A method for detecting material breakage in a cement raw material conveyor belt

Abstract

The present invention discloses a method for detecting material breakage of a cement raw material conveyor belt, and the field of deep learning image processing technology, including a method for detecting material breakage of a cement raw material conveyor belt, which includes 5 modules, namely, an image acquisition module, an AI classification module, a conveyor state analysis module, an automatic alarm module and an equipment linkage module; the present invention proposes a complete set of conveyor breakage discrimination methods, which can automatically monitor the state of materials on the conveyor in real time throughout the day through real-time intelligent analysis of monitoring cameras, and once a material breakage occurs on the conveyor, an alarm is immediately issued to solve the problem in time. Manual inspections are reduced, the quality of cement production is improved, and the continuous cement production is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of deep learning image processing, and in particular to a method for detecting material breakage in a cement raw material conveyor belt. Background Art

[0002] With the improvement of economy and the advancement of technology, the production scale and construction scale of cement plants are also expanding, and the production equipment is constantly iterating towards automation and unmanned operation, so there is a need to improve the productivity of cement. As the most widely used conveying equipment in the cement production process, the state of the material on the conveyor affects the cement production status of the entire factory. If the material is broken during the transmission process of the conveyor, it will have a certain impact on the final production of cement, and even cause the cement production line to stop operating, directly causing incalculable economic losses. According to statistics, more than 90% of the reduction in cement production efficiency due to conveyor accidents is related to untimely and inadequate inspections.

[0003] At present, the situation of missing materials in the conveyor transportation process is mainly monitored by manual inspection. This monitoring method cannot report and process the material breakage information in a timely and accurate manner at the first time when the material breakage occurs, which will cause immeasurable production and economic losses. The manual inspection method for conveyor material breakage is not only extremely inefficient and wastes human resources, but also cannot detect the conveyor material breakage in time, and can no longer meet the production requirements of larger scale and higher quality. Summary of the invention

[0004] The purpose of the present invention is to solve the problem in the prior art that the material status on the conveyor cannot be discovered in time and cannot be quickly processed during cement raw material transportation, resulting in increased production costs caused by delays in the production process, and a method for detecting material breakage in a cement raw material conveyor belt is proposed.

[0005] A method for detecting material breakage of a cement raw material conveyor belt, including a method for detecting material breakage of a cement raw material conveyor belt, including 5 modules, namely, an image acquisition module, an AI classification module, a conveyor state analysis module, an automatic alarm module and an equipment linkage module, and the following steps are performed through the above modules:

[0006] a. Start the system and load computer and hardware resources, as well as the classification model;

[0007] b. Obtain the image to be classified. First, decode the video stream transmitted by the surveillance camera into image data. Then, cut out the area included in the ROI from the image according to the preset rectangular ROI area.

[0008] c. Send the ROI image cropped in step b to the YOLOv8 classification network for AI analysis to obtain the conveyor status classification result;

[0009] d. According to the conveyor classification results obtained by reasoning, analyze and determine the state information of the conveyor when it is started and the conditions that are satisfied when the material is cut off during operation;

[0010] e. Determine whether the material status is normal when the conveyor is started based on the initial status analysis. If it is normal, return to step b to start a new round of status analysis; if it is abnormal, push the information to the equipment personnel for inspection through the equipment alarm;

[0011] f. Read the data of DCS system in real time to check whether the conveyor material is in normal condition;

[0012] g. If it is in a normal state, no alarm will be sounded; if an abnormal signal is received, an alarm will be sounded immediately and the abnormal signal will be pushed to the equipment management personnel for remote intervention, and the abnormal information location will be processed in time.

[0013] In the above-mentioned method for detecting material breaks in a cement raw material conveyor belt, in the image acquisition module, a monitoring camera is used that is directly aimed at the conveyor and adjusted to capture the entire conveyor as much as possible. The video stream data of the monitoring camera is read in real time and decoded into single image data. Then, a rectangular ROI area is manually or automatically selected on the image data, and the coordinates of the rectangular ROI area are set to (x, y, w, h) relative to the image coordinates, where (x, y) represents the center coordinates of the rectangular area, and (w, h) represent the width and height of the rectangular area, respectively.

