Chute angle abnormality diagnosis method and system based on deep time series image learning
Through the depth timing image learning technology, the blast furnace chute angle is judged in real time, which solves the problem of checking the blast furnace during rest in the existing technology, and realizes the chute angle verification during operation, improving convenience and timeliness.
Patent Information
- Application Number
- CN202311375875.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-10-23
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2043-10-23
AI Technical Summary
The existing blast furnace chute angle calibration method needs to be carried out during blast furnace rest maintenance, resulting in insufficient timeliness and convenience, affecting the operation of the blast furnace.
The chute angle abnormality diagnosis method based on deep timing image learning is adopted. By collecting the top monitoring video data of blast furnace furnace, preprocessing and model training, the LSTM network is used to extract spatiotemporal features, and whether the chute angle is abnormal in real time is determined.
Real-time verification of the chute angle during chute operation is achieved, reducing verification difficulty, saving costs, improving convenience and timeliness, and reducing potential hidden dangers.
Smart Images

Figure CN117475203B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of blast furnace smelting technology and image processing technology, and in particular to a method and system for diagnosing abnormal chute angles based on deep temporal image learning. Background Art
[0002] Blast furnace charging is the core component of blast furnace smelting process, and charging chute is the key component of blast furnace charging system. In order to make the raw materials in the blast furnace fully contact with the gas, the top charging system needs to accurately adjust the charging angle. In blast furnace ironmaking production, due to equipment wear and data signal drift, the actual angle of the chute may deviate from the displayed angle. Among them, the chute α angle (such as Figure 1 The control of the angle of the chute is an important part of the charging system. In addition, since the chute of the blast furnace is in a closed operating environment with high temperature, high pressure and high dust, and the frequent impact of the falling charge on the chute, the actual execution angle of the blast furnace chute will be inconsistent with the theoretical setting angle. In order to avoid the chute angle "deviation", the chute angle needs to be checked.
[0003] At present, the calibration technologies for blast furnace chute angles are mainly divided into three types: measurement inside the furnace when the blast furnace is shut down, contact measurement of mechanical devices outside the furnace, and non-contact laser measurement outside the furnace.
[0004] The most direct and traditional method of measuring the angle of a chute is to build an inclined bridge between the inspection door and the chute with steel plates. The measurement personnel enter the furnace through the inclined bridge and use a protractor to measure the angle of the chute. This method requires opening a large square hole on the furnace roof for inspection and building a measurement operation platform. The measurement is highly disturbing and there are also operational safety risks.
[0005] The contact measurement technology of mechanical devices outside the furnace uses the manhole of the blast furnace to place the measuring device close to the back of the chute to measure the angle of the chute, avoiding the shortcomings of opening the large square hole on the furnace roof for maintenance and setting up a measurement operation platform. The measurement process is safe and simple. However, the diameter of the manhole is limited, and the above devices are mostly difficult to insert, and the operation is complicated. In addition, in order to ensure the measurement accuracy, it is necessary to ensure that the measuring device is close to the back of the chute. However, in actual applications, it is often difficult to observe whether the above requirements are met, and the measurement accuracy and reliability are not high.
[0006] The non-contact laser measurement technology outside the furnace uses a laser rangefinder to measure the angle between the laser and the vertical support through the blast furnace manhole to obtain the chute angle, or to sample points on the blast furnace chute through the laser to indirectly calculate the chute angle. However, this technology is complex to operate and has high costs. In addition, the above method also requires opening the inspection door or observation hole, and the angle can only be checked during the blast furnace maintenance.
[0007] In order to avoid abnormal angles of blast furnace chutes, relevant studies have proposed chute angle calibration technologies based on contact measurement of mechanical devices and non-contact measurement based on lasers. However, the above methods all require diagnosis of chute "deviation" when the blast furnace is shut down for maintenance. Since blast furnaces are generally shut down 2-3 times per year, when the chute is found to be "deviant", it has already caused adverse effects on the operation of the blast furnace. Therefore, the current research methods are insufficient in timeliness and convenience, and it is urgent to design a method that can determine whether the chute angle is abnormal in real time. Summary of the invention
[0008] To this end, an embodiment of the present invention provides a method and system for diagnosing abnormal chute angles based on deep time series image learning, which is used to solve the problem that the existing chute angle calibration methods in the prior art all require the blast furnace chute angle calibration to be performed when the blast furnace is shut down for maintenance. Since the blast furnace is generally shut down 2-3 times / year, when an abnormal chute angle is found, it has already caused adverse effects on the blast furnace operation. Therefore, the current research method has deficiencies in timeliness, convenience and other aspects.
