Boiler air bellow ash removal system, method and device

Abstract

The invention relates to the technical field of air bellow deashing structures, in particular to a boiler air bellow deashing system, method and deashing device.The boiler air bellow deashing system comprises an image acquisition module used for acquiring images of areas, prone to ash deposition, of historical flues and images in the current use state and conducting image preprocessing to obtain an image set; the feature extraction module is used for extracting physical features and regional organization features from an image set, and the image model learning module is used for selecting and training machine models by utilizing the image set, the physical features and the regional organization features, and then integrating the trained machine models to obtain a deep learning model; due to the fact that the air cap structure has a large coverage range, a single air cap structure can cover a large air box side wall, the number and the positions of the air caps can be determined according to an RGB luminosity comparison method, historical data, the easily-accumulated portion and the thickness of dust easy to butt joint, and the functions of installing the fewest air caps and maximizing the coverage range are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of bellows dust cleaning structures, and particularly relates to a boiler bellows dust cleaning system, method and its dust cleaning device. Background Art

[0002] After the unit is put into production, during several recent major and minor repairs, it has been found that there is a large amount of ash accumulation in several layers of the boiler bellows. In particular, the ash accumulation in the burner bellows of the upper layers reaches more than 1.5 meters, and the ash accumulation area on a single layer and one side sometimes reaches more than 180 cubic meters. At present, there are three layers of bellows in a certain unit with the above situation. This not only increases the workload of manual ash cleaning, but also reduces the secondary air volume during operation, affecting the combustion air distribution of the boiler and having a certain impact on combustion. Therefore, it is planned to add a set of dust cleaning system in the third layer of the bellows with serious ash accumulation to eliminate the phenomenon of a large amount of ash accumulation at the bottom of the boiler bellows.

[0003] Among the three commonly used existing methods, the pneumatic ash blowing system scheme with compressed air is the most investment-friendly and economical, and has a good cleaning effect. However, in the existing compressed air equipment, there is a lack of a device for high-pressure blowing at the dead corners of the flue gas ducts where ash accumulates. Summary of the Invention

[0004] Technical Problem to be Solved

[0005] Aiming at the above-mentioned shortcomings of the existing technology, the present invention provides a boiler bellows dust cleaning system, method and its dust cleaning device, which can effectively solve the problem that in the existing compressed air equipment, there is a lack of a device for high-pressure blowing at the dead corners of the flue gas ducts where ash accumulates.

[0006] Technical Solution

[0007] To achieve the above objectives, the present invention is realized through the following technical solutions:

[0008] The present invention provides a boiler bellows dust cleaning system, including an image acquisition module: used to obtain images of the areas in the historical flue that are prone to ash accumulation and images in the current usage state, and perform image preprocessing to obtain an image set;

[0009] A feature extraction module: used to extract physical features and regional tissue features from the image set,

[0010] An image model learning module: used to select and train a machine model by using the image set, physical features and regional tissue features, and then integrate the trained machine model to obtain a deep learning model;

[0011] An image analysis module: used to analyze a newly input clean-state flue gas duct picture by using the deep learning model to obtain an evaluation result of the flue gas duct picture;

[0012] Visualization module: used to generate a visualization report based on the results of analysis and prediction.

[0013] Furthermore, the physical characteristics include the sizes of the bellow flue pipes, the range and thickness of the accumulated ash; the tissue characteristics include pores and granularity.

[0014] Furthermore, the evaluation results of the flue duct pictures include: one is the proportion of dust in the picture size within the picture range, and the other is the brightness and darkness of the dust. The soot cleaning range and the soot cleaning time are determined according to the proportion number and the brightness and darkness of the dust.

[0015] A method for cleaning soot in a boiler wind box, which analyzes the position comparison between the easily soot-accumulating positions of the flue duct in the unused state and the historical dust-covered positions to determine whether it is necessary to modify and optimize the soot cleaning points; if not, determine the minimum installation quantity of the wind caps and their maximum coverage area, and if necessary, adjust the quantity of the wind caps according to the historical dust coverage area to achieve a complete coverage effect, and add wind caps to the overlapping coverage part of the historical dust in the unused flue duct to complete the installation of the soot cleaning structure;

[0016] Analyze the results according to the overlapping coverage part of the historical dust, mark this position, determine the positions where dust is easily accumulated in the flue pipes, analyze the brightness and darkness of the pictures of the overlapping coverage part of the historical dust to obtain the overlapping coverage brightness and darkness value, and judge the dust accumulation thickness level of the flue pipes according to the dust coverage brightness and darkness value, which are the first class, the second class, the third class and the fourth class, and the thickness levels gradually become thinner according to the class order, and draw the dust thickness level contour lines on the pictures.