[0014] In the above-mentioned cement raw material conveyor belt break detection method, in the AI ​​classification module, a rectangular ROI area is cut out from the image and the cut ROI image is set as I i ∈R 2 , while I i Scaling and padding are performed to unify the image size to 224x224;

[0015] The YOLOv8 classification network consists of a backbone network and a prediction network;

[0016] The backbone network downsamples the 224x224 input image five times to convert it into three feature maps of different sizes: 7x7, 14x14, and 28x28;

[0017] The prediction network classifies the feature representation output by the backbone network and outputs the classification result;

[0018] Before AI analysis, we need to perform model training of the YOLOv8 network.

[0019] In the above-mentioned method for detecting material breakage in a cement raw material conveyor belt, in the conveyor state analysis module, when the conveyor starts to operate, it is necessary to satisfy the following conditions and pre-define parameters to determine which state the conveyor belongs to: C max is the maximum number of statistically valid frames in the initial running state of the conveyor, f breakless f is the effective frame number when the conveyor is initially in the state of having materials, break The effective frame number when the initial state of the conveyor is material breaking;

[0020] Determine the initial state of the conveyor: After obtaining the valid frame number C max If the initial state of the conveyor is that there is material, the effective frame number f breakless Greater than the effective frame number f when the initial state of the conveyor is material breaking break ,Right now:

[0021] f breakless >f break

[0022] If the conditions are met, it can be judged that the initial state of the conveyor is that there is material. If not, the initial state of the conveyor is that there is no material.

[0023] The above judgment is based on the initial state when the conveyor starts running;

[0024] For the conveyor to meet the continuous condition during operation, the following conditions must be met to determine whether a material break occurs. The parameters are predefined: f max is the maximum frame number threshold, f min is the minimum frame number threshold, u1 is the number of frames for continuous detection of material failure, and u2 is the number of frames for continuous detection of material presence;

[0025] When the conveyor changes from material supply to material shortage during operation, the material shortage detection frame number u1 is greater than the maximum frame number f max ,Right now:

[0026] u1>f max

[0027] When the above conditions are met, the conveyor status is judged as material shortage and a warning is issued;

[0028] When the conveyor changes from material-free to material-filled during operation, the material-filled detection frame number u2 is less than the minimum frame number f min ,Right now:

[0029] u2 <f min

[0030] When the above conditions are met, the conveyor status is judged as material shortage and a warning is issued.

[0031] In the above-mentioned method for detecting material breakage in a cement raw material conveyor belt, in the automatic alarm module, when the classification model determines that a material breakage has occurred in the conveyor, the algorithm judgment result can be used to issue an alarm through the early warning platform, and then the alarm screen and alarm details can be pushed to relevant equipment management personnel in the form of a client or PC.

[0032] In the above-mentioned method for detecting material breakage on a cement raw material conveyor belt, in the equipment linkage module, the DCS system interface is connected to the video monitoring system and the equipment control system. When the video monitoring system determines through classification results that the conveyor has material breakage, while pushing the alarm information to the equipment management personnel, the equipment abnormality signal can be transmitted to the DCS system through the DCS interface. The distributed control system can immediately perform automatic equipment control such as stopping the equipment operation according to the received equipment abnormality signal.

[0033] In the above-mentioned method for detecting material breakage in a cement raw material conveyor belt, a method of random cropping augmentation combined with jigsaw augmentation is used during the training process of YOLOv8.

[0034] Compared with the prior art, the present invention has the following advantages:

[0035] 1. The present invention proposes a complete set of conveyor material breakage identification methods. Through real-time intelligent analysis of monitoring cameras, the status of materials on the conveyor can be automatically monitored in real time throughout the day. Once the conveyor breaks, an alarm will be immediately issued to solve the problem in time. This reduces manual inspections, improves the quality of cement production, and ensures the continuous progress of cement production.

[0036] 2. The conveyor material breaking classification model designed by the present invention improves the robustness of the conveyor material breaking classification model by using the method of random cutting augmentation combined with jigsaw augmentation. The model of the present invention can obtain a clear, complete and accurate conveyor diagram, providing powerful and accurate data for the subsequent classification model. Through actual testing of conveyor material breaking, the accuracy rate can reach more than 99%.

