A sintering machine grate bar fault detection method

Abstract

The present application belongs to the technical field of image processing, and discloses a sintering machine grate bar fault detection method. The method proposes a grate bar real-time detection method based on dynamic target detection mask. Starting from the motion state of the trolley and the static state of the stain shielding, the accuracy and false positive rate of the online detection of the sintering machine grate bar fault are balanced, the frame difference method is used as the basic method, and the morphological processing and grate bar image structure information are used as the auxiliary to confirm the grate bar fault condition. In the method, the frame difference method and the threshold method are used as the basis for fault detection of the grate bar, the target detection mask for camera shielding detection is generated and updated, the shielding and interference features in the grate bar image are excluded from the fault detection range, thereby reducing the false positive rate of fault detection. The present application can avoid the interference of environmental factors on the grate bar fault detection result, and improve the detection accuracy of the grate bar fault.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, and in particular relates to a method for detecting faults in the grate bars of a sintering machine. Background Technology

[0002] In sintering production, the sintering machine trolley is constantly exposed to high temperatures and corrosive gas flows, which can easily cause the grate bars to thin and break. Simultaneously, external forces such as trolley movement and uneven material distribution can lead to increased grate bar gaps or even detachment. These conditions can worsen the trolley's sealing performance, increase air leakage, cause sintered ore to fall out, affect the quality of the sintered ore, and even cause secondary accidents. Therefore, real-time monitoring of excessive grate bar gaps and breakage in the sintering machine is an essential part of the intelligent upgrading and transformation of sintering processes.

[0003] Due to the generally high temperatures and large amounts of dust in sintering workshops, the operating environment is quite harsh. Currently, the common practice for monitoring the operating status of sintering machine grates is to conduct regular manual inspections at the steel plant site or to use cameras for real-time manual observation. However, with the intelligent upgrading and transformation of sintering, the adoption of machine vision-based online monitoring methods is gradually becoming an inevitable trend. Pan Dong et al. proposed a deep network based on an attention mechanism that integrates motion-guided feature modules and stripe feature modules for online detection of grate defects in sintering trolley furnaces, solving the technical problem of low accuracy in online detection of grate defects in sintering trolley furnaces. Qin Lipeng developed a grate fault monitoring device based on image processing technology. This device acquires grate images through camera calibration, extracts the grate contour after image denoising, and detects grate faults based on the grate contour spacing information. These methods can more effectively monitor the fault status of sintering machine grates. Currently, the experimental environment for studying grate bar fault detection in sintering production is relatively ideal. However, in actual real-time grate bar detection, complex environmental factors can interfere with the detection results, causing false alarms and reducing system efficiency. Therefore, how to avoid the interference of environmental factors on grate bar fault detection results and improve the accuracy of grate bar fault detection is a problem that needs to be solved. Summary of the Invention

[0004] This invention aims to provide a method for detecting grate bar faults in sintering machines, addressing the problem that complex environmental factors can interfere with real-time grate bar detection, affecting the detection results and causing false alarms in the detection system. This method proposes a real-time grate bar detection approach based on a dynamic target detection mask. Starting with the motion state of the trolley and the static state obscured by dirt, the goal is to balance the accuracy and false alarm rate of online grate bar fault detection. Frame difference is used as the basic method, supplemented by morphological processing and grate bar image structure information to confirm the grate bar fault status.

[0005] The technical solution of the present invention:

[0006] A method for detecting faults in sintering machine grate bars includes the following steps:

[0007] Step 1: Real-time acquisition of the original grate bar image, and extraction of the detection mask from the original grate bar image. The detection mask is a mask of the original grate bar image that is calibrated.

[0008] Step 2: Perform CLAHE equalization on the original grate image to equalize the image brightness of the original grate image, and use the equalized original grate image as the equalized image;

[0009] Step 3: Due to the complex environment of the production site, the camera lens is contaminated. It is necessary to remove and suppress occlusion and interference factors, which requires occlusion area detection. The occlusion area detection method is as follows: the first frame of the grating image acquired by the detection system is thresholded to equalize the first frame of the grating image to determine the binary image; the image pixel values ​​of the binary image are determined. When the image pixel values ​​are lower than the target pixel threshold (i.e., the custom threshold), the area corresponding to the image pixel values ​​lower than the target pixel threshold is taken as the occlusion area.

