A method for solving entry section threading abnormalities by using AI technology

Abstract

The application discloses a kind of methods for solving entry section threading abnormality using AI technology, comprising the following steps: installing camera step: multiple cameras are arranged at the entry section of the color coater unit, and the angle of the camera is adjusted, so that the entry of the color coater unit is located in the video area of multiple cameras;Networking step: multiple cameras are networked, then multiple cameras are time adjusted, so that the time of multiple cameras is synchronized, and a shared space and a world coordinate are constructed according to the viewing angle of multiple cameras;Calibration step: multiple cameras are sequentially coded, and each camera is initialized, then the camera that has completed initialization is calibrated with the world coordinate.The application identifies and analyzes abnormal state through algorithm, and can automatically analyze and judge the probability of threading abnormality such as steel jam according to incoming material, give judgment and assist decision suggestion, finally realize reducing the probability of threading abnormality at the entry section, improve production efficiency and save maintenance cost.

Description

Technical Field

[0001] This invention relates to the field of metallurgical industry technology, specifically to a method for solving abnormal belt threading at the inlet section using AI technology. Background Technology

[0002] Color-coated steel sheets are produced on continuous mills using cold-rolled strip steel or galvanized strip steel (electroplated and hot-dip galvanized) as substrates. After surface pretreatment (degreasing and chemical treatment), one or more layers of liquid coating are applied using a roller coating method. The sheets are then baked and cooled. The processing of color-coated steel sheets requires the use of cold-rolled color-coating mills.

[0003] For example, CN201610085913.7 discloses a method and device for detecting deformation of heat-treated steel strip based on machine vision. The front camera and the rear camera respectively acquire image information of the steel strip before and after straightening, and then transmit the information to the processor for processing. The processor compares and analyzes the image information to obtain the straightening result, and reports the straightening result to the straightening machine. The straightening machine automatically adjusts the straightening process parameters according to the straightening result.

[0004] During strip threading at the inlet section of a cold-rolled color coating mill, strip jamming and piling often occur. These phenomena are rarely predictable, requiring frequent maintenance and handling of the cold-rolled color coating mill, thus reducing its efficiency and ultimately lowering production output. Therefore, there is an urgent need to design a method using AI technology to address these abnormal strip threading issues at the inlet section. Summary of the Invention

[0005] The purpose of this invention is to provide a method for solving the abnormal tape-carrying at the entrance section using AI technology, so as to overcome the above-mentioned shortcomings in the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for resolving tape-carrying anomalies at the entrance segment using AI technology includes the following steps:

[0008] Camera installation steps: Install multiple cameras at the entrance section of the color coating unit and adjust the angle of the cameras so that the entrance of the color coating unit is located in the video area of ​​multiple cameras;

[0009] Network setup steps: Connect multiple cameras to a network, then adjust the time of the multiple cameras to synchronize their time, and build a shared space and world coordinates based on the perspectives of the multiple cameras;

[0010] Calibration steps: Sequence the multiple cameras, initialize each camera, and then calibrate the initialized cameras with world coordinates;

[0011] Data collection steps: The camera captures images of materials passing through the inlet of the color coating unit and automatically generates speed, progress, and position parameters. These parameters are then synchronized with the backend, and the operating parameters of the color coating unit at the same time are also recorded and synchronized to the backend.

[0012] Database setup steps: The backend establishes the database based on material parameters and color coating unit operating parameters;

[0013] AI algorithm analysis steps: Use AI technology to filter and organize data in the database, and use algorithms to identify and analyze patterns such as the correlation and association of abnormal states;

[0014] Judgment steps: Based on the correlation and association of abnormal states, the incoming material situation is judged, and the probability of abnormal threading state is automatically analyzed and judged, and judgment and auxiliary decision-making suggestions are given.

[0015] Furthermore, the camera tracks the threading process in real time, monitors situations such as steel strip jamming and stacking, and records and stores data such as material speed, progress, position parameters, and unit operating parameters at the same time in the database.

[0016] Furthermore, in the AI ​​algorithm analysis step, the method for filtering and organizing the data in the database is as follows:

[0017] The data threshold is calculated using AI algorithms: select some groups of data and randomly initialize the center point of each group of data;

[0018] Calculate the distance from each data point to the center point, and group the data points closest to the center point into the same category;

[0019] Calculate the center point in each class as the new center point, and repeat the above steps until the center of each class does not change much after each iteration.