[0017] Meanwhile, the process of judging whether to modify and optimize the soot cleaning points is as follows: Take any point in the picture of the historical dust-covered position and perform color difference analysis of the brightness and darkness of the same point in the flue duct in the unused state. If the color difference analysis coincides, it is an easily covered position and is marked as an easily covered point. If it does not coincide, it is marked as a not easily covered point, and perform RGB brightness and darkness model analysis on the picture:

[0018]

[0019] The value range of L is 0 to 1 to obtain the brightness and darkness coincidence value L, where R, G, and B are the RGB values of the picture detection points. Compare the detected brightness and darkness coincidence value L with the detection coincidence threshold. If it is greater than the coincidence threshold, it is necessary to set the positions of the wind caps. If it is not greater than the coincidence threshold, it is not necessary to set the positions of the wind caps.

[0020] Meanwhile, obtain the internal image of the air duct at the historical dust-covered position, evenly divide it into an image array with equal spacing, based on any one image array, perform RGB ratio processing on the flue at the same position in the unused state. In one image array, further evenly divide it into standard squares with a quantity of 10 to obtain 10 groups of RGB values for this image array, and sum and average according to these 10 groups of RGB values to obtain t. Obtain several groups of image array averages of the historical dust-covered image, marked as t1, t2, t3....tn, to form a historical dataset, train a convolutional neural network on the historical dataset, and determine a value according to the output result, and compare it with the threshold of the hood blowing range. If it is greater than the hood blowing range threshold, it is necessary to open the hood fixing points according to the minimum range of the hood blowing. If it is not greater than the hood blowing range threshold, the hood fixing points can be opened according to the maximum range of the hood.

[0021] Meanwhile, the convolutional neural network includes two sets of convolutional neural modules. Each set of convolutional neural modules includes an input layer, multiple convolutional layers, a pooling layer, and a fully connected layer connected in sequence. First, input the historical dataset into the two input layers respectively. Each convolutional neural network includes a first convolutional layer, a second convolutional layer, a max pooling layer, a third convolutional layer, a fourth convolutional layer, a fifth convolutional layer, a first fully connected layer, a second fully connected layer, and an output layer. Input the image array average t into the first convolutional layer of the two convolutional modules respectively, and bring the feature data into the max pooling layer to enable the fully connected layer to output the category of the structural defect corresponding to the sample. The feature data includes the flue size and the flatness of the flue, where: Input layer: Receive the original image data or other types of grid structure data. Convolutional layer: Extract the local features of the input data through convolutional operations. Each convolutional kernel in the convolutional layer can extract a specific feature, and multiple convolutional kernels can work in parallel to extract different types of features. Pooling layer: Perform downsampling (or dimensionality reduction) on the output of the convolutional layer to reduce the number of parameters and improve the calculation efficiency. Common pooling operations include max pooling and average pooling. Fully connected layer: Combine the features extracted by the previous layers for tasks such as classification or regression. Each neuron in the fully connected layer is connected to all neurons in the previous layer.

[0022] Meanwhile, within the first convolutional layer of each group of convolutional neural networks, there are 81 groups of convolutional kernels with a size of 10*10, a stride of 2, and a zero padding at the edges; the window size of the first pooling layer is 3*3 with a stride of 2; within the second convolutional layer, there are 93 groups of convolutional kernels with a size of 10*10, a stride of 3, and a padding of 1 at the edges; within the third convolutional layer, there are 81 groups of convolutional kernels with a size of 12*12, a stride of 3, and a padding of 1 at the edges; within the fourth convolutional layer, there are 108 groups of convolutional kernels with a size of 11*11, a stride of 2, and a zero padding at the edges; within the fifth convolutional layer, there are 126 groups of convolutional kernels with a size of 11*11, a stride of 2, and a zero padding at the edges.

