analyze and design the path from different macro levels, which can effectively improve the efficiency of path design. For example, 1000 coke units are divided into 50 areas, each area is surrounded by 8 areas, each area is used as a node, and a path is set. The number of path solutions set by the area is ,exist The shortest path is obtained from the path schemes, and then the path design is performed for each area. The path scheme generated in each area is analyzed. Assuming that there are 20 coke units in each area, the path scheme generated in each area is ,exist Get a shortest path from the path solutions, and then connect the shortest paths in the area to get the defocused path. path solutions, so the number of path solutions is much smaller than , so this method is more efficient in obtaining the shortest path.

[0077] Preferably, as an example, each category in the first layer clustering result is obtained, each category in the first layer clustering result is used as a node, each two adjacent nodes are connected, the distance between each two adjacent nodes is obtained as the edge weight of the connecting edge, an undirected graph is constructed, and the shortest path is obtained, which is recorded as the shortest path of the first layer clustering result, including:

[0078] Get each category in the first layer clustering results, use each category in the first layer clustering results as a node, connect every two adjacent nodes, get the distance between every two adjacent nodes as the edge weight of the connecting edge, build an undirected graph, and use the Dijkstra algorithm to get the shortest path, which is recorded as the shortest path of the first layer clustering results.

[0079] The above embodiments involve the distance between every two adjacent nodes. The method for obtaining the distance between every two adjacent nodes will be described below.

[0080] Optionally, as an example, a method for obtaining the distance between every two adjacent nodes includes:

[0081] The Euclidean distance between the geometric center of each node and the geometric center of the adjacent node is taken as the distance between each node and the adjacent node.

[0082] It should be noted that the Euclidean distance between the geometric centers of the nodes is used as the distance between every two adjacent nodes. This method treats the nodes as point masses, but in fact, there are many coke units inside the nodes, and the distances between nodes calculated based on different coke units are different. In particular, when the number of coke units contained in the nodes is large, the calculated distances are quite different. There are also many coke units inside the nodes, so it is necessary to design paths for the coke units inside the nodes. The endpoints of the paths designed for the coke units inside the nodes will affect the calculation of the distance relationship between the nodes. Therefore, when calculating the distance between nodes, the path design of the coke units inside the nodes needs to be considered.

[0083] Preferably, as an example, the method for obtaining the distance between every two adjacent nodes includes:

[0084] Get each category in the last layer of clustering results and record them as unit categories;

[0085] Each pixel in the unit category is taken as an analysis unit, and the distance between the analysis units is taken as the comprehensive distance; based on the comprehensive distance between the analysis units, the comprehensive distance between the categories of the last layer of clustering results is obtained; the clustering results of the previous layer are obtained, the unit category is taken as the analysis unit, and the comprehensive distance between the categories of the previous layer of clustering results is obtained based on the comprehensive distance between the analysis units. By analogy, the clustering results of the nodes are obtained, each category in the clustering results of the nodes is taken as an analysis unit, and the comprehensive distance between adjacent nodes is obtained based on the comprehensive distance between the analysis units, which is recorded as the distance between adjacent nodes.

[0086] The comprehensive distance between the categories of the last layer of clustering results is obtained based on the comprehensive distance between the analysis units, including:

[0087] Obtain each category in the last layer of clustering results as a unit category;

[0088] The outermost circle of analysis units in the unit category is taken as an optional point, and any two optional points in the unit category form a point pair; the analysis unit is taken as a node, and the two analysis units in the point pair are taken as the starting point and the end point of the path, and the Dijkstra algorithm is used to obtain the shortest path formed by all nodes in the unit category, which is recorded as the analysis path of the point pair;

[0089] The point pairs of two adjacent unit categories are combined into a point pair combination, the distance between each pixel in one point pair in the point pair combination and each pixel in another point pair in the point pair combination is obtained, and the minimum distance is taken as the minimum distance of the point pair combination; the average of the analysis path lengths of the two point pairs in the point pair combination is taken as the comprehensive path length of the point pair combination; with the comprehensive path length as the weight, the minimum distances of all point pair combinations are weighted and summed to obtain the comprehensive distance of the two adjacent unit categories.

[0090] It can be understood that the comprehensive distance calculated in this way takes into account the internal path design of the node, so that the macro path designed based on the comprehensive distance takes into account both the detailed distribution and the macro distribution, and thus the path designed based on the comprehensive distance is better.

[0091] S31: Obtain the connected nodes of each node on the shortest path of the first-level clustering result, obtain the category closest to the connected node in the clustering result of a node and record it as the lower endpoint of the node, and use the lower endpoint as the lower connected point of the node; use each category in the clustering result of the node as a child node, connect every two adjacent child nodes, use the distance between every two adjacent child nodes as the edge weight, construct an undirected graph of the node, obtain the shortest path with the lower connection point of the node as the starting point and end point of the path, and record it as the shortest path of the second-level clustering result.