[0009] In order to solve the above problems, an embodiment of the present invention provides a method for diagnosing abnormal chute angle based on deep time series image learning, the method comprising:
[0010] Step S1: Collect blast furnace top monitoring video data to obtain blast furnace chute videos under different angles and different environments;
[0011] Step S2: preprocessing the blast furnace chute video to obtain a chute angle video data set;
[0012] Step S3: inputting the chute angle video dataset into a pre-built chute angle recognition model based on deep time series image learning for model training, thereby obtaining a trained chute angle recognition model based on deep time series image learning;
[0013] Step S4: inputting the real-time collected chute image into the trained chute angle recognition model based on deep time series image learning to perform prediction and obtain the final classification result;
[0014] Step S5: Compare the final classification result with the set chute angle to determine whether the chute angle is abnormal.
[0015] Preferably, in step S2, the method for preprocessing the blast furnace chute video comprises:
[0016] The blast furnace chute video is sampled in a time series window. In the blast furnace chute video, the number of frames for each angle of the chute to rotate one circle is N, and the number of sampled image window frames is M. By taking an image frame every N / M and adding it to the window, one video takes N / M windows, which can retain the video features to the greatest extent.
[0017] Preferably, in step S3, the chute angle video dataset is input into a pre-built chute angle recognition model based on deep time series image learning for model training, and the method for obtaining the trained chute angle recognition model based on deep time series image learning specifically includes:
[0018] S31: dividing the chute angle video data set in proportion to obtain a training sample set and a test sample set;
[0019] S32: inputting the training sample set into a pre-built chute angle recognition model based on deep time series image learning to perform model training;
[0020] S33: Use the cross entropy function to calculate the loss of the predicted value and the true value, save the model weight when the loss function converges, and obtain the trained chute angle recognition model based on deep time series image learning.
[0021] Preferably, in step S32, the method of inputting the training sample set into a pre-built chute angle recognition model based on deep time series image learning for model training specifically includes:
[0022] Firstly, the LSTM network model is used to extract the spatiotemporal feature information of the chute image. The last LSTM unit is selected to output the feature vector. Then, the fully connected layer is used to map the extracted feature vector to the category of the chute angle. Finally, the Softmax function is used to normalize the score of each category to obtain the probability of each category, and the category with the maximum probability is output as the predicted value of the chute angle.
[0023] Preferably, the LSTM network model introduces three gating mechanisms, including an input gate, a forget gate and an output gate, to control the input, update and output of information; wherein the input gate is used to control the input of new information, the forget gate is used to control the retention of historical information, and the output gate is used to control the generation of output, and the three gating mechanisms together constitute a hidden unit of the LSTM network model.
[0024] Preferably, in step S4, the chute image collected in real time is input into a trained chute angle recognition model based on deep time series image learning for prediction, and the method for obtaining the final classification result specifically includes:
[0025] The real-time collected chute images are input into the trained chute angle recognition model based on deep time series image learning for prediction. Since the video of the chute rotating one circle is divided into multiple time series windows, when multiple different category results appear in one circle, the category result with the largest output ratio is used as the final classification result of the chute angle in that circle according to the ratio of category results.
[0026] When multiple different category results appear in one circle, according to the category result ratio, the category result with the largest output ratio is used as the final classification result of the chute angle of the circle, specifically including:
[0027] Assume that N represents the total number of time series windows for one rotation of the chute, c i represents the diagnosis result of the i-th time series window, then the number of times each category appears in the whole cycle count(c) is:
[0028]
[0029] Among them, [c i =c] is the indicator function, c is the actual angle of the chute, when c i The value is 1 when it is equal to c, otherwise it is 0;
[0030] Next, calculate the proportion p(c) of each category in the entire circle as:
[0031]
[0032] Finally, output the final result of the circle, that is, the category with the highest proportion for:
[0033]
[0034] Among them, argmax is a function that finds parameters for a function and is used to output the category with the largest number of occurrences.