[0037] 3. The present invention can more reliably complete the analysis and processing of abnormal signals by linking the video image classification algorithm results with the DCS system. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a module diagram of the overall structure of the present invention.

[0039] Figure 2 This is a structural diagram of the classification network YOLOv8 in the present invention.

[0040] Figure 3 It is a flow chart of the image acquisition module and AI classification module in the present invention.

[0041] Figure 4 Schematic diagram of the random cropping and augmentation method in the present invention.

[0042] Figure 5 Schematic diagram of the jigsaw puzzle augmentation method of the present invention. DETAILED DESCRIPTION

[0043] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0044] Reference Figure 1-5 As shown, a method for detecting material breakage of a cement raw material conveyor belt includes five modules that are performed in sequence, namely, an image acquisition module, an AI classification module, a conveyor state analysis module, an automatic alarm module, and an equipment linkage module. The following steps are performed through the above modules:

[0045] a. Start the system and load computer and hardware resources, as well as the classification model;

[0046] b. Obtain the image to be classified. First, decode the video stream transmitted by the surveillance camera into image data. Then, cut out the area included in the ROI from the image according to the preset rectangular ROI area.

[0047] c. Send the ROI image cropped in step b to the YOLOv8 classification network for AI analysis to obtain the conveyor status classification result;

[0048] d. According to the conveyor classification results obtained by reasoning, analyze and determine the state information of the conveyor when it is started and the conditions that are satisfied when the material is cut off during operation;

[0049] e. Determine whether the material status is normal when the conveyor is started based on the initial status analysis. If it is normal, return to step b to start a new round of status analysis; if it is abnormal, push the information to the equipment personnel for inspection through the equipment alarm;

[0050] f. Read the data of DCS system in real time to check whether the conveyor material is in normal condition;

[0051] g. If it is in a normal state, no alarm will be sounded; if an abnormal signal is received, an alarm will be sounded immediately and the abnormal signal will be pushed to the equipment management personnel for remote intervention, and the abnormal information location will be processed in time.

[0052] In the image acquisition module, a surveillance camera is used that is directly aimed at the conveyor and adjusted to capture the entire conveyor as much as possible. The video stream data of the surveillance camera is read in real time and decoded into single image data. Then, a rectangular ROI area is manually or automatically selected on the image data. The area needs to be able to completely contain the conveyor. The coordinates of the rectangular ROI area are set relative to the image coordinates as (x, y, w, h), where (x, y) represents the center coordinates of the rectangular area, and (w, h) represent the width and height of the rectangular area, respectively, to ensure that the ROI area can completely cover the conveyor.

[0053] In the AI ​​classification module, the rectangular ROI area is cropped from the image and the cropped ROI image is set as I i ∈R 2 , while I i Scaling and padding are performed to unify the image size to 224x224;

[0054] The YOLOv8 classification network consists of a backbone network and a prediction network;

[0055] The backbone network downsamples the 224x224 input image five times to convert it into three feature maps of different sizes: 7x7, 14x14, and 28x28;

[0056] The prediction network classifies the feature representation output by the backbone network and outputs the classification result.

[0057] Through this simple method, the network can achieve target classification tasks. Its main idea is to obtain feature information of different scales through the backbone network and perform the final classification through the prediction network. This method is simple and effective and is suitable for various target classification tasks.

[0058] Before AI analysis, we need to train the YOLOv8 network model. During the YOLOv8 training process, we do not use the original images obtained by directly inputting the surveillance camera. Instead, we use the random cropping augmentation combined with the jigsaw augmentation method. Through the above augmentation method, we can improve the robustness and anti-interference of the conveyor breaking model, while increasing the diversity of the data set.