[0023] Meanwhile, a boiler air box soot cleaning device includes a control component structure, a control board, and an intelligent control component. Both ends of the control board are electrically connected to the intelligent control component and the control component structure. The intelligent control component is connected to multiple valves and a solenoid valve structure communicating with miscellaneous gas. One end of the valve is connected to the wind cap structure. The wind cap structure is a hollow cavity structure without a bottom surface at the bottom, and multiple groups of air outlets are provided on the side wall of the wind cap structure. The side wall of the air outlet is fixedly connected to the side cylinder, and the air outlet is in communication with the side cylinder.

[0024] Meanwhile, a connection port is provided at the bottom end of the air outlet on the wind cap structure, and the wind cap structure is cast from ZG40CI25NI20SI2Re material. The side cylinder forms a 45° angle with the side wall support of the wind cap structure.

[0025] Beneficial effects

[0026] The technical solution provided by the present invention has the following beneficial effects compared with the known public technologies:

[0027] Through the connection and cooperation of DCS / PLC, the present invention intelligently and rationally blows and cleans the flue. Using compressed air as the purging medium, the purging wind caps are accurately arranged in the ash accumulation area of the flue. Through the intelligent control compressed air purging system, the ash accumulation is purged and agitated at fixed times and locations, and the ash is carried away by the flue gas flow to achieve the function of preventing furnace ash from depositing.

[0028] In the present invention, through the provided soot cleaning system and its method structure, by using the convolutional neural network and the RGB brightness contrast method, the appropriate number of wind cap structures are determined to be placed at the dead corners of the air box pipeline. Multiple groups of wind caps are provided at the places where ash is likely to accumulate, and at the places where ash is not likely to accumulate, blowing is carried out through the characteristics of the wind cap structure. Since this wind cap structure has a large coverage range, a single wind cap structure can cover a large side wall of the air box. According to the RGB brightness contrast method and historical data, at the parts where ash is likely to accumulate, according to the thickness of the ash that is easy to accumulate, the number and position of the wind caps are determined, achieving the function of installing the fewest wind caps and maximizing the coverage range. Description of the Drawings

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention, and for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0030] Figure 1 It is a structural block diagram of the present invention;

[0031] Figure 2 It is one of the schematic diagrams of the wind cap structure of the present invention;

[0032] Figure 3 It is the second schematic diagram of the wind cap structure in the present invention;

[0033] Figure 4 It is the schematic sectional structure diagram of the wind cap structure in the present invention;

[0034] Figure 5 It is the flow chart of the steps in the present invention;

[0035] Figure 6 It is the block diagram of the system structure connection in the present invention

[0036] The reference numerals in the drawings respectively represent: 1. Wind cap structure; 11. Connection port; 12. Air outlet; 2. Side cylinder. Detailed Embodiments

[0037] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0038] The following further describes the present invention with reference to the embodiments.

[0039] Embodiment 1: A boiler air box ash cleaning system, referring to the attached Figure 6 , including an image acquisition module: used to obtain images of the areas in the historical flue that are prone to ash accumulation and images in the current usage state, and perform image preprocessing to obtain an image set;

[0040] A feature extraction module: used to extract physical features and regional organization features from the image set,

[0041] Image model learning module: It is used to select and train a machine model by using an image set, physical features, and regional tissue features, and then integrate the trained machine model to obtain a deep learning model;

[0042] Image analysis module: It is used to analyze a newly input clean flue gas duct picture by using the deep learning model to obtain an evaluation result of the flue gas duct picture;

[0043] Visualization module: It is used to generate a visualization report based on the analysis and prediction results;

[0044] The physical features include the sizes of the air box flue pipes, the ash accumulation range and thickness; the tissue features include pores and granularity.

[0045] The evaluation results of the flue gas duct picture include: one is the proportion of dust in the picture size within the picture range, and the other is the brightness and darkness of the dust. The soot cleaning range and soot cleaning time are determined according to the proportion number and the brightness and darkness of the dust.