[0035] The embodiment of the present invention further provides a chute angle abnormality diagnosis system based on deep time series image learning, which is used to implement the above-mentioned chute angle abnormality diagnosis method based on deep time series image learning, and the system includes:
[0036] The chute data acquisition module is used to collect blast furnace top monitoring video data and obtain blast furnace chute videos under different angles and environments;
[0037] A data preprocessing module, used for preprocessing the blast furnace chute video to obtain a chute angle video data set;
[0038] A model training module is used to input the chute angle video data set into a pre-built chute angle recognition model based on deep time series image learning for model training, so as to obtain a trained chute angle recognition model based on deep time series image learning;
[0039] The prediction and classification module is used to input the real-time collected chute images into the trained chute angle recognition model based on deep time series image learning to make predictions and obtain the final classification results;
[0040] The angle abnormality diagnosis module is used to compare the final classification result with the set chute angle to determine whether the chute angle is abnormal.
[0041] An embodiment of the present invention also provides an electronic device, which includes a processor, a memory and a bus system, wherein the processor and the memory are connected through the bus system, the memory is used to store instructions, and the processor is used to execute the instructions stored in the memory to implement the above-mentioned chute angle abnormality diagnosis method based on deep time series image learning.
[0042] An embodiment of the present invention also provides a computer storage medium, which stores a computer software product. The computer software product includes a number of instructions for enabling a computer device to execute the above-mentioned chute angle abnormality diagnosis method based on deep time series image learning.
[0043] It can be seen from the above technical solutions that the present invention has the following advantages:
[0044] The embodiment of the present invention provides a method and system for diagnosing abnormal chute angles based on deep time series image learning. By collecting thermal imaging videos of the blast furnace chute rotating at different angles and analyzing video features, the method of the present invention has strong real-time performance. Compared with the previous mechanical measurement method, it can verify the chute angle during the operation of the chute, greatly reducing the difficulty of chute angle verification, saving costs, reducing potential hidden dangers, and realizing convenient and efficient chute angle verification. BRIEF DESCRIPTION OF THE DRAWINGS
[0045] In order to more clearly illustrate the implementation cases of the present invention or the technical solutions in the prior art, the following is a brief description of the drawings required for use in the embodiments. By referring to the drawings, the features and advantages of the present invention will be more clearly understood. The drawings are schematic and should not be understood as limiting the present invention in any way. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:
[0046] Figure 1 It is a schematic diagram of the blast furnace chute angle in the background technology;
[0047] Figure 2 It is a flowchart of a method for diagnosing abnormal chute angle based on deep time series image learning provided in an embodiment;
[0048] Figure 3 It is a schematic diagram of the technical solution framework for the chute angle identification problem in the embodiment;
[0049] Figure 4 It is a structural diagram of the LSTM unit in the embodiment;
[0050] Figure 5 is a block diagram of a chute angle abnormality diagnosis system based on deep temporal image learning provided in an embodiment;
[0051] Figure 6 It is a schematic diagram of the interface of the chute angle abnormality diagnosis system based on deep temporal image learning in the embodiment. DETAILED DESCRIPTION
[0052] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0053] Embodiment 1
[0054] The existing chute angle calibration methods all require the blast furnace chute angle calibration to be performed when the blast furnace is shut down for maintenance. Since the blast furnace is generally shut down 2-3 times per year, when the chute angle is found to be abnormal, it has already had a bad impact on the operation of the blast furnace. Therefore, the current research methods are insufficient in terms of timeliness and convenience. Figure 2 As shown, an embodiment of the present invention proposes a chute angle abnormality diagnosis method based on deep time series image learning, which is used to determine in real time whether the chute angle is abnormal. The method includes:
[0055] Step S1: Collect blast furnace top monitoring video data to obtain blast furnace chute videos under different angles and different environments;
[0056] Step S2: preprocessing the blast furnace chute video to obtain a chute angle video data set;
[0057] Step S3: inputting the chute angle video dataset into a pre-built chute angle recognition model based on deep time series image learning for model training, thereby obtaining a trained chute angle recognition model based on deep time series image learning;
[0058] Step S4: inputting the real-time collected chute image into the trained chute angle recognition model based on deep time series image learning to perform prediction and obtain the final classification result;
[0059] Step S5: Compare the final classification result with the set chute angle to determine whether the chute angle is abnormal.