[0059] After the classification results are obtained through training model prediction, there are two initial states of the conveyor: one is that there is material on the conveyor; the other is that there is no material on the conveyor. For the two states, when the conveyor starts to operate, the following conditions need to be met to determine which state the conveyor belongs to. The parameters are predefined: C max is the maximum number of statistically valid frames in the initial running state of the conveyor, f breakless f is the effective frame number when the conveyor is initially in the state of having materials, breakThe effective frame number when the initial state of the conveyor is material breaking;

[0060] Determine the initial state of the conveyor: After obtaining the valid frame number C max If the initial state of the conveyor is that there is material, the effective frame number f breakless Greater than the effective frame number f when the initial state of the conveyor is material breaking break ,Right now:

[0061] f breakless >f break

[0062] If the conditions are met, it can be judged that the initial state of the conveyor is that there is material. If not, the initial state of the conveyor is that there is no material.

[0063] The above judgment is based on the initial state when the conveyor starts running;

[0064] For the conveyor to meet the continuous condition during operation, the following conditions must be met to determine whether a material break occurs. The parameters are predefined: f max is the maximum frame number threshold, f min is the minimum frame number threshold, u1 is the number of frames for continuous detection of material failure, and u2 is the number of frames for continuous detection of material presence;

[0065] When the conveyor changes from material supply to material shortage during operation, the material shortage detection frame number u1 is greater than the maximum frame number f max ,Right now:

[0066] u1>f max

[0067] When the above conditions are met, the conveyor status is judged as material shortage and a warning is issued;

[0068] When the conveyor changes from material-free to material-filled during operation, the material-filled detection frame number u2 is less than the minimum frame number f min ,Right now:

[0069] u2 <f min

[0070] When the above conditions are met, the conveyor status is judged as material shortage and a warning is issued.

[0071] When the classification model determines that the conveyor is broken, the algorithm judgment result can be used to issue an alarm through the early warning platform, and then the alarm screen and alarm details can be pushed to the relevant equipment managers in the form of a client or PC. The managers can check the push information and the alarm location on site, make corresponding adjustments to the equipment according to the alarm information, and ensure that the equipment managers can understand the specific situation of the alarm conveyor at the first time and anytime and anywhere.

[0072] In the face of the timeliness issue from alarm to processing of conveyor material shortage, timely and effective control can be achieved by linking the classification results with the conveyor equipment. As an important part of the construction of intelligent factories, the DCS system realizes real-time monitoring and decentralized remote control of equipment parameters in the factory, so the DCS system interface can be connected to the video monitoring system and equipment control system.

[0073] When the video monitoring system determines that the conveyor is broken through the classification results, the alarm information can be pushed to the equipment management personnel, and the abnormal equipment signal can be transmitted to the DCS system through the DCS interface. The distributed control system can immediately stop the equipment operation and other automatic equipment control according to the received abnormal equipment signal. The management personnel can also perform manual intervention through the DCS system, which reduces the delay time from the AI ​​determining the existence of broken material information and automatically alarming to the equipment management personnel to deal with it on site, and improves production efficiency.

[0074] Example

[0075] Step 1:

[0076] In the image acquisition module, in addition to acquiring video stream data in real time, the training phase also requires collecting a large number of conveyor pictures from surveillance cameras and classifying them. Specifically, it is necessary to distinguish between two categories of pictures: pictures with material and pictures of broken material, i.e., pictures of conveyors with material and pictures of conveyors with broken material. In this embodiment, 9396 pictures of broken material and 9475 original pictures of material are used for training. The categories in this embodiment are divided into two categories: "with material" and "broken material".

[0077] Step 2:

[0078] Before AI analysis, we need to train the model. In this process, we adopted the original training method of YOLOv8 and added the method of online random cropping ROI (region of interest) combined with jigsaw augmentation. In this way, we can make better use of data resources and improve the generalization ability and accuracy of the model.

[0079] As attached Figure 4 As shown, the steps of random cropping method are as follows:

[0080] (1) It is known that a ROI is selected in a monitoring image, namely, the conveyor area;

[0081] (2) Perform translation transformation on the selected ROI area so that the center point of the conveyor is randomly offset by several pixels in the horizontal and vertical directions to obtain a new ROI area. In this way, α different ROI areas can be obtained, and their positions are different;

[0082] (3) Scaling the ROI region in the original image. According to the different ROI regions obtained in step (2), scaling is performed on each new ROI region. After scaling, β ROI regions of different sizes can be obtained. By changing the size of the ROI region, the diversity of the data is increased.