[0046] Embodiment 2: A boiler air box soot cleaning method, referring to the attached Figure 5 , including a boiler air box soot cleaning system in Embodiment 1,

[0047] Step 1: By performing a position comparison analysis on the soot accumulation prone positions and historical dust coverage positions of the flue gas duct in the unused state, determine whether it is necessary to modify and optimize the soot cleaning points; if not, determine the minimum installation quantity and maximum coverage area of the wind caps. If necessary, adjust the quantity of the wind caps according to the historical dust coverage area to achieve a complete coverage effect, and add wind caps to the overlapping coverage part of the historical dust in the unused flue gas duct to complete the installation of the soot cleaning structure;

[0048] Step 2: Analyze the results based on the overlapping coverage part of the historical dust, mark this position, determine the soot accumulation prone points of the flue pipes, perform a picture brightness and darkness analysis on the overlapping coverage part of the historical dust to obtain the overlapping coverage brightness and darkness value, and judge the dust accumulation thickness level of the flue pipes according to the dust coverage brightness and darkness value, which are the first class, second class, third class, and fourth class, and the thickness level gradually becomes thinner according to the level order, and draw the dust thickness level contour lines on the picture.

[0049] The process of judging whether to modify and optimize the soot cleaning points is as follows: Take any point in the picture of the historical dust coverage position and perform a color difference analysis of the picture brightness and darkness with the same point in the flue gas duct in the unused state. If the color difference analysis coincides, it is an easily covered position and is marked as an easily covered point. If not, it is marked as a not easily covered point, and perform an RGB brightness and darkness model analysis on the picture:

[0050]

[0051] The value range of L is from 0 to 1 to obtain the light and darkness coincidence value L, where R, G, and B are the RGB values of the picture detection points. Compare the detected light and darkness coincidence value L with the detection coincidence threshold. If it is greater than the coincidence threshold, the position of the wind cap needs to be set. If it is not greater than the coincidence threshold, the position of the wind cap does not need to be set.

[0052] Step 3: Obtain the internal image of the air duct at the historical dust-covered position, evenly divide it into an image array at equal intervals. Based on any one piece of the image array, perform RGB ratio processing on the flue at the same position in the unused state. In one piece of the image array, further evenly divide it into standard squares with a quantity of 10 to obtain 10 groups of RGB values for this image array, and sum and average according to these 10 groups of RGB values to get t. Obtain several groups of image array averages of the historical dust-covered image, marked as t1, t2, t3....tn, to form a historical data set. Then train a convolutional neural network on the historical data set, and determine the value according to the output result, and compare it with the wind cap blowing range threshold. If it is greater than the wind cap blowing range threshold, the wind cap fixing points need to be opened according to the minimum range of the wind cap blowing. If it is not greater than the wind cap blowing range threshold, the wind cap fixing points can be opened according to the maximum range of the wind cap.

[0053] And the convolutional neural network includes two groups of convolutional neural modules. Each group of convolutional neural modules includes an input layer, multiple convolutional layers, a pooling layer, and a fully connected layer connected in sequence. First, input the historical data set into the two input layers respectively. Each convolutional neural network includes a first convolutional layer, a second convolutional layer, a max pooling layer, a third convolutional layer, a fourth convolutional layer, a fifth convolutional layer, a first fully connected layer, a second fully connected layer, and an output layer. Input the image array average value t into the first convolutional layer of the two convolutional modules respectively, and bring the feature data into the max pooling layer to make the fully connected layer output the category of the structural defect corresponding to the sample. The feature data includes the flue size and the flatness of the flue, where: Input layer: Receive the original image data or other types of grid structure data. Convolutional layer: Extract the local features of the input data through convolutional operations. Each convolutional kernel in the convolutional layer can extract a specific feature, and multiple convolutional kernels can work in parallel to extract different types of features. Pooling layer: Perform downsampling (or called dimensionality reduction) on the output of the convolutional layer to reduce the number of parameters and improve the calculation efficiency. Common pooling operations include max pooling and average pooling. Fully connected layer: Synthesize the features extracted by the previous layers for tasks such as classification or regression. Each neuron in the fully connected layer is connected to all neurons in the previous layer;

[0054] And within the first convolutional layer of each group of convolutional neural networks, there are 81 groups of convolutional kernels with a size of 10*10, a stride of 2, and a padding of 0 at the edges; the window size of the first pooling layer is 3*3 and the stride is 2; within the second convolutional layer, there are 93 groups of convolutional kernels with a size of 10*10, a stride of 3, and a padding of 1 at the edges; within the third convolutional layer, there are 81 groups of convolutional kernels with a size of 12*12, a stride of 3, and a padding of 1 at the edges, within the fourth convolutional layer, there are 108 groups of convolutional kernels with a size of 11*11, a stride of 2, and a padding of 0 at the edges, and within the fifth convolutional layer, there are 126 groups of convolutional kernels with a size of 11*11, a stride of 2, and a padding of 0 at the edges.