[0060] From the above technical solution, it can be seen that the present invention is a method for diagnosing abnormal chute angles based on deep time series image learning. By collecting thermal imaging videos of the blast furnace chute rotating at different angles and analyzing video features, the method of the present invention has strong real-time performance. Compared with the previous mechanical measurement method, it can verify the chute angle during the operation of the chute, which greatly reduces the difficulty of chute angle verification, saves costs, reduces potential hidden dangers, and realizes convenient and efficient chute angle verification.
[0061] From the chute movement video collected by the furnace top imaging system, it can be found that the size of the chute in the video increases with the increase of the chute angle, which has obvious time-series image characteristics. Based on the above analysis, we classify the chute angle recognition problem as a classification problem. Through image collection, data processing, model training and online recognition, we can achieve real-time and convenient detection of the chute angle. The complete technical solution framework is as follows Figure 3 shown.
[0062] In this embodiment, in step S1, by collecting the blast furnace top monitoring video data, the blast furnace chute video under different angles and different environments can be obtained. Since each angle rotation contains about 100 frames of images, if these images are input into the model at one time, the computer resources required are extremely large. In addition, for a certain frame image in the chute video, the changes in the images of the nearby frames are small, and inputting the data at one time will cause data redundancy. Therefore, in the chute angle diagnosis task, the timing window is sampled first.
[0063] In this embodiment, in step S2, the blast furnace chute video is preprocessed to obtain a chute angle video data set. Specifically, in the blast furnace chute video, the number of frames for each angle of the chute to rotate one circle is N, and the number of sampled image window frames is M. The sampling methods of the image window can be divided into two types. The first method is to select M consecutive frames in the video and combine them into a window, so one video can take N / M windows; the second method is to take a frame of image every N / M and add it to the window, and similarly one video can take N / M windows. Through the visual feature analysis of the blast furnace chute angle, the first method of continuous frame taking often loses part of the spatiotemporal information, thereby affecting the judgment; while the second method of interval frame taking can retain the video features to the maximum extent and provide more information. Therefore, this embodiment adopts the second method, and interval frames are added to the image window.
[0064] In this embodiment, in step S3, the chute angle video dataset is input into a pre-built chute angle recognition model based on deep temporal image learning for model training, and a trained chute angle recognition model based on deep temporal image learning is obtained, which specifically includes:
[0065] S31: dividing the chute angle video data set in proportion to obtain a training sample set and a test sample set;
[0066] S32: Input the training sample set into a pre-built chute angle recognition model based on deep time series image learning to perform model training, specifically including:
[0067] Firstly, the LSTM network model is used to extract the spatiotemporal feature information of the chute image, and the last LSTM unit is selected to output the feature vector. Then, the fully connected layer is used to map the extracted feature vector to the category of the chute angle. Finally, the Softmax function is used to normalize the score of each category to obtain the probability of each category, and the category with the maximum probability is output as the predicted value of the chute angle.
[0068] S33: Use the cross entropy function to calculate the loss of the predicted value and the true value, save the model weight when the loss function converges, and obtain the trained chute angle recognition model based on deep time series image learning.
[0069] This embodiment uses a long short-term memory network (Long Short-Term Memory, LSTM) which is a special recurrent neural network that solves the gradient vanishing problem of the recurrent neural network (Recurrent Neural Network, RNN) when processing long sequence data.
[0070] The LSTM network model introduces three gating mechanisms, including input gate, forget gate and output gate, to control the input, update and output of information; the input gate is used to control the input of new information, the forget gate is used to control the retention of historical information, and the output gate is used to control the generation of output. The three gating mechanisms together constitute a hidden unit of the LSTM network model. Figure 4 As shown, the structure of the LSTM unit is as follows:
[0071] f t =σ(W f ·[h t-1 ,x t ]+b f )
[0072] i t =σ(W i ·[h t-1 ,x t ]+b i )
[0073]
[0074]
[0075] Ot =σ(W O ·[h t-1 ,x t ]+b O )
[0076] h t =O t *tanh(C t )
[0077] Among them, f t ,i t , O t All are internal variables of LSTM units; W f , W i , W c , W O is a learnable weight parameter; b f , b i , b c , b o is a learnable bias parameter; x t is the chute image input at time t; h t is the hidden state at time t; [h t-1 ,x t ] is h t-1 With x t The merging operation; σ is the sigmoid nonlinear unit, σ(x)=(1+e -x ) -1 , tanh is the hyperbolic tangent function, and each LSTM unit outputs a hidden variable h t , and C t Enter into the next unit.