[0083] (4) Combining the above two steps, the original ROI area is transformed (α·β) times, thereby obtaining (α·β) ROIs with different sizes and positions from the initial ROI area;

[0084] (5) After random cropping and augmentation, the position of the conveyor in the ROI area is not necessarily in the center of the image, and some conveyor information may be lost. At this time, it is necessary to constrain the size of the ROI area to be larger than the size of the conveyor to ensure that most of the conveyor information can be included.

[0085] The detection of the conveyor belt status is affected by the quality of the image. The slight differences between the conveyor belts affect the detection results, which are all limited by the learning ability of the CNN network. Further processing of the acquired images can improve the effective learning of the CNN grid on the original images.

[0086] Using the conveyor belt puzzle as input data for training can enable the grid model to learn more and richer conveyor belt status information. The image set D obtained by random cropping and augmentation k , for each randomly cropped and augmented image ω∈D k 3 ×W×H , after jigsaw augmentation, such as the attached Figure 5 As shown, the training data set for the conveyor classification model is obtained, and the steps of the jigsaw augmentation method are as follows:

[0087] (1) Divide ω into n×n small blocks, each of which is

[0088] (2) Randomly select some small blocks from the n×n small blocks, shuffle them, and move them to new positions;

[0089] (3) Repeat step (2) until all small blocks are rearranged and spliced ​​into a new training image P(ω,n);

[0090] In order to prevent the grid model from focusing too much on local information, the puzzle piece size is gradually increased during the training process. This allows the grid model to first learn smaller local fine-grained information, then learn coarse-grained information, and finally learn the global information of the entire image.

[0091] In this embodiment, the number of translation transformations α=4, the number of scale scaling β=4, and the number of randomly augmented pictures obtained is α·β=4·4=16, that is, through a cropped ROI area, D k = 16 ROI areas of different sizes and positions.

[0092] The jigsaw puzzler generates conveyor belts of information of different scales. According to the scaled image size W = H = 224, as the jigsaw puzzler input data, each image uses n = 2, 4, 8, and can generate three conveyor belt puzzles of different sizes. The image is divided into 8×8 puzzle pieces (a total of 64 puzzle pieces), and these 64 puzzle pieces are randomly decomposed, reconstructed into puzzle pieces and sent to the training network for feature extraction.

[0093] As the training network progresses, the length and width of the puzzle pieces are doubled, and the image is divided into 4×4 puzzle pieces for further training. Similarly, the image is further divided into 2×2 puzzle pieces for training. Finally, the puzzle piece size is expanded to the puzzler input image size and sent to the network model training.

[0094] After using data augmentation on a cropped ROI area, we obtain s=16×4=64 images.

[0095] Through the above-mentioned augmentation method, the diversity of conveyor data is greatly improved, and the robustness and anti-interference ability of the conveyor breaking model are improved.

[0096] Step 3:

[0097] After completing the model training, we started the actual calculation work of the model. In order to obtain accurate data, we used a surveillance camera that was facing the conveyor and was adjusted to be able to capture the entire conveyor.

[0098] By reading the video stream data of the surveillance camera in real time and decoding the video stream into single picture data every 1 second, the real-time image of the conveyor is obtained. In order to focus only on the conveyor part, we need to manually or automatically select a rectangular area (ROI) on the picture, which should be able to completely contain the conveyor. In this embodiment, we select 5 adjacent points at the beginning and the end to define the ROI area. In this way, we can ensure that we focus on the conveyor part instead of other background information.

[0099] [(70,240),(500,240),(500,360),(70,360),(70,240)]

[0100] The coordinates of the rectangular ROI area relative to the image coordinates are (x, y, w, h), where (x, y) represent the center coordinates of the rectangular area, and (w, h) represent the width and height of the rectangular area.

[0101]

[0102]

[0103] Step 4:

[0104] According to the previously obtained image data and rectangular ROI area coordinates, we cropped the ROI image. Then the ROI image is scaled and filled to make its size uniform to the target image size of 224x224. In this embodiment, we use the calculated ROI coordinates to crop the ROI image, and finally obtain a target image of size 215x60.