[0055] Embodiment 3: A boiler air box soot cleaning device, as Figures 1-4 shown. On the basis of Embodiment 1 and Embodiment 2, it includes a control component structure, a control board, and an intelligent control component. Both ends of the control board are electrically connected to the intelligent control component and the control component structure. The intelligent control component is connected to a plurality of valves and a solenoid valve structure communicating with miscellaneous gas. One end of the valve is connected to the wind cap structure 1. The wind cap structure 1 is a hollow cavity structure without a bottom surface at the bottom, and a plurality of air outlet openings 12 are provided on the side wall of the wind cap structure 1. The side wall of the air outlet opening 12 is fixedly connected to the side cylinder 2, and the air outlet opening 12 is in communication with the side cylinder 2;

[0056] There are multiple groups of the air outlet openings 12 provided on the wind cap structure 1 at intervals, and a connection port 11 is provided at the bottom end of the wind cap structure 1. The wind cap structure 1 is cast from ZG40CI25NI20SI2Re material, and the side cylinder 2 forms a 45° angle with the side wall support of the wind cap structure 1.

[0057] By setting the soot cleaning system and its method structure, using the method of convolutional neural networks and RGB brightness contrast, it is determined to place an appropriate number of wind cap structures 1 at the dead corners of the air box pipeline. At the places where accumulation is likely to occur, multiple groups of wind cap structures 1 are provided. At the places where accumulation is not likely to occur, blowing is carried out through the characteristics of the wind cap structure 1. Since this wind cap structure 1 has a large coverage range, a single wind cap structure 1 can cover a large side wall of the air box. According to the method of RGB brightness contrast and historical data, at the parts where accumulation is likely to occur, according to the thickness of the dust that is easily docked, the number and position of the wind caps can be determined, achieving the function of installing the fewest wind caps and maximizing the coverage range.

[0058] Therefore, through on-site survey, the size of the area with accumulated ash in the secondary air box is: the width is 36000mm, the depth is 2500mm, and the depth of accumulated ash is 1500mm. According to the characteristics of the accumulated ash in the secondary air box, the wind cap layout plan is designed as follows:

[0059] In the direction of the secondary air box depth of 2500 mm, a row of branch pipes with a total of three nozzles is arranged. The first nozzle is arranged 500 mm away from the air box wall surface. The second nozzle is 750 mm away from the first nozzle. The third nozzle is 750 mm away from the second nozzle and 500 mm away from the air box wall surface.

[0060] In the direction of the secondary air box width of 36000 mm, the first row of branch pipes is arranged 500 mm away from the air box wall surface. The second row of branch pipes is arranged 1000 mm behind the first row, with a total of 36 rows of branch pipes. Each row of branch pipes has a total of 3 nozzles, with a total of 108 nozzles. Every two rows of branch pipes are grouped into one set, and one solenoid valve is set for each set, with a total of 18 sets. The installation heights of the nozzles in the two rows of branch pipes in each set are: the nozzles of the first row of branch pipes are 300 mm away from the bottom of the air box, and the nozzles of the second row of branch pipes are 500 mm away from the bottom of the air box, arranged in a staggered manner. According to the situation of ash accumulation and ash cleaning process in the on-site secondary air box, some nozzles are appropriately arranged in the key gas flow dead corner areas. At the same time, considering the nozzle reservation and spare parts situation, our side promises that the total number of nozzles in this project is not less than 180.

[0061] Using compressed air as the medium for soot blowing, the compressed air enters from the bottom of the air cap structure 1 and flows out from the four air outlets 12 at the upper part of the air chamber, flowing into the four side cylinders 2. The medium is ejected at high speed from the bottom of the side cylinder 2. Since the side cylinder 2 forms a 45-degree angle with the side wall of the air cap structure 1, the compressed air is directionally sprayed towards the flue wall surface through the side cylinder 2, blowing and fluidizing the accumulated ash in the flue. The four side cylinders 2 ensure that all directions around the nozzle can be blown to. The directional jet of the side cylinder 2 enhances the disturbance of the accumulated ash, effectively solving the problem of accumulated ash accumulation in the flue.