[0078] Specifically, the chute image (specification is W*H) is input into the LSTM network model to extract the time series features of the chute sequence image. Since each group of chute time series windows contains N images, the input size of the LSTM network model is (N, W*H). Taking N=16 as an example, it means that each sample contains 16 time series images, and the LSTM network model will extract features from the above time series images. In terms of network structure, the present invention selects a classic LSTM network structure with a hidden layer size of 256 and a unit layer number of 2. The last LSTM unit is selected as the output vector, that is, the output vector size is 256. Further, the vector is input into the fully connected network, and the fully connected layer output vector is the category of the chute angle. If the chute angle is from 20°-40°, the output of the fully connected network is 21 categories, each category corresponds to an angle, and the network output value represents the score of each category. Finally, the score of each category is normalized by the Softmax function to obtain the probability of each category, and the category with the maximum probability is output as the predicted value of the chute angle. In this embodiment, we use the LSTM network model, but the method of the present invention is also applicable to other methods that can perform time series image classification.
[0079] In order to train the chute angle recognition model based on deep time series image learning, it is necessary to define a loss function and an optimizer. Considering that the chute calibration problem of the present invention can be classified as a classification problem, the loss of the predicted value and the true value is calculated by the cross entropy function. In addition, the commonly used reverse gradient propagation method (such as the Adam function) can be used to train the model.
[0080] In this embodiment, in step S4, the chute image collected in real time is input into a trained chute angle recognition model based on deep time series image learning for prediction to obtain a final classification result.
[0081] Specifically, after the model training is completed, the chute image can be input into the model to realize the chute angle diagnosis. Considering that the video of the chute rotating in one circle can be divided into multiple time series windows, when multiple different category results appear in one circle, the one with the largest proportion will be output as the final classification result of the chute angle in that circle according to the proportion of category results. This shows that even if some time series windows have output errors in one circle, as long as the error rate is under a certain proportion, it will not affect the final chute angle diagnosis result. The specific calculation process is as follows:
[0082] Assume that N represents the total number of time series windows for one rotation of the chute, c i represents the diagnosis result of the i-th time series window, then the number of times each category appears in the whole cycle count(c) is:
[0083]
[0084] Among them, [c i=c] is the indicator function, c is the actual angle of the chute, and takes the value of 1 when cx is equal to c, otherwise it takes the value of 0;
[0085] Next, calculate the proportion p(c) of each category in the entire circle as:
[0086]
[0087] Finally, output the final result of the circle, that is, the category with the highest proportion for:
[0088]
[0089] Among them, argmax is a function that finds parameters for a function and is used to output the category with the largest number of occurrences.
[0090] In this embodiment, in step S5, the final classification result is compared with the set chute angle to determine whether the chute angle is abnormal.
[0091] For example, for a video of a chute with a chute angle of 22° moving in one circle, after window processing, it contains 12 window samples. The above samples are input into the model, and 4 samples are output as 22 degrees, and 8 samples are output as 23 degrees. According to the above formula, count(22)=4, count(23)=8, p(22)=0.333, p(23)=0.667, That is, the final judgment is that the chute angle is 23 degrees. By comparing the chute angle with the chute angle of the PLC, it is possible to finally check whether the chute angle is abnormal.
[0092] In order to further illustrate the advantages of the present invention, a description is given below in conjunction with specific experiments.
[0093] 1. Data preparation and preprocessing
[0094] After collecting the blast furnace chute movement video through the blast furnace top monitoring equipment, the video is preliminarily split into long videos into short videos, each of which corresponds to one degree of rotation of the chute. The number of video angles collected is shown in Table 1 below.
[0095] Table 1
[0096]
[0097]
[0098] From Table 1 above, we can see that the number of circles for chute angles of 20°, 25°, 27°, 28°, 29°, 30°, and 31° is 18 each, and there are 16 circles for other angles, with a total of 350 videos collected. First, these videos are divided according to the ratio of training set: test set = 0.85:0.15, resulting in 298 training set videos and 52 test set videos. In addition, the chute video is split into image frames, generating a total of 2045 groups of samples. The training set data size is 1758, and similarly, the test set data size is 287. Sample examples are shown in Table 2 below.