[0010] Step 4: Remove occluded regions from the mask to be detected and determine the target detection mask. Based on the target detection mask, extract regions from the equalized image to remove occluded regions. Identify the n target pixels with the smallest pixel values ​​in the occluded regions, perform dilation-erosion-dilation morphological processing on these target pixels, determine the connected regions of the processed target pixels, and determine whether there are any abnormal grate patterns in the connected regions based on their aspect ratio.

[0011] The beneficial effects of this invention are as follows: When detecting faults in grate bars, based on the frame difference method and threshold method, by generating and updating the target detection mask for camera occlusion detection, occlusion and interference features in the grate bar image are excluded from the fault detection range, thereby reducing the false alarm rate of fault detection. This invention can avoid the interference of environmental factors on the grate bar fault detection results and improve the accuracy of grate bar fault detection. Attached Figure Description

[0012] Figure 1 This is a flowchart of a method for detecting faults in the grate bars of a sintering machine.

[0013] Figure 2 A flowchart for determining the mask area to be detected.

[0014] Figure 3 Example images are generated for the target detection mask of the starting frame, where (a) is the difference image of the starting frame and (b) is the target detection mask of the starting frame.

[0015] Figure 4 A flowchart for determining the occlusion area of ​​the second frame of the grating image.

[0016] Figure 5 A flowchart for determining the occlusion area of ​​the intermediate frame grating image.

[0017] Figure 6 Here are example images of candidate detection regions, where (a) is the original candidate region, (b) is the target detection mask, and (c) is the processed candidate region. Detailed Implementation

[0018] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings and technical solutions.

[0019] The flowchart of the sintering machine grate bar fault detection method provided by this invention is as follows: Figure 1 As shown, the specific content is as follows:

[0020] (1) Image equalization processing:

[0021] The original grate bar image is acquired in real time, and CLAHE equalization is performed on the original grate bar image to equalize the image brightness. The original grate bar image after equalization is used as the equalized image.

[0022] Specifically, the original grate image is divided into several sub-images, and the gray-level histogram of each sub-image is calculated. Adaptive equalization is then applied to the gray-level histograms, and linear interpolation is performed on each sub-image based on its location within the region to determine the enhanced image. This effectively enhances the local features of the grate, highlighting detailed information while reducing noise interference.

[0023] (2) Generate the detection mask for the original comb image:

[0024] The original grate image's detection mask represents the stationary interference area obstructing the lens during the trolley's movement of the grate. The detection mask is used to select areas with good imaging performance from the grate image with uneven illumination distribution for grate detection. The detection mask is obtained through comprehensive statistical calculation based on the direction of the grate's dynamic movement and the brightness distribution of the grate image.

[0025] Specifically, such as Figure 2As shown, firstly, the direction of the dynamic movement of the grate bars is determined. Specifically, two relatively close frames of original grate bar images are selected, for example, two frames with a two-second time interval. The SIFT algorithm is used to detect key points in the two relatively close frames, and descriptors for the detected key points are calculated to determine the descriptors of the two frames. A matching algorithm is then used to match the descriptors of the two frames to determine matching points. These matching points are then connected, and the angle of motion of the two frames is determined based on the connection results. Finally, the boundary angle of the mask to be detected is determined based on the angle of motion of the two frames. It should be noted that the SIFT algorithm detects key points in the original grate bar images. These key points are points with salientity and uniqueness in the original grate bar images, such as edges and corners. The number of key points detected can be adjusted according to parameters. For each detected key point, the SIFT algorithm calculates a descriptor. A descriptor is a vector that describes the content and structure of a local region of an image around a keypoint. Since the SIFT algorithm is scale-invariant, descriptors can handle image scaling variations. Multiple sets of data with complete grate areas are selected for comprehensive statistical calculation of the grate image's brightness distribution. Pixel thresholds for different brightness stages are selected to obtain information feature maps of brightness changes in the grate image. Based on the brightness distribution and information feature maps of brightness changes in the original grate image, the changes in highlight information at different brightness stages are determined, thereby generating the direction vector of brightness changes in the original grate image. The corresponding mask region to be detected in the original grate image is determined based on this direction vector.

