A converter flue cinder monitoring system and method based on machine vision
By automatically monitoring converter flue slagging using a machine vision system, the problems of large errors in human observation and the inability to monitor around the clock have been solved. This has enabled efficient and low-cost flue slagging alarms, thereby improving smelting efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- МААНЬШАНЬ АЙРОН ЭНД СТИЛ КО ЛТД
- Filing Date
- 2023-02-08
- Publication Date
- 2026-07-07
AI Technical Summary
In existing technologies, monitoring of slagging in converter flues relies on human observation, which has large errors and cannot be monitored around the clock, leading to flue blockages and water leaks that affect smelting efficiency.
A machine vision-based converter flue slagging monitoring system is adopted. Visible light and infrared video are collected by a binocular monitoring camera, and image stitching and slagging detection are performed by the central control server. The slagging size is determined by OTSU and semantic segmentation algorithms, and an audible and visual alarm is triggered for timely alarm.
It achieves all-weather automated monitoring, reduces human error, improves the accuracy and timeliness of flue gas slagging monitoring, and reduces costs.
Smart Images
Figure CN116342481B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of safe operation and maintenance technology, specifically to a machine vision-based converter flue slagging monitoring system and method. Background Technology
[0002] Converters are crucial equipment in metallurgical enterprises, and their smelting flue gas deposits in the upper flue. When the converter is in operation, the slag hanging below the flue is visible to the naked eye. If excessive slag is not cleaned in time, it can lead to flue blockage, water leakage, and other malfunctions, thereby affecting smelting efficiency.
[0003] From a routine maintenance perspective, most metallurgical enterprises visually observe the deposits in the flue through cameras installed near the converter. However, visual observation relies too much on experience, has a large margin of error, and cannot monitor the deposits around the clock, making it easy to miss detections and trigger alarms in a timely manner.
[0004] From a visual inspection perspective, the slag below the flue exhibits an irregular, jagged shape, making its image features quite distinct. Therefore, it is entirely feasible to design a machine vision algorithm to detect the slag condition by using a camera to capture video footage of the area below the flue during converter rotation. Due to the complex operating conditions of the converter, it is necessary to rationally configure the video acquisition method to reduce environmental interference. Summary of the Invention
[0005] The purpose of this invention is to provide a machine vision-based converter flue slagging monitoring system and method to improve the automation level of converter flue slagging monitoring and solve the problems mentioned in the background art.
[0006] To achieve the above objectives, the present invention provides the following technical solution:
[0007] A machine vision-based converter flue slagging monitoring system includes binocular monitoring cameras, a central control server, and audible and visual alarms. The binocular monitoring cameras are installed on the left, right, and rear sides of the converter flue and can simultaneously capture visible light images and infrared video frames. The central control server is installed in the central control room. The audible and visual alarms are installed in the central control room and at the production site, with the number and location of the audible and visual alarms corresponding to the number and location of the binocular monitoring cameras.
[0008] Furthermore, the binocular monitoring cameras installed on the left, right, and rear sides of the converter flue wait for the limit switch signal from the converter rotation PLC. When the signal is captured, each binocular monitoring camera is triggered to simultaneously capture visible light and infrared video; otherwise, each binocular monitoring camera is in sleep mode. The central control server installed in the central control room simultaneously receives the video captured by each binocular monitoring camera, detects the slagging status of the flue, judges the detection results, and triggers the audible and visual alarms installed in the central control room and the converter production site as appropriate.
[0009] This invention provides another technical solution: a method for monitoring slagging in converter flue gas based on machine vision, comprising the following steps:
[0010] S1: Each binocular monitoring camera captures real-time video of the converter flue within its field of view, including the opening and closing areas of the converter and the converter flue, and sends the video of each opening and closing area to the central control server.
[0011] S2: The central control server calls the spectral segment video frame stitching algorithm to stitch all the opening and closing areas into one visible light panoramic video frame and one infrared visible light panoramic video frame. Based on the field of view center of each binocular monitoring camera, the empirical size value of the opening and closing area, and the preset threshold, the region of interest of the visible light panoramic video frame is set, and each region ID number is assigned.
[0012] S3: The central control server calls the slag detection algorithm to detect the slag size information inside the region of interest. If the slag size exceeds the warning value, it goes to S4; otherwise, it goes to S1.
[0013] S4: Trigger the audible and visual alarms in the central control room and converter production site based on the ID number of the region of interest where the slag exceeds the standard.
[0014] Furthermore, in S1, when the converter rotates, the PLC controller sends a limit switch signal to trigger each binocular monitoring camera to simultaneously collect real-time visible light and infrared video of the converter and the opening and closing area of the converter flue; when the converter resets, each binocular monitoring camera is turned off.