[0099] Table 2
[0100]
[0101] 2. Experimental results and analysis
[0102] To verify the recognition effect of the model, we collected videos of 15 chute angles at different times and performed the same preprocessing operation. The test results are shown in Table 3 below, where the diagnosis rate (by window) is the ratio of the number of correctly predicted time series windows to the number of all windows. Take 23° in Table 3 as an example. The test data contains a video of the chute angle running 5 times in a production environment. The video is split into 30 time series image windows, of which 28 are diagnosed as 23° and 2 are diagnosed incorrectly, so the diagnosis rate is 28 / 30=0.933. Table 3 shows that 11 of the 15 angles are correctly predicted, while the other angles show small errors. The overall accuracy of this batch of data is 97.5%.
[0103] Since one circle of the chute movement can be divided into multiple time series windows, the angle prediction result of this circle is determined by the prediction results of multiple windows, rather than the results of a single window. Therefore, the diagnostic rate needs to be further calculated in order to evaluate the performance of the model in engineering applications. Taking 23° as an example, the blast furnace chute rotates 5 circles and records 23°. Each circle can be divided into 6 time series windows, of which 1 window prediction error is in the first circle and 1 window prediction error in the second circle. Substituting into the above category calculation formula, the predicted angle is still 23°, indicating that the prediction result is correct. Similarly, we can calculate the diagnostic rate of other angles. Since the angle of each circle is predicted correctly, the diagnostic rate of this batch of test data is 100%.
[0104] Table 3
[0105]
[0106] Embodiment 2
[0107] like Figure 5 As shown, the present invention provides a chute angle abnormality diagnosis system based on deep time series image learning, the system comprising:
[0108] The chute data acquisition module 10 is used to collect blast furnace top monitoring video data and obtain blast furnace chute videos under different angles and different environments;
[0109] A data preprocessing module 20 is used to preprocess the blast furnace chute video to obtain a chute angle video data set;
[0110] A model training module 30 is used to input the chute angle video data set into a pre-built chute angle recognition model based on deep time series image learning for model training, so as to obtain a trained chute angle recognition model based on deep time series image learning;
[0111] The prediction and classification module 40 is used to input the real-time collected chute image into the trained chute angle recognition model based on deep time series image learning to make predictions and obtain the final classification results;
[0112] The angle abnormality diagnosis module 50 is used to compare the final classification result with the set chute angle to determine whether the chute angle is abnormal.
[0113] The interface of the chute angle abnormality diagnosis system based on deep temporal image learning proposed in this embodiment is as follows: Figure 6 shown.
[0114] A chute angle abnormality diagnosis system based on deep time series image learning in this embodiment is used to implement the aforementioned chute angle abnormality diagnosis method based on deep time series image learning. Therefore, the specific implementation method of the chute angle abnormality diagnosis system based on deep time series image learning can be seen in the embodiment part of the chute angle abnormality diagnosis method based on deep time series image learning in the previous text. For example, the chute data acquisition module 10, the data preprocessing module 20, the model training module 30, the prediction and classification module 40, and the angle abnormality diagnosis module 50 are respectively used to implement steps S1, S2, S3, S4, and S5 in the aforementioned chute angle abnormality diagnosis method based on deep time series image learning. Therefore, its specific implementation method can refer to the description of the corresponding various partial embodiments. In order to avoid redundancy, it will not be repeated here.
[0115] Embodiment 3
[0116] An embodiment of the present invention also provides an electronic device, which includes a processor, a memory and a bus system, wherein the processor and the memory are connected through the bus system, the memory is used to store instructions, and the processor is used to execute the instructions stored in the memory to implement the above-mentioned chute angle abnormality diagnosis method based on deep time series image learning.
[0117] Embodiment 4
[0118] An embodiment of the present invention also provides a computer storage medium, which stores a computer software product. The computer software product includes a number of instructions for enabling a computer device to execute the above-mentioned chute angle abnormality diagnosis method based on deep time series image learning.
[0119] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.
[0120] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.
[0121] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0122] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0123] Obviously, the above embodiments are merely examples for clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived from these are still within the protection scope of the invention.