[0092] Preferably, as an example, the connected nodes of each node on the shortest path of the first-layer clustering result are obtained, and the category closest to the connected node in the clustering result of a node is obtained and recorded as the lower connected point of the node; each category in the clustering result of the node is used as a child node, and each two adjacent child nodes are connected, and the distance between each two adjacent child nodes is used as the edge weight, and an undirected graph of the node is constructed, and the shortest path with the lower connection point of the node as the starting point and the end point of the path is obtained, which is recorded as the shortest path of the second-layer clustering result, including:

[0093] Get the connected nodes of each node on the shortest path of the first-layer clustering result; get each category in the clustering result of each node, calculate the distance between the geometric center of each category of a node and the geometric center of each category of the connected node, and get the two categories corresponding to the minimum distance, and use the category belonging to the node in the two categories as the lower connected point of the node;

[0094] Take each category in the clustering result of the node as a child node, connect every two adjacent child nodes, take the distance between every two adjacent child nodes as the edge weight, construct an undirected graph of the node, and use the Dijkstra algorithm to obtain the shortest path with the lower connection point of the node as the starting point and end point of the path, which is recorded as the shortest path of the second-level clustering result.

[0095] S4: This process is repeated in this way until the analysis of the last layer of clustering results is completed, and the shortest paths of all layers of clustering results are connected together based on the lower connected points, and the obtained path is used as the decoking path to achieve decoking processing.

[0096] Preferably, as an example, and by analogy, until the analysis of the last layer of clustering results is completed, the shortest paths of all layers of clustering results are connected together based on the lower connection points, and the obtained path is used as the decoking path to achieve the decoking process, including:

[0097] And so on, until the analysis of the last layer of clustering results is completed;

[0098] The lower connected points of each category in the last layer of clustering results are used as the starting and ending points of the path, and the shortest path of each category in the last layer of clustering results is obtained, which is recorded as the unit path;

[0099] The last layer is recorded as the first upper layer, the layer above the last layer is recorded as the second upper layer, the shortest path of the clustering result of the second upper layer is used as the reference path, the connection points on the reference path are obtained, and the unit paths corresponding to the connection points are connected together using the lower connection points to obtain the comprehensive path of the second upper layer;

[0100] The upper layer of the second upper layer is used as the third upper layer, the shortest path of the clustering result of the third upper layer is obtained as the reference path, the connection points on the reference path are obtained, and the comprehensive paths corresponding to the connection points are connected together using the lower connection points to obtain the comprehensive path of the third upper layer;

[0101] This process is repeated until the connection processing of the first-layer clustering results is completed, and the resulting path is used as the defocusing path. Figure 2 The decoking path connection diagram is shown below: Figure 2 The defocusing path formed by connecting the clustering results of different layers is shown in Figure .

[0102] The dry ice decoking robot moves along the decoking path and uses dry ice to perform decoking.

[0103] An embodiment of the present invention also discloses a robot-based furnace dry ice intelligent decoking system, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the robot-based furnace dry ice intelligent decoking method according to the present invention is implemented.

[0104] The above system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface, and their configuration and functions are known in the art, so they will not be described in detail here.

[0105] In the present invention, the aforementioned memory may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium may be any suitable magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory, a dynamic random access memory, a static random access memory, an enhanced dynamic random access memory, a high bandwidth memory, a hybrid storage cube, etc., or any other medium that can be used to store the required information and can be accessed by an application, a module, or both. Any such computer storage medium may be part of the device or accessible or connectable to the device.

[0106] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

[0107] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A robot-based furnace dry ice intelligent decoking method, characterized in that: Includes steps: Collect furnace images; Segment the furnace image to obtain coke units; cluster the coke units to obtain a first-level clustering result, and in response to the distance difference between each category in the clustering result being no greater than a preset difference threshold, cluster the coke units in each category to obtain a second-level clustering result, and in response to the distance difference between each category in the clustering result being greater than a preset difference threshold, clustering ends; Obtain each category in the first-layer clustering result, take each category in the first-layer clustering result as a node, connect every two adjacent nodes, obtain the distance between every two adjacent nodes as the edge weight of the connecting edge, construct an undirected graph, obtain the shortest path, and record it as the shortest path of the first-layer clustering result; obtain the connected nodes of each node on the shortest path of the first-layer clustering result, obtain the category closest to the connected node in the clustering result of a node and record it as the lower endpoint of the node, and use the lower endpoint as the lower connection point of the node; take each category in the clustering result of the node as a child node, connect every two adjacent child nodes, take the distance between every two adjacent child nodes as the edge weight, construct an undirected graph of the node, obtain the shortest path with the lower connection point of the node as the starting point and end point of the path, and record it as the shortest path of the second-layer clustering result; This process is repeated in this way until the analysis of the last layer of clustering results is completed. The shortest paths of all layers of clustering results are connected together based on the lower connected points, and the obtained path is used as the decoking path to achieve decoking processing.