[0015] Furthermore, S2 employs a spectral segment video frame stitching algorithm to stitch visible light video frames and infrared video frames separately. In the visible light image, feature points of the furnace body and flue area are collected as key information for video frame matching; in the infrared image, feature points of the flame area are collected as key information for video frame matching.
[0016] Furthermore, the opening and closing area size of the converter in S2 is unstable with each rotation, and the camera's field of view is larger than this opening and closing area size, with the center of the field of view of camera i (x i y i Based on this, empirical size values (Width, Height) of the opening and closing region are introduced, and then a threshold θ is set to expand these size values as the region of interest (Rect) of the visible light panoramic video frame. i The details are as follows:
[0017] Rect i =[x i y i Width+θ, Height+θ]
[0018] In the formula, x i and y iLet θ be the center coordinates of the i-th region of interest, and Width+θ and Height+θ be the width and height of the i-th region of interest, respectively.
[0019] Furthermore, the central control server in S3 first uses the OTSU algorithm to segment the infrared panoramic video frames, sets a redundant area mask M based on the bright areas in the result image, and then removes the redundant areas in the visible light panoramic video frames based on the mask M.
[0020] Furthermore, in S3, a semantic segmentation algorithm is used to segment the i-th region of interest in the visible light panoramic video frame to obtain the furnace body outline, flue outline, and suspected slagging outline, and the j-th suspected slagging outline is detected. Bottom pixel chain x j Coordinates and y j The variances of the coordinates, D(x) and D(y), are such that if both variances are greater than the threshold ρ... x and ρ y This confirms that the suspected slag-forming area is indeed slag-forming, and it can be marked. Set it to 1, otherwise set it to 0; the method is as follows:
[0021]
[0022] Based on this, the profile of the j-th suspected slag is calculated. area value When the area value exceeds the warning value, an alarm signal and the ID number of the region of interest where the slag is located are given.
[0023] Furthermore, in S4, the alarm signal first triggers the audible and visual alarm installed in the central control room, and then, based on the ID number of the area of interest where the slag is located, it triggers the audible and visual alarm near the corresponding binocular monitoring camera to guide maintenance personnel to check and repair.
[0024] Compared with the prior art, the beneficial effects of the present invention are:
[0025] 1. The present invention provides a machine vision-based converter flue slagging monitoring system and method, which is small in size, easy to install, does not need to contact the converter flue body, and has little impact on the production line.
[0026] 2. The present invention provides a machine vision-based converter flue slagging monitoring system and method, which can replace the traditional human eye observation method, realize all-weather flue slagging monitoring and alarm, and is highly efficient and low in cost. Attached Figure Description
[0027] Figure 1 This is a side view of the system structure of the present invention;
[0028] Figure 2This is a top view of the system structure of the present invention;
[0029] Figure 3 This is a flowchart of the method of the present invention.
[0030] In the picture: 1. Binocular monitoring camera; 2. Central control server; 3. Audible and visual alarm; 4. Converter flue. Detailed Implementation
[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0032] Please see Figure 1-2 This invention provides a machine vision-based converter flue slagging monitoring system, including a binocular monitoring camera 1, a central control server 2, and an audible and visual alarm 3. The binocular monitoring camera 1 is installed on the left, right, and rear sides of the converter flue 4, and can simultaneously acquire visible light video and infrared video. The central control server 2 stores a spectral segmentation image stitching algorithm, a converter flue slagging detection algorithm, and an alarm mechanism, and can analyze the detection results to generate alarm information. The central control server 2 is installed in the central control room. The audible and visual alarm 3 is installed in the central control room and at the production site, and the number and location of the audible and visual alarm 3 correspond to the number and location of the binocular monitoring camera 1. The audible and visual alarm 3 triggers an alarm according to the alarm information.
[0033] In the above embodiment, the binocular monitoring cameras 1 installed on the left, right and rear sides of the converter flue 4 wait for the limit switch signal of the converter rotation PLC. When the signal is captured, each binocular monitoring camera 1 is triggered to simultaneously shoot visible light and infrared video; otherwise, each binocular monitoring camera 1 is in a sleep state. The central control server 2 installed in the central control room simultaneously receives the video captured by each binocular monitoring camera 1, detects the slagging status of the flue, judges the detection results and triggers the audible and visual alarms 3 installed in the central control room and the converter production site as appropriate.
[0034] Please see Figure 3 To further explain the embodiments of the present invention, a method for monitoring slagging in converter flue gas based on machine vision is also provided, comprising the following steps:
[0035] Step 1: Each binocular monitoring camera 1 captures real-time video of the converter flue 4 within its field of view, including the opening and closing areas of the converter and converter flue 4, and sends the video of each opening and closing area to the central control server 2.