[0026] For example, pixel thresholds for different brightness stages can be preset, such as setting three levels of pixel thresholds to determine which level each pixel in the original grate image belongs to, and determining the information feature map of brightness change of the grate image based on the pixel level determination results of each pixel.

[0027] (3) Extract the occluded areas from the original comb image:

[0028] Due to the complex environment of the production site, camera lenses may become contaminated. It is necessary to remove and suppress obstructions and interference factors, and thus, obstruction area detection is required.

[0029] The method for detecting occlusion areas is as follows: determine whether the original grating image acquired in the current frame is the first frame grating image. If so, perform threshold segmentation on the equalized image of the first frame grating image to determine the binary image; determine the image pixel values ​​of the binary image. When the image pixel values ​​of the binary image are lower than the third pixel threshold, the area corresponding to the image pixel values ​​lower than the third pixel threshold is taken as the occlusion area.

[0030] Specifically, each pixel value in the binary image is compared with a third pixel threshold, and the region corresponding to the pixel value below the third pixel threshold is determined as the occlusion region of the first frame of the grating image.

[0031] It should be noted that each time a grate bar image is captured, an original grate bar image can be captured. The first frame of the grate bar image refers to the original grate bar image captured at the very beginning.

[0032] (4) Generate the target detection mask region:

[0033] The target detection mask region refers to the mask region determined by removing the occluded areas from the mask to be detected.

[0034] If the original grate image captured in the current frame is the first frame grate image, then the occlusion area is determined according to the threshold method. That is, the gray value of each pixel in the original grate image is determined, and the gray value of each pixel is compared with the preset gray value threshold. The area corresponding to the pixel with the gray value greater than the gray value threshold is determined as the occlusion area.

[0035] The target detection mask region is generated based on the occlusion area and the mask to be detected in the first frame of the grating image.

[0036] If the original grate image captured in the current frame is not the first frame grate image, determine the mask region to be detected corresponding to the original grate image. Then, based on the original grate image and the previous frame grate image, determine the occlusion region of the original grate image. Remove the occlusion region from the mask region to be detected, and determine the target detection mask region. An example image of target detection mask generation for the starting frame is shown below. Figure 3 As shown.

[0037] It should be noted that, as Figure 4 As shown, if the original grate image of the current frame is the second frame grate image, then first determine the target detection mask of the original grate image of the previous frame, then determine the occlusion area of ​​the original grate image of the current frame based on the original grate image and the first frame grate image, remove the occlusion area from the target detection mask, and determine the target detection mask of the original grate image.

[0038] like Figure 5 As shown, Figure 5An image that reflects the degree of change refers to an image that reflects the changes in the grayscale values ​​of the image. If the original grate image of the current frame is neither the first frame grate image nor the second frame grate image, the area with lower grayscale values ​​and smaller changes in the original grate image of the current frame can be determined as the occlusion area based on the inter-frame difference between the original grate image of the current frame and the original grate image of the previous frame. Specifically, the process involves: determining an equalized image of the original grate image based on the current frame's original grate image; determining an equalized image of the previous frame's original grate image based on the previous frame's original grate image; determining the grayscale value variation information of the original grate image based on the equalized images of the original and previous frames; identifying regions with low grayscale values ​​and minimal variation in the current frame's original grate image based on the grayscale value variation information of the current and previous frames' original grate images; and designating these regions as occlusion regions. Finally, a target detection mask for the current frame's original grate image is determined based on the target mask region and the occlusion region.

[0039] As the target detection mask is generated iteratively, its representation of interference features becomes increasingly accurate.

[0040] (5) Validity detection of the original comb image:

[0041] The ratio of the number of pixels smaller than a first pixel threshold in the equalized image of the original grate image to the total number of pixels in the original grate image is determined. This ratio can be used as the percentage of completely black areas in the original grate image. If the percentage of completely black areas is greater than a preset percentage threshold, the original grate image is determined to be invalid, and the next frame of the original grate image is acquired.