[0020] Based on the two types of points obtained, recalculate the centroids of the two types of points and reassign all points to one of the two new centroids. Repeat the above process until the centroid of each type does not change much after each iteration.

[0021] Furthermore, in the process of calculating the data threshold using the AI ​​algorithm, the center point is the position with the same vector length as each data point, the number of classes is a known number, the number of classes is the number of center points, and the centroid is the point with the shortest distance to all points of that class.

[0022] Furthermore, in the AI ​​algorithm analysis step: the AI ​​algorithm includes one of the following: linear regression algorithm, logistic regression algorithm, support vector machine algorithm, decision tree algorithm, random forest algorithm, and neural network algorithm.

[0023] Furthermore, in the AI ​​algorithm analysis step: the AI ​​algorithm adopts the logistic regression algorithm, inputs a y to the logistic regression algorithm, defines it as divided into two categories, and records the predicted classification results through the field fcst.

[0024] Furthermore, when performing analysis using the logistic regression algorithm, logistic regression employs two basic assumptions: the first assumption is that the data follows a Bernoulli distribution; the second assumption is that the probability of a sample being positive conforms to the output of the sigmoid function.

[0025] Furthermore, the logistic regression algorithm continuously approximates the optimal solution by performing gradient descent on the sigmoid function. The gradient descent method includes: stochastic gradient descent, batch gradient descent, and mini-batch gradient descent.

[0026] Furthermore, the loss function used in the logistic regression algorithm generally includes any one of the following: 0-1 loss function, squared loss function, absolute value loss function, logarithmic loss function, hinge loss function, and maximum likelihood function.

[0027] Furthermore, the classification result only has 0 and 1, and the two categories defined are normal and abnormal, with more being normal and fewer being abnormal.

[0028] In the above technical solution, the present invention provides a method for solving the abnormal tape-crossing at the entrance segment using AI technology, which has the following beneficial effects:

[0029] (1) This invention identifies and analyzes abnormal states through algorithms, and can automatically analyze and judge the probability of abnormal states such as steel jamming based on the incoming material situation, and provide judgment and auxiliary decision-making suggestions, so as to reduce the probability of abnormal occurrence of threading in the inlet section, improve production efficiency, and save maintenance costs.

[0030] (2) This invention uses the logistic regression algorithm to predict the probability of anomalies, which makes the model very interpretable. The training speed is further improved by stacking machines, and the resource consumption is small, especially memory, which makes it easy to adjust the output results.

[0031] (3) By installing multiple cameras, the present invention can provide more data when detecting abnormal tape threading at the entrance section of the color coating unit, making the detection of abnormal tape threading at the entrance section of the color coating unit more accurate. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0033] Figure 1 This is a flowchart illustrating an embodiment of a method for resolving tape-crossing anomalies at the entrance segment using AI technology, according to the present invention.

[0034] Figure 2 This is a data graph showing the predicted calculation results provided by an embodiment of the method for solving the abnormal tape crossing at the entrance section using AI technology according to the present invention. Detailed Implementation

[0035] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0036] like Figure 1-2 As shown in the figure, an embodiment of the present invention provides a method for solving the abnormal tape-crossing at the entrance section using AI technology, which includes the following steps:

[0037] Camera installation steps: Install multiple cameras at the entrance section of the color coating unit and adjust the angle of the cameras so that the entrance of the color coating unit is located in the video area of ​​multiple cameras;

[0038] Network setup steps: Connect multiple cameras to a network, then adjust the time of the multiple cameras to synchronize their time, and build a shared space and world coordinates based on the perspectives of the multiple cameras;

[0039] Calibration steps: Sequence the multiple cameras, initialize each camera, and then calibrate the initialized cameras with world coordinates;

[0040] Data collection steps: The camera captures images of materials passing through the inlet of the color coating unit and automatically generates speed, progress, and position parameters. These parameters are then synchronized with the backend, and the operating parameters of the color coating unit at the same time are also recorded and synchronized to the backend.

[0041] Database setup steps: The backend establishes the database based on material parameters and color coating unit operating parameters;

[0042] AI algorithm analysis steps: Use AI technology to filter and organize data in the database, and use algorithms to identify and analyze patterns such as the correlation and association of abnormal states;

[0043] Judgment steps: Based on the correlation and association of abnormal states, the incoming material situation is judged, and the probability of abnormal threading state is automatically analyzed and judged, and judgment and auxiliary decision-making suggestions are given.