[0062] During on-site operation, the compressed air for the soot blowing system is taken from the compressed air provided by the newly added air compressor. The compressed air pressure at the outlet of the air compressor should not be less than 1 MPa. The compressed air main pipe is a DN50 stainless steel pipe, and the soot blowing branch pipe is a DN20 stainless steel pipe. The medium in the pipe is 20°C compressed air (considered according to the highest compressed air pressure of 1 MPa). The supports and hangers of the soot blowing compressed air branch pipes are hung on the rigid beam at the bottom of the horizontal flue with rigid hangers. The flue soot blowing method is intermittent soot blowing. Each time, half a row is purged for 1 minute, and the interval between each row is 2 minutes, with a total of 20 rows. Completing a full soot blowing process takes 60 minutes, and soot blowing is carried out once every 3 days.

[0063] Therefore, through the connection and cooperation of DCS / PLC, the flue is cleaned by intelligent and rational soot blowing. Using compressed air as the purging medium, the purging air caps are accurately arranged in the flue ash accumulation area. Through the intelligent control compressed air purging system, the accumulated ash is purged and agitated at fixed times and points, and the accumulated ash is taken away by the flue gas flow, realizing the function of non-deposition of furnace ash.

[0064] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. A boiler air box ash cleaning system, characterized in that It includes an image acquisition module: which is used to obtain images of the areas prone to ash accumulation in the historical flue and images in the current usage state, and perform image preprocessing to obtain an image set; A feature extraction module: which is used to extract physical features and regional tissue features from the image set, An image model learning module: which is used to select and train a machine model by using the image set, physical features and regional tissue features, and then integrate the trained machine model to obtain a deep learning model; An image analysis module: which is used to analyze newly input clean-state flue pictures by using the deep learning model to obtain an evaluation result of the flue pictures; A visualization module: which is used to generate a visualization report based on the analysis and prediction results.

2. The boiler air box ash cleaning system according to claim 1, characterized in that, The physical features include the sizes of the air box and the flue pipe, the ash accumulation range and thickness; the tissue features include pores and granularity.

3. The boiler air box ash cleaning system according to claim 1, wherein The evaluation result of the flue picture includes: one is the proportion of dust in the picture size within the picture range, and the other is the brightness and darkness of the dust. The ash cleaning range and ash cleaning time are determined according to the proportion number and the brightness and darkness of the dust.

4. A method for cleaning the ash of a boiler air box, comprising a boiler air box ash cleaning system according to any one of claims 1-3, characterized in that, By performing position comparison and analysis on the ash accumulation prone positions in the unused flue and the historical dust covered positions, it is judged whether it is necessary to modify and optimize the ash cleaning points; if not, determine the minimum installation quantity of the wind caps and their maximum coverage area. If necessary, adjust the quantity of the wind caps according to the historical dust coverage area to achieve a complete coverage effect, and add wind caps at the overlapping coverage parts of the historical dust in the unused flue to complete the installation of the ash cleaning structure; Perform result analysis on the overlapping coverage part of the historical dust, mark this position, determine the positions where dust is prone to accumulate in the flue pipe, perform brightness and darkness analysis on the overlapping coverage part of the historical dust to obtain the overlapping coverage brightness and darkness value, and judge the dust accumulation thickness level of the flue pipe according to the dust coverage brightness and darkness value, which are the first level, the second level, the third level and the fourth level, and the thickness level gradually becomes thinner according to the level order, and draw the dust thickness level contour lines on the picture.

5. A method for cleaning ash from a boiler air box according to claim 4, characterized in that The process of judging whether to modify and optimize the ash cleaning points is as follows: Take any point in the picture of the historical dust covered position and perform color difference analysis on the brightness and darkness of the same point in the flue in the unused state. If the color difference analysis coincides, it is an easily covered position and is marked as an easily covered point. If it does not coincide, it is marked as a not easily covered point, and perform RGB brightness and darkness model analysis on the picture: The value range of L is 0 to 1 to obtain the brightness and darkness coincidence value L, where R, G, and B are the RGB values of the picture detection point. Compare the detected brightness and darkness coincidence value L with the detection coincidence threshold. If it is greater than the coincidence threshold, it is necessary to set the wind cap points. If it is not greater than the coincidence threshold, it is not necessary to set the wind cap points.