[0042] Furthermore, since light shining on dust can cause white smoke-like areas to appear on the grate image, affecting the grate fault detection results, it is also necessary to determine whether white smoke-like areas exist in the original grate image. Specifically, if the proportion of completely black areas in the equalization image is less than or equal to a preset area proportion threshold, then the areas containing pixels with pixels smaller than the first pixel threshold in the equalization image are considered invalid areas and removed from the equalization image to determine the original detection area.

[0043] The process involves determining whether the original grate image is the first frame of the acquired grate image. If so, the original detection region is used as a candidate detection region. Otherwise, the target detection mask of the previous frame of the original grate image is obtained, and this mask is removed from the original detection region to determine the candidate detection region. An example of candidate detection region determination is shown in the image below. Figure 6As shown. The average pixel value of the candidate detection region is calculated. If the average pixel value of the candidate detection region is greater than the preset second pixel threshold, it is determined that there is a white smoke-like region in the candidate detection region, and the original grating image is determined to be an invalid image. The next frame of the original grating image is then acquired.

[0044] (6) Transition region detection of the original comb image

[0045] In actual detection, there may be instances where a transition area of ​​the grate bars is mistaken for a broken area. Therefore, it is necessary to determine whether the edge region of the candidate detection area is indeed a transition area. Specifically, the edge region of the candidate detection area is extracted, and the proportion of the candidate detection area is determined based on the ratio of the number of pixels in the edge region to the number of pixels in the target detection mask of the original grate bar image in the current frame. That is, the proportion of the candidate detection area = number of pixels in the edge region / number of pixels in the mask. If the proportion of the candidate detection area is lower than a preset value, the edge region is determined to be a transition area, meaning the grate bar is not broken. Conversely, if the proportion of the candidate detection area is higher than or equal to the preset value, the edge region is determined to be the edge of a broken grate bar, meaning the grate bar is broken.

[0046] (7) Detection of abnormalities in the comb bars:

[0047] The n target pixels with the smallest pixel values ​​in the candidate detection region are identified. It should be noted that these n target pixels with the smallest pixel values ​​are generally located at the notch positions. Morphological processing (dilation-erosion-dilation) is performed on the extracted target pixels. Based on the morphological processing results, the connected regions of the processed target pixels are determined, and the aspect ratio of each connected region is also determined. There must be at least one connected region. The aspect ratio of each connected region is calculated and determined. Based on the aspect ratio of each connected region, it is determined whether any grate anomalies exist within the connected region.

[0048] Specifically, if the aspect ratio of the connected region is less than a preset aspect ratio threshold, then the pixel value of the connected region is determined to be within the specified threshold range and meets the requirements for the average pixel value of the grate mask area, the proportion of black areas in the mask, and the number of edge pixels in the mask area. In other words, the connected region has a grate abnormality, and a grate abnormality signal can be output. If the aspect ratio of the connected region is greater than or equal to the preset aspect ratio threshold, then the connected region does not have a grate abnormality, and a grate normal signal can be output.