[0044] Specifically, this embodiment includes the following steps:

[0045] Camera installation steps: Install multiple cameras at the entrance section of the color coating unit and adjust the angle of the cameras so that the entrance of the color coating unit is located in the video area of ​​multiple cameras;

[0046] Network setup steps: Connect multiple cameras to a network, then adjust the time of the multiple cameras to synchronize their time, and build a shared space and world coordinates based on the perspectives of the multiple cameras;

[0047] Calibration steps: Sequence the multiple cameras, initialize each camera, and then calibrate the initialized cameras with world coordinates;

[0048] Data collection steps: The camera captures images of materials passing through the inlet of the color coating unit and automatically generates speed, progress, and position parameters. These parameters are then synchronized with the backend, and the operating parameters of the color coating unit at the same time are also recorded and synchronized to the backend.

[0049] Database setup steps: The backend establishes the database based on material parameters and color coating unit operating parameters;

[0050] AI algorithm analysis steps: AI technology is used to filter and organize data in the database; algorithms are then used to identify and analyze patterns such as the correlation and association of abnormal states.

[0051] Judgment steps: Based on the correlation and association of abnormal states, the incoming material situation is judged, and the probability of abnormal threading state is automatically analyzed and judged, and judgment and auxiliary decision-making suggestions are given.

[0052] This invention provides a method for solving tape threading abnormalities at the entry section using AI technology. This invention identifies and analyzes abnormal states through algorithms, and can automatically analyze and judge the probability of tape threading abnormalities such as steel jamming based on the incoming material situation, and provide judgment and auxiliary decision-making suggestions, ultimately reducing the probability of tape threading abnormalities at the entry section, improving production efficiency, and saving maintenance costs.

[0053] In another embodiment of the present invention, the camera tracks the threading process in real time, monitors situations such as strip jamming and stacking, and records and stores data such as material speed, progress, position parameters and unit operating parameters at the same time in the database.

[0054] In another embodiment of the present invention, the method for filtering and organizing data in the database during the AI ​​algorithm analysis step is as follows:

[0055] The data threshold is calculated using AI algorithms: select some groups of data and randomly initialize the center point of each group of data;

[0056] Calculate the distance from each data point to the center point, and group the data points closest to the center point into the same category;

[0057] Calculate the center point in each class as the new center point, and repeat the above steps until the center of each class does not change much after each iteration.

[0058] The center point was also initialized randomly multiple times, and then the one with the best result was selected.

[0059] Based on the two types of points obtained, recalculate the centroids of the two types of points and reassign all points to one of the two new centroids. Repeat the above process until the centroid of each type does not change much after each iteration.

[0060] In another embodiment of the present invention, in the data threshold calculated by the AI ​​algorithm, the center point is the position with the same length as the vector of each data point, the number of classes is the known number, the number of classes is the number of center points, and the centroid is the point with the shortest distance to all points of that class.

[0061] In another embodiment of the present invention, in the AI ​​algorithm analysis step: the AI ​​algorithm uses logistic regression, and the input y to the logistic regression algorithm is defined as two categories. The specific principle of the algorithm is as follows:

[0062] import pandas as pd

[0063] import numpy as np

[0064] from sklearn.cluster import KMeans;

[0065] The algorithm is as follows:

[0066] y =[470, 509, 500, 511,435,489,483,483,482. 483, 504, 472,464,481, 501,581, 507,558, 569,467,38

[0067] km = KMeans(n_clusters-2)

[0068] km.fit(y)

[0069] y['fcst'] = km.predict(y)

[0070] The input y is defined as being divided into two categories (one normal and one abnormal). The field fcst records the predicted classification result. There are two categories, 0 and 1, but we do not know which category is considered normal. We make a reasonable assumption that the category with more people is normal and the category with fewer people is abnormal.