[0036] Step 2: The central control server 2 calls the spectral segment video frame stitching algorithm to stitch all the opening and closing areas into one visible light panoramic video frame and one infrared visible light panoramic video frame. Based on the field of view center of each binocular monitoring camera 1, the empirical size value of the opening and closing area, and the preset threshold, the region of interest of the visible light panoramic video frame is set, and each region ID number is assigned.
[0037] Step 3: The central control server 2 calls the slagging detection algorithm to detect the slagging size information inside the region of interest. If the slagging size exceeds the warning value, it goes to S4; otherwise, it goes to S1.
[0038] Step 4: Trigger the audible and visual alarms 3 in the central control room and converter production site based on the ID number of the area of interest where the slag exceeds the standard.
[0039] In step one above, when the converter rotates, the PLC sends a limit switch signal to trigger the binocular monitoring camera 1 to collect real-time video of the opening and closing areas of the converter and flue within its field of view. This video includes visible light video and infrared video. Subsequently, the video of each opening and closing area is sent to the central control server 2. When the converter resets, the PLC controller sends a limit switch signal again to trigger the binocular monitoring camera 1 to enter sleep mode.
[0040] In step two above, the central control server 2 calls the spectral segment video frame stitching algorithm to stitch all open and closed areas into one visible light panoramic video frame and one infrared-visible light panoramic video frame. The specific implementation steps are as follows: SURF feature points are detected in the visible light video frame. Since the pixels of the furnace body and flue are relatively dark, while the pixels of the flame area are bright red, there is a significant difference between the two. Therefore, SURF feature points in the flame area can be filtered out, while those belonging to the furnace body and flue are retained as key information for matching the visible light video frame. Similarly, SURF feature points in the infrared video frame are detected. Color differences are also used to filter out SURF feature points belonging to the furnace body and flue, while those belonging to the flame area are retained as key information for matching the infrared video frame. Based on this, image matching is performed in the visible light video frame and the infrared video frame using the descriptors of the SURF feature points, respectively, to generate the visible light panoramic video frame and the infrared-visible light panoramic video frame.
[0041] In step two above, based on the field of view center of each camera, the empirical size value of the opening and closing area, and a preset threshold, the region of interest (ROI) for the visible light panoramic video frame is set, and an ID number is assigned to each region. The specific implementation steps are as follows: Using the field of view center (xi) of the i-th camera... i y i Based on this, empirical size values (Width, Height) of the opening and closing region are introduced, and then a threshold θ is set to expand these size values as the region of interest (Rect) of the visible light panoramic video frame. i The details are as follows:
[0042] Rect i =[x i y i Width+θ, Height+θ]
[0043] In the formula, x i and y i Let θ be the center coordinates of the i-th region of interest, and Width+θ and Height+θ be the width and height of the i-th region of interest, respectively.
[0044] In step three above, the central control server 2 invokes the slag detection algorithm to detect the slag size information within the region of interest. The specific implementation steps are as follows: The central control server 2 first uses the OTSU algorithm to segment the infrared panoramic video frames, and sets a redundant region mask M based on the highlighted areas in the result image. Then, redundant regions in the visible light panoramic video frames are removed based on the mask M.
[0045] Specifically, a semantic segmentation algorithm is used to segment the i-th region of interest in a visible light panoramic video frame, obtaining the furnace body outline, flue outline, and suspected slagging outline. The j-th suspected slagging outline is then detected. Bottom pixel chain x j Coordinates and y j The variances of the coordinates, D(x) and D(y). If both variances are simultaneously greater than the threshold ρ... x and ρ y This confirms that the suspected slag-forming area is indeed slag-forming, and it can be marked. Set it to 1, otherwise set it to 0. The method is as follows:
[0046]
[0047] Based on this, the profile of the j-th suspected slag is calculated. area value When the area value exceeds the warning value, an alarm signal and the ID number of the region of interest where the slag is located are given.
[0048] In step four above, the audible and visual alarms 3 in the central control room and converter production site are triggered based on the ID number of the region of interest where the slag exceeds the standard. The specific implementation steps are as follows: the audible and visual alarm 3 installed in the central control room is first triggered based on the alarm signal, and then the audible and visual alarm 3 near the corresponding binocular monitoring camera 1 is triggered based on the ID number of the region of interest where the slag is located, guiding maintenance personnel to check and repair.
[0049] The above description is merely an example to clearly illustrate the content involved in this invention and is not intended to limit the implementation of this invention. For example, the binocular surveillance camera is not limited to binocular cameras capable of simultaneously acquiring visible light and infrared video, but also includes any integrated or combined instrument capable of simultaneously acquiring visible light and infrared video; the binocular surveillance camera can also be equipped with cooling and self-cleaning devices according to the actual working conditions on site. The method of transmitting video to the central control server includes wired transmission and wireless transmission. Obvious system structures, methods, and applications made by those skilled in the art based on the above embodiments are still within the protection scope of this invention.