6. A method for cleaning the ash of a boiler air box according to claim 4, characterized in that, Obtain the internal image of the air duct at the historical dust-covered position, evenly divide it into an image array with equal spacing. Based on any one piece of the image array, perform RGB ratio processing on the flue at the same position in the unused state. In one piece of the image array, further evenly divide it into standard squares with a quantity of 10 to obtain 10 groups of RGB values for this image array, and sum and average these 10 groups of RGB values to get t. Obtain several groups of image array averages for the historical dust-covered image, marked as t1, t2, t3....tn, to form a historical dataset. Then train a convolutional neural network on the historical dataset, and determine the value according to the output result, and compare it with the air cap blowing range threshold. If it is greater than the air cap blowing range threshold, the air cap fixing points need to be opened according to the minimum range of the air cap blowing. If it is not greater than the air cap blowing range threshold, the air cap fixing points can be opened according to the maximum range of the air cap.

7. A method for cleaning ash from a boiler air box according to claim 6, characterized in that, The convolutional neural network includes two sets of convolutional neural modules. Each set of convolutional neural modules includes an input layer, multiple sets of convolutional layers, a pooling layer, and a fully connected layer connected in sequence. First, input the historical dataset into the two input layers respectively. Each set of convolutional neural networks includes a first convolutional layer, a second convolutional layer, a max pooling layer, a third convolutional layer, a fourth convolutional layer, a fifth convolutional layer, a first fully connected layer, a second fully connected layer, and an output layer. Input the image array average t into the first convolutional layer of the two convolutional modules respectively, and bring the feature data into the max pooling layer to make the fully connected layer output the category of the structural defect corresponding to the sample. The feature data includes the flue size and the flatness of the flue, where: Input layer: Receive the original image data or other types of grid structure data. Convolutional layer: Extract the local features of the input data through convolution operations. Each convolutional kernel in the convolutional layer can extract a specific feature, and multiple convolutional kernels can work in parallel to extract different types of features. Pooling layer: Perform downsampling (or dimensionality reduction) on the output of the convolutional layer to reduce the number of parameters and improve the calculation efficiency. Common pooling operations include max pooling and average pooling. Fully connected layer: Combine the features extracted by the previous layers for tasks such as classification or regression. Each neuron in the fully connected layer is connected to all neurons in the previous layer.

8. A method for cleaning ash from a boiler air box according to claim 7, characterized in that, In the first convolutional layer of each set of convolutional neural networks, there are 81 convolutional kernels with a size of 10*10, a stride of 2, and a padding of 0 at the edge; the window size of the first pooling layer is 3*3 and the stride is 2; in the second convolutional layer, there are 93 convolutional kernels with a size of 10*10, a stride of 3, and a padding of 1 at the edge; in the third convolutional layer, there are 81 convolutional kernels with a size of 12*12, a stride of 3, and a padding of 1 at the edge; in the fourth convolutional layer, there are 108 convolutional kernels with a size of 11*11, a stride of 2, and a padding of 0 at the edge; in the fifth convolutional layer, there are 126 convolutional kernels with a size of 11*11, a stride of 2, and a padding of 0 at the edge.

9. A boiler air box ash cleaning device, comprising any one of the boiler air box ash cleaning methods described in claims 4-8, characterized in that, It includes a control component structure, a control board, and an intelligent control component. Both ends of the control board are electrically connected to the intelligent control component and the control component structure. In the intelligent control component, there are structures of multiple valves and solenoid valves communicating with miscellaneous gas. One end of the valve is in communication with the wind cap structure (1). The wind cap structure (1) is a hollow cavity structure without a bottom surface at the bottom, and multiple groups of air outlets (12) are provided on the side wall of the wind cap structure (1). The side wall of the air outlet (12) is fixedly connected to the side cylinder (2), and the air outlet (12) is in communication with the side cylinder (2).

10. A method for cleaning ash from a boiler air box according to claim 4, characterized in that, Multiple groups of the air outlets (12) are provided on the wind cap structure (1) at intervals. A connection port (11) is provided at the bottom end of the wind cap structure (1). The wind cap structure (1) is cast with ZG40CI25NI20SI2Re material. The side cylinder (2) forms a 45° angle with the side wall support of the wind cap structure (1).

Images