[0105] The fixed-size input data after scaling and padding is sent to the YOLOv8 network. In this embodiment, we adopt the method of enlarging first and then padding. First, the long side is enlarged to 224 in proportion, and then the short side is filled with fixed pixels (114, 114, 114) up and down, and finally a fixed-size 224x224 image is obtained.

[0106] (1) In the YOLOv8 network, we first process the input data using the backbone network to obtain three conveyor feature maps of different sizes: 7x7, 14x14, and 28x28. These feature maps can be used to predict the results. Figure 2 The backbone part is implemented in .

[0107] (2) Finally, we use the prediction network to perform classification prediction on each fused feature map. The prediction network consists of the attached Figure 2 For each target category and confidence level, the prediction network will give the corresponding prediction result.

[0108] Step 5:

[0109] Determine the initial state of the conveyor: After obtaining the valid frame number C max If the initial state of the conveyor is that there is material, the effective frame number f breakless Greater than the effective frame number f when the initial state of the conveyor is material breaking break ,Right now:

[0110] f breakless >f break

[0111] If the conditions are met, it can be judged that the initial state of the conveyor is that there is material. If not, the initial state of the conveyor is that there is no material.

[0112] The above judgment is based on the initial state when the conveyor starts running;

[0113] For the conveyor to meet the continuous condition during operation, the following conditions must be met to determine whether a material break occurs. Predefined parameters: f max is the maximum frame number threshold, and u is the number of continuous detection frames.

[0114] When the conveyor changes from material supply to material shortage during operation, the material shortage detection frame number u1 is greater than the maximum frame number f max ,Right now:

[0115] u1>f max

[0116] When the conditions are met, the conveyor status is judged as material shortage and a warning is issued;

[0117] When the conveyor changes from material-free to material-filled during operation, the material-filled detection frame number u2 is less than the minimum frame number f min ,Right now:

[0118] u2 <f min

[0119] When the above conditions are met, the conveyor status is judged as material shortage and a warning is issued.

[0120] For the above parameter settings, the effective frame number C max =5, maximum frame number f max =10, minimum frame number f min =5.

[0121] Step 6:

[0122] The automatic alarm module uses the classification model to determine the material shortage of the conveyor, and can issue a real-time alarm through the early warning platform, and push the alarm screen and details to the relevant equipment managers. The managers can handle the alarm in time according to the pushed information to ensure the normal operation of the equipment.

[0123] Step 7:

[0124] The AI ​​classification model is used to determine whether the conveyor is broken and generate corresponding alarm information; the alarm information is pushed to the equipment management personnel, and the abnormal signal is transmitted to the DCS system through the DCS system interface; after the DCS system receives the abnormal signal, it immediately stops the operation of the equipment and takes other automatic control measures to avoid equipment damage and production stagnation;

[0125] When the DCS system obtains that the conveyor is in working condition, if the above analysis shows that the conveyor is in normal condition, no alarm will be given; if the above analysis shows that the conveyor is moving abnormally, an alarm will be given in time to notify the equipment manager to check. The equipment manager can manually intervene through the DCS system to further adjust and maintain the equipment.

[0126] In this embodiment, two methods are used to issue an alarm: one is to display the alarm information on the screen through a large screen, and the other is to obtain the alarm information through a message receiving method.

[0127] It is known from common technical knowledge that the present invention can be implemented by other embodiments that do not deviate from its spirit or essential features. Therefore, the above disclosed embodiments are only illustrative in all respects and are not exclusive. All changes within the scope of the present invention or within the scope equivalent to the present invention are included in the present invention.