Claims
1. A method for monitoring slagging in converter flue gas based on machine vision, characterized in that: The monitoring system includes binocular monitoring cameras (1), a central control server (2), and an audible and visual alarm (3). The binocular monitoring cameras (1) are installed on the left, right, and rear sides of the converter flue (4) and can simultaneously capture visible light images and infrared video frames. The central control server (2) is installed in the central control room. The audible and visual alarm (3) is installed in the central control room and the production site, and the number and location of the audible and visual alarm (3) correspond to the number and location of the binocular monitoring cameras (1). The binocular monitoring cameras (1) installed on the left, right, and rear sides of the converter flue (4) wait for the converter rotation PLC limit switch signal. When the signal is captured, each binocular monitoring camera (1) is triggered to simultaneously capture visible light and infrared video. Otherwise, each binocular monitoring camera (1) is in a dormant state. The central control server (2) installed in the central control room simultaneously receives the video captured by each binocular monitoring camera (1), detects the slagging status of the flue, judges the detection results, and triggers the audible and visual alarm (3) installed in the central control room and the converter production site as appropriate, using the following steps: S1: Each binocular monitoring camera (1) captures the converter flue (4), collects real-time video of the opening and closing areas of the converter and converter flue (4) within the field of view, and sends the video of each opening and closing area to the central control server (2). S2: The central control server (2) calls the spectral segment video frame splicing algorithm to splice all the open and closed areas into a visible light panoramic video frame and an infrared visible light panoramic video frame. Based on the field of view center of each binocular monitoring camera (1), the empirical size value of the open and closed area, and the preset threshold, the region of interest of the visible light panoramic video frame is set, and each region ID number is assigned. S3: The central control server (2) first uses the OTSU algorithm to segment the infrared panoramic video frame, sets a redundant region mask M according to the bright areas in the result image, and then removes the redundant areas in the visible light panoramic video frame according to the mask M; then it uses the semantic segmentation algorithm to segment the i-th region of interest in the visible light panoramic video frame to obtain the furnace body outline, flue outline and suspected slagging outline, and detects the j-th suspected slagging outline. Bottom pixel chain coordinates and The variances of the coordinates, D(x) and D(y), if both variances are greater than a threshold. and This confirms that the suspected slag-forming area is indeed slag-forming, and it can be marked. Set it to 1, otherwise set it to 0; the method is as follows: Based on this, the profile of the j-th suspected slag is calculated. area value When the area value is greater than the warning value, an alarm signal and the ID number of the region of interest where the slag is located are given. The size information of the slag inside the region of interest is detected. If the size of the slag exceeds the warning value, the process goes to S4; otherwise, it goes to S1. S4: Trigger the audible and visual alarms (3) in the central control room and converter production site based on the ID number of the region of interest where the slag exceeds the standard.
2. The method for monitoring slagging in converter flue gas based on machine vision as described in claim 1, characterized in that, When the converter rotates in S1, the PLC controller gives a limit switch signal to trigger each binocular monitoring camera (1) to simultaneously collect real-time visible light and infrared video of the opening and closing areas of the converter and converter flue (4); when the converter is reset, each binocular monitoring camera (1) is turned off.
3. The method for monitoring slagging in converter flue gas based on machine vision as described in claim 1, characterized in that, In S2, a segmented video frame stitching algorithm is used to stitch visible light video frames and infrared video frames separately. In the visible light image, feature points of the furnace body and flue area are collected as key information for video frame matching; in the infrared image, feature points of the flame area are collected as key information for video frame matching.
4. The method for a converter flue gas slagging monitoring system based on machine vision as described in claim 1, characterized in that, The opening and closing area size of the S2 converter is unstable with each rotation, and the camera's field of view is larger than this opening and closing area size, with the center of camera i's field of view... Based on this, empirical dimensional values for the opening and closing areas are introduced. Then set the threshold. Expand this size value as the region of interest for a visible light panoramic video frame. The details are as follows: In the formula, and Let i be the coordinates of the center of the i-th region of interest. and These are the width and height of the i-th region of interest, respectively.
5. The method for a converter flue gas slagging monitoring system based on machine vision as described in claim 1, characterized in that, S4 first triggers the audible and visual alarm (3) installed in the central control room based on the alarm signal. Then, based on the ID number of the area of interest where the slag is located, it triggers the audible and visual alarm (3) near the corresponding binocular monitoring camera (1) to guide maintenance personnel to check and repair.