Claims

1. A method for detecting faults in the grate bars of a sintering machine, characterized in that, The steps are as follows: Step 1: Acquire the original grate bar image in real time and extract the mask to be detected from the original grate bar image. The mask to be detected in the original grate bar image is used to represent the static interference area that is blocked in front of the lens during the operation of the trolley driving the grate bar. The detection mask is used to select areas with good imaging performance from the original grate image with uneven illumination distribution for grate detection. The detection mask is obtained by comprehensive statistical calculation based on the direction of the dynamic movement of the grate and the brightness distribution of the original grate image. Step 2: Perform CLAHE equalization on the original grate image to equalize the image brightness of the original grate image, and use the equalized original grate image as the equalized image; Step 3: Due to the complex environment of the production site, the camera lens is contaminated. It is necessary to remove and suppress occlusion and interference factors, and occlusion area detection is required. The method for occlusion area detection is as follows: the first frame of the baffle image is acquired, and the equalized image of the first frame of the baffle image is thresholded to determine the binary image; the image pixel values ​​of the binary image are determined, and when the image pixel values ​​are lower than the target pixel threshold, the area corresponding to the image pixel values ​​lower than the target pixel threshold is taken as the occlusion area. Step 4: Remove occluded areas from the mask to be detected and determine the target detection mask; extract region images from the equalized image based on the target detection mask, and extract the unoccluded areas from the equalized image; determine the n target pixels with the smallest pixel values ​​in the unoccluded areas, perform dilation-erosion-dilation morphological processing on the target pixels, determine the connected regions of the processed target pixels, and determine whether there are any abnormal grate phenomena in the connected regions based on the aspect ratio of the connected regions; (4.1) Generate target detection mask region The target detection mask region refers to the mask region determined by removing the occluded areas from the mask to be detected; (4.2) Validity detection of the original comb image The ratio of the number of pixels smaller than the first pixel threshold in the equalized image of the original grate image to the total number of pixels in the original grate image is determined. This ratio is used as the proportion of completely black areas in the original grate image. If the proportion of completely black areas is greater than a preset area proportion threshold, the original grate image is determined to be an invalid image. The next frame of the original grate image is then acquired and the next frame of the original grate image is then detected. The detection steps are the same as those for the current frame of the original grate image. (4.3) Transition region detection of the original comb image Determine whether the edge region of the candidate detection area is a transition region of the baffle. Specifically, extract the edge region of the candidate detection area, and determine the proportion of the candidate area based on the ratio of the number of pixels in the edge region to the number of pixels in the target detection mask of the original baffle image in the current frame; that is, the proportion of the candidate area = the number of pixels in the edge region / the number of pixels in the mask. If the proportion of the candidate area is lower than a preset value, the edge region is determined to be a transition region, that is, the baffle is determined to be without breakage; conversely, if the proportion of the candidate area is higher than or equal to the preset value, the edge region is determined to be the edge of the baffle breakage, that is, the baffle is broken. (4.4) Detection of abnormal grate bars Identify the n target pixels with the smallest pixel values ​​in the candidate detection region. These n target pixels with the smallest pixel values ​​are generally located at the gap positions. Perform dilation-erosion-dilation morphological processing on the extracted target pixels. Based on the morphological processing results, determine the connected regions of the processed target pixels and the aspect ratio of the connected regions. There must be at least one connected region. Calculate the aspect ratio of each connected region to determine whether there is a grate anomaly in the connected region.

2. The method for detecting faults in the sintering machine grate bars according to claim 1, characterized in that, The specific implementation process of step 1 is as follows: First, determine the direction of the dynamic movement of the grate bars. Select two original grate bar images that are close to each other. Use the SIFT algorithm to detect key points in the two original grate bar images that are close to each other, and calculate the descriptors of the detected key points to determine the descriptors of the two original grate bar images. Match the descriptors of the two original grate bar images using a matching algorithm to determine the matching points on the two original grate bar images based on the matching results. Connect the matching points and determine the angle of movement of the two original grate bar images based on the connection results. Determine the boundary angle of the mask to be detected based on the angle of movement of the two original grate bar images. Multiple sets of data with complete grate bar areas are selected for comprehensive statistical calculation of the brightness distribution of the grate bar image; pixel thresholds for different brightness stages are selected to obtain the information feature map of brightness change in the original grate bar image; based on the brightness distribution and the information feature map of brightness change in the original grate bar image, the change of highlight information in different brightness stages is determined, thereby generating the direction vector of brightness change in the original grate bar image; based on the direction vector of brightness change in the original grate bar image, the corresponding mask area to be detected in the original grate bar image is determined. Pixel thresholds for different brightness stages are preset. Based on the preset pixel thresholds, the stage to which each pixel in the original grate image belongs is determined. Based on the pixel stage determination results for each pixel, the information feature map of brightness change in the grate image is determined.

3. The method for detecting faults in the sintering machine grate bars according to claim 1, characterized in that, The specific implementation process of step 2 is as follows: Divide the original comb image into several sub-images, calculate the gray-level histogram of each sub-image, perform adaptive equalization processing on the gray-level histogram, and perform linear interpolation processing on the sub-images according to the location of each sub-image to determine the enhanced image.