Claims

1. A method for detecting material breakage in a cement raw material conveyor belt, characterized in that: It includes 5 modules, namely image acquisition module, AI classification module, conveyor status analysis module, automatic alarm module and equipment linkage module. The following steps are performed through the above modules: a. Start the system and load computer and hardware resources, as well as the classification model; b. Obtain the image to be classified. First, decode the video stream transmitted by the surveillance camera into image data. Then, cut out the area included in the ROI from the image according to the preset rectangular ROI area. c. Send the ROI image cropped in step b to the YOLOv8 classification network for AI analysis to obtain the conveyor status classification result; d. According to the conveyor classification results obtained by reasoning, analyze and determine the state information of the conveyor when it is started and the conditions that are satisfied when the material is cut off during operation; e. Determine whether the material status is normal when the conveyor is started based on the initial status analysis. If it is normal, return to step b to start a new round of status analysis; if it is abnormal, push the information to the equipment personnel for inspection through the equipment alarm; f. Read the data of DCS system in real time to check whether the conveyor material is in normal condition; g. If it is in a normal state, no alarm will be sounded; if an abnormal signal is received, an alarm will be sounded immediately and the abnormal signal will be pushed to the equipment management personnel for remote intervention, and the abnormal information will be handled in time; In the image acquisition module, a surveillance camera is used to aim directly at the conveyor and adjust its position to capture the entire conveyor as much as possible. The video stream data of the surveillance camera is read in real time and decoded into a single picture data. Then, a rectangular ROI area is manually or automatically selected on the picture data, and the coordinates of the rectangular ROI area are set relative to the image coordinates as ( x, y, z, h ),in( x,y ) represents the center coordinates of the rectangular area, ( w,h ) represent the width and height of the rectangular area respectively; In the AI ​​classification module, a rectangular ROI area is cropped from the image and the cropped ROI image is set as I i ∈R 2 , while I i Scaling and padding are performed to unify the image size to 224x224; The YOLOv8 classification network consists of a backbone network and a prediction network; The backbone network downsamples the 224x224 input image five times to convert it into three feature maps of different sizes: 7x7, 14x14, and 28x28; The prediction network classifies the feature representation output by the backbone network and outputs the classification result; Before AI analysis, we need to train the YOLOv8 network model; In the conveyor state analysis module, when the conveyor starts to operate, the following conditions need to be met to determine which state the conveyor belongs to, and the parameters are predefined: C max The maximum number of valid frames in the initial running state of the conveyor. f breakless It is the effective frame number when the conveyor is initially in the state of having materials. f break The effective frame number when the initial state of the conveyor is material breaking; Determine the initial state of the conveyor: After obtaining the valid frame number C max If the initial state of the conveyor is that there is material, the effective frame number is f breakless Greater than the effective frame number when the conveyor is initially in the broken state f break ,Right now: f breakless ＞ f break If the conditions are met, it can be judged that the initial state of the conveyor is that there is material. If not, the initial state of the conveyor is that there is no material. The above judgment is based on the initial state when the conveyor starts running; For the conveyor to meet the continuous condition during operation, the following conditions must be met to determine whether a material break occurs, and the parameters are predefined: f max is the maximum frame number threshold, f min is the minimum frame number threshold, u1 is the number of frames for continuous detection of material failure, and u2 is the number of frames for continuous detection of material presence; When the conveyor changes from material supply to material shortage during operation, the material shortage detection frame number u1 is greater than the maximum frame number f max ,Right now: u1＞ f max When the above conditions are met, the conveyor status is judged as material shortage and a warning is issued; When the conveyor changes from material-free to material-filled during operation, the material-filled detection frame number u2 is less than the minimum frame number f min ,Right now: u2＜ f min When the above conditions are met, the conveyor status is judged as material shortage and a warning is issued.

2. A method for detecting material breakage in a cement raw material conveyor belt according to claim 1, characterized in that: In the automatic alarm module, when the classification model determines that the conveyor is out of material, the algorithm judgment result can be used to issue an alarm through the early warning platform, and then the alarm screen and alarm details can be pushed to the relevant equipment management personnel in the form of a client or PC.

3. The method for detecting material breakage in a cement raw material conveyor belt according to claim 1, characterized in that: In the equipment linkage module, the DCS system interface is connected to the video monitoring system and the equipment control system. When the video monitoring system determines through classification results that the conveyor is out of material, while pushing the alarm information to the equipment management personnel, the equipment abnormality signal can be transmitted to the DCS system through the DCS interface. The distributed control system can immediately stop the equipment operation and other automatic equipment controls according to the received equipment abnormality signal.

4. A method for detecting material breakage in a cement raw material conveyor belt according to claim 1, characterized in that: During the training process of YOLOv8, a random cropping augmentation method combined with a jigsaw augmentation method was used.

Images