2. The robot-based furnace dry ice intelligent decoking method according to claim 1, characterized in that: The step of segmenting the furnace image to obtain the coke unit includes: Get a pre-trained coke segmentation network; The furnace image is input into the pre-trained coke segmentation network to segment the coke pixels, and the connected domain composed of the coke pixels is obtained as the coke area. The coke area is evenly divided into several blocks, and each block is recorded as a coke unit.

3. The robot-based furnace dry ice intelligent decoking method according to claim 1, characterized in that: The clustering process for the coke units to obtain the first layer of clustering results includes: Several categories are preset, and based on each category, the coke units are clustered, the category interval is calculated, and the clustering result of the category with the largest category interval is used as the first-level clustering result.

4. The robot-based furnace dry ice intelligent decoking method according to claim 3, characterized in that: The method for obtaining the category interval includes: Get the adjacent categories of each category, record the distance between the nearest pixels of each two adjacent categories as the inter-class distance, calculate the distance between each adjacent pixel in the category, take the mean of the distances between all two adjacent pixels in the category as the intra-class distance, divide the mean of the inter-class distances of all two adjacent categories by the mean of the intra-class distances of all categories, and then normalize them to get the category interval.

5. The robot-based furnace dry ice intelligent decoking method according to claim 1, characterized in that: The method for obtaining the distance between every two adjacent nodes includes: Get each category in the last layer of clustering results and record them as unit categories; Each pixel in the unit category is taken as the analysis unit, and the distance between the analysis units is taken as the comprehensive distance. Based on the comprehensive distance between the analysis units, the comprehensive distance between the categories of the last layer of clustering results is obtained; the clustering results of the previous layer are obtained, and the unit category is taken as the analysis unit. Based on the comprehensive distance between the analysis units, the comprehensive distance between the categories of the previous layer of clustering results is obtained. By analogy, the clustering results of the nodes are obtained, and each category in the clustering results of the nodes is taken as the analysis unit. Based on the comprehensive distance between the analysis units, the comprehensive distance between adjacent nodes is obtained, which is recorded as the distance between adjacent nodes.

6. The robot-based furnace dry ice intelligent decoking method according to claim 5, characterized in that: The step of obtaining the comprehensive distance between categories of the last layer of clustering results based on the comprehensive distance between the analysis units includes: Obtain each category in the last layer of clustering results as a unit category; The outermost analysis unit in the unit category is taken as an optional point, and any two optional points in the unit category form a point pair; the analysis unit is taken as a node, and the two analysis units in the point pair are taken as the starting point and the end point of the path, and the shortest path formed by all the nodes in the unit category is obtained, which is recorded as the analysis path of the point pair; The point pairs of two adjacent unit categories are combined into a point pair combination, the distance between each pixel in one point pair in the point pair combination and each pixel in another point pair in the point pair combination is obtained, and the minimum distance is taken as the minimum distance of the point pair combination; the average of the analysis path lengths of the two point pairs in the point pair combination is taken as the comprehensive path length of the point pair combination; with the comprehensive path length as the weight, the minimum distances of all point pair combinations are weighted and summed to obtain the comprehensive distance of the two adjacent unit categories.

7. The robot-based furnace dry ice intelligent decoking method according to claim 1, characterized in that: The method for obtaining the decoking path comprises: The lower-level connection points of each category in the last layer of clustering results are used as the starting and ending points of the path, and the shortest paths of each category of the last layer of clustering results are obtained, which are recorded as unit paths; the last layer is recorded as the first upper layer, and the upper layer of the last layer is recorded as the second upper layer. The shortest path of the clustering result of the second upper layer is used as the reference path, and the connection points on the reference path are obtained. The unit paths corresponding to the connection points are connected together using the lower-level connection points to obtain the comprehensive path of the second upper layer; the upper layer of the second upper layer is used as the third upper layer, and the shortest path of the clustering result of the third upper layer is obtained as the reference path. The connection points on the reference path are obtained, and the comprehensive paths corresponding to the connection points are connected together using the lower-level connection points to obtain the comprehensive path of the third upper layer; and so on, until the connection processing of the first layer of clustering results is completed, and the obtained path is used as the defocused path.

8. The robot-based furnace dry ice intelligent decoking method according to claim 1, characterized in that: The decoking process comprises: The dry ice decoking robot moves along the decoking path and uses dry ice to perform decoking.

9. A robot-based furnace dry ice intelligent decoking system, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the robot-based furnace dry ice intelligent decoking method according to any one of claims 1 to 8 is implemented.

Images