4. The method for detecting faults in the sintering machine grate bars according to claim 1, characterized in that, The method for detecting occlusion areas in step 3 is as follows: determine whether the original grating image acquired in the current frame is the first frame grating image. If so, perform threshold segmentation on the equalized image of the first frame grating image to determine the binary image; determine the image pixel value of the binary image. When the image pixel value of the binary image is lower than the third pixel threshold, the area corresponding to the image pixel value lower than the third pixel threshold is taken as the occlusion area.

5. The method for detecting faults in sintering machine grate bars according to claim 1, characterized in that, The specific steps for generating the target detection mask region are as follows: If the original grate image captured in the current frame is the first frame grate image, then the occlusion area is determined according to the threshold method, that is, the gray value of each pixel in the original grate image is determined, and the gray value of each pixel is compared with the preset gray value threshold to determine the area corresponding to the pixel whose gray value is less than the gray value threshold as the occlusion area. Generate a target detection mask region based on the occlusion area and the mask to be detected in the first frame of the grating image; If the original grate image captured in the current frame is not the first frame grate image, determine the mask area to be detected corresponding to the original grate image, and determine the occlusion area of ​​the original grate image based on the original grate image and the previous frame grate image of the original grate image. Remove the occlusion area from the mask area to be detected and determine the target detection mask area. If the original grate image of the current frame is the second frame grate image, then first determine the detection mask of the original grate image of the previous frame, then determine the occlusion area of ​​the original grate image of the current frame based on the original grate image and the first frame grate image, remove the occlusion area from the detection mask, and determine the target detection mask of the original grate image. An image reflecting the degree of change refers to an image that reflects the changes in the grayscale values ​​of an image. If the original grate image of the current frame is neither the first nor the second frame, the region with lower grayscale values ​​and smaller changes in the original grate image of the current frame is determined as the occlusion region based on the inter-frame difference between the original grate image of the current frame and the original grate image of the previous frame. Specifically, the equalization image of the original grate image is determined based on the original grate image of the current frame, and the equalization image of the previous grate image is determined based on the original grate image of the previous frame. Based on the equalized images of the original and previous frames of the grate image, the grayscale value change information of the original grate image is determined. Based on the grayscale value change information of the current frame's original grate image and the grayscale value change information of the previous frame's original grate image, the region with low grayscale value and small change in the current frame's original grate image is determined, and the region with low grayscale value and small change in the current frame's original grate image is taken as the occlusion region. Based on the target mask region and the occlusion region of the current frame's original grate image, the target detection mask of the current frame's original grate image is determined.

6. The method for detecting faults in the sintering machine grate bars according to claim 1, characterized in that, The validity detection of the original grate image also needs to determine whether there are white smoke-like areas in the original grate image; specifically, if the proportion of completely black areas in the equalization image is less than or equal to the preset area proportion threshold, then the area where the pixel corresponding to the pixel point less than the first pixel threshold in the equalization image is located is regarded as an invalid area, and the invalid area is removed from the equalization image to determine the original detection area. Determine whether the original grate image is the first frame of the acquired grate image. If so, use the original detection region as a candidate detection region. If not, obtain the target detection mask of the previous frame of the original grate image of the currently processed original grate image, and remove the target detection mask of the previous frame of the original grate image from the original detection region to determine the candidate detection region. Calculate the average pixel value of the candidate detection region. If the average pixel value of the candidate detection region is greater than the preset second pixel threshold, it is determined that there is a white smoke-like region in the candidate detection region, and the original grate image is determined to be an invalid image. Continue to acquire the next frame of the original grate image.

7. The method for detecting faults in sintering machine grate bars according to claim 1, characterized in that, The specific determination of the grate bar anomaly detection is as follows: if the aspect ratio of the connected region is less than the preset aspect ratio threshold, then the pixel value of the connected region is determined to be within the specified threshold range and meets the requirements of the average pixel value of the grate bar mask area, the proportion of black area in the mask, and the number of edge pixels in the mask area. That is, there is a grate bar anomaly in the connected region, and a grate bar anomaly signal can be output at this time; if the aspect ratio of the connected region is greater than or equal to the preset aspect ratio threshold, then there is no grate bar anomaly in the connected region, and a grate bar normal signal can be output at this time.

Images