Claims
1. A chute angle abnormality diagnosis method based on deep temporal image learning, characterized in that: include: Step S1: Collect blast furnace top monitoring video data to obtain blast furnace chute videos under different angles and different environments; Step S2: preprocessing the blast furnace chute video to obtain a chute angle video data set, specifically comprising: sampling the blast furnace chute video in a time series window, wherein the number of frames for each angle of the chute to rotate one circle in the blast furnace chute video is n, the number of sampled image window frames is M, and a frame of image is added to the window by taking an image every n / M intervals, and taking n / M windows for one video, so as to retain the video features to the greatest extent; Step S3: inputting the chute angle video dataset into a pre-built chute angle recognition model based on deep temporal image learning for model training, and obtaining a trained chute angle recognition model based on deep temporal image learning, specifically including: S31: dividing the chute angle video data set in proportion to obtain a training sample set and a test sample set; S32: Input the training sample set into a pre-built chute angle recognition model based on deep time series image learning to perform model training, specifically including: Firstly, the LSTM network model is used to extract the spatiotemporal feature information of the chute image, and the last LSTM unit is selected to output the feature vector. Then, the fully connected layer is used to map the extracted feature vector to the category of the chute angle. Finally, the Softmax function is used to normalize the score of each category to obtain the probability of each category, and the category with the maximum probability is output as the predicted value of the chute angle. S33: Calculate the loss of the predicted value and the true value using the cross entropy function, save the model weight when the loss function converges, and obtain the trained chute angle recognition model based on deep time series image learning; Step S4: input the real-time collected chute image into the trained chute angle recognition model based on deep time series image learning for prediction to obtain the final classification result, which specifically includes: The real-time collected chute images are input into the trained chute angle recognition model based on deep time series image learning for prediction. Since the video of the chute rotating in one circle is divided into multiple time series windows, when multiple different category results appear in one circle, the category result with the largest output ratio is used as the final classification result of the chute angle in this circle according to the category result ratio; Step S5: Compare the final classification result with the set chute angle to determine whether the chute angle is abnormal.
2. The method for diagnosing abnormal chute angle based on deep temporal image learning according to claim 1, characterized in that: The LSTM network model introduces three gating mechanisms, including an input gate, a forget gate, and an output gate, to control the input, update, and output of information; wherein the input gate is used to control the input of new information, the forget gate is used to control the retention of historical information, and the output gate is used to control the generation of output. The three gating mechanisms together constitute a hidden unit of the LSTM network model.
3. The method for diagnosing abnormal chute angle based on deep temporal image learning according to claim 1, characterized in that: When multiple different category results appear in one circle, according to the category result ratio, the category result with the largest output ratio is used as the final classification result of the chute angle of the circle, specifically including: Assume that N represents the total number of time series windows for one rotation of the chute, c i represents the diagnosis result of the i-th time series window, then the number of times each category appears in the whole cycle count(c) is: Among them, [c i =c] is the indicator function, c is the actual angle of the chute, when c i The value is 1 when it is equal to c, otherwise it is 0; Next, calculate the proportion p(c) of each category in the entire circle as: Finally, output the final result of the circle, that is, the category with the highest proportion for: Among them, argmax is a function that finds parameters for a function and is used to output the category with the largest number of occurrences.
4. A chute angle abnormality diagnosis system based on deep temporal image learning, characterized in that: The system for implementing the method for diagnosing abnormal chute angle based on deep time series image learning according to any one of claims 1 to 3 comprises: The chute data acquisition module is used to collect blast furnace top monitoring video data and obtain blast furnace chute videos under different angles and environments; A data preprocessing module, used for preprocessing the blast furnace chute video to obtain a chute angle video data set; A model training module is used to input the chute angle video data set into a pre-built chute angle recognition model based on deep time series image learning for model training, so as to obtain a trained chute angle recognition model based on deep time series image learning; The prediction and classification module is used to input the real-time collected chute images into the trained chute angle recognition model based on deep time series image learning to make predictions and obtain the final classification results; The angle abnormality diagnosis module is used to compare the final classification result with the set chute angle to determine whether the chute angle is abnormal.
5. An electronic device, characterized in that: The electronic device includes a processor, a memory and a bus system, the processor and the memory are connected through the bus system, the memory is used to store instructions, and the processor is used to execute the instructions stored in the memory to implement the chute angle abnormality diagnosis method based on deep time series image learning as described in any one of claims 1 to 3.
6. A computer storage medium, characterized in that: The computer storage medium stores a computer software product, and the computer software product includes a number of instructions for enabling a computer device to execute the chute angle abnormality diagnosis method based on deep time series image learning as described in any one of claims 1 to 3.