[0071] The anomaly detection and handling calculations are as follows:

[0072] labele = y[y.fcst==0]['fcst'].count()

[0073] label1= y[y.fcst==1]['fcst'].count()

[0074] if labell <= labele:

[0075] y['isAbnormal']=y['fcst']

[0076] else:

[0077] y.1oc[y['fcst'] == 0, 'isAbnormal']=1

[0078] y.1oc[y['fcst'] == 1, 'isAbnormal']=0

[0079] y.columns = ['data','fcst',' isAbnormal']

[0080] y[' isAbnormal']= y[ ' isAbnormal ']. astype(int)

[0081] y = y['data',' isAbnormal'll

[0082] The system records the number of items classified as 1 and the number of items classified as 0. The `isAbnormal` field records whether an item is abnormal, with 0 indicating normal and 1 indicating abnormal. The `fcst` field records the predicted classification result, which has only two possible values: 0 and 1. The two categories are defined as normal and abnormal, with more items in each category indicating normality and fewer items indicating abnormality. When performing analysis, the logistic regression algorithm uses two basic assumptions: first, that the data follows a Bernoulli distribution; and second, that the probability of a positive sample conforms to the output of the sigmoid function. The logistic regression algorithm continuously approaches the optimal solution by applying gradient descent to the sigmoid function. Gradient descent methods include stochastic gradient descent, batch gradient descent, and mini-batch gradient descent.

[0083] Batch gradient descent: It can obtain the global optimum, but the disadvantage is that it requires traversing all the data when updating each parameter, which will result in a large amount of computation and slow updates for each parameter.

[0084] Stochastic gradient descent updates parameters using only one sample point at a time. Therefore, the cost can fluctuate significantly.

[0085] Mini-batch gradient descent combines the advantages of SGD and batch GD, using n samples per update. This reduces the number of parameter updates, resulting in more stable convergence. The final partitioning typically uses a threshold; predicted values ​​above this threshold belong to one class, and predicted values ​​below it belong to another. The threshold is adjusted based on the specific situation, but 0.5 is commonly chosen. In logistic regression, loss functions typically include 0-1 loss, squared loss, absolute value loss, logarithmic loss, hinge loss, and maximum likelihood function. Taking the logarithm of the maximum likelihood function is equivalent to the logarithmic loss function. In logistic regression, the logarithmic loss function is relatively fast for training and parameter calculation. Furthermore, it is independent of the gradient of the sigmoid function itself, ensuring a consistently stable update speed.

[0086] Example 1

[0087] A method for resolving tape-carrying anomalies at the entrance segment using AI technology includes the following steps:

[0088] Camera installation steps: Install multiple cameras at the entrance section of the color coating unit and adjust the angle of the cameras so that the entrance of the color coating unit is located in the video area of ​​multiple cameras;

[0089] Network setup steps: Connect multiple cameras to a network, then adjust the time of the multiple cameras to synchronize their time, and build a shared space and world coordinates based on the perspectives of the multiple cameras;

[0090] Calibration steps: Sequence the multiple cameras, initialize each camera, and then calibrate the initialized cameras with world coordinates;

[0091] Data collection steps: The camera captures images of materials passing through the inlet of the color coating unit and automatically generates speed, progress, and position parameters. These parameters are then synchronized with the backend, and the operating parameters of the color coating unit at the same time are also recorded and synchronized to the backend.

[0092] Database setup steps: The backend establishes the database based on material parameters and color coating unit operating parameters;

[0093] AI algorithm analysis steps: AI technology is used to filter and organize data in the database; algorithms are then used to identify and analyze patterns such as the correlation and association of abnormal states.

[0094] Judgment Steps: Based on the correlation and association patterns of abnormal states, determine the incoming material situation, automatically analyze and judge the probability of abnormal threading conditions, and provide judgment and auxiliary decision-making suggestions.

[0095] Example 2

[0096] A method for resolving tape threading anomalies at the entry section using AI technology is proposed in this embodiment, which further refines the method based on Embodiment 1. The camera tracks the tape threading process in real time, monitoring for issues such as strip jamming and stacking, and records and stores data such as material speed, progress, position parameters, and unit operating parameters within the same time frame in a database. In the AI ​​algorithm analysis step, the method for filtering and organizing the data in the database is as follows:

[0097] The data threshold is calculated using AI algorithms: select some groups of data and randomly initialize the center point of each group of data;

[0098] Calculate the distance from each data point to the center point, and group the data points closest to the center point into the same category;

[0099] Calculate the center point in each class as the new center point, and repeat the above steps until the center of each class does not change much after each iteration.

[0100] The center point was also initialized randomly multiple times, and then the one with the best result was selected.

[0101] Based on the two types of points obtained, the centroids of the two types of points are recalculated, and all points are reassigned to one of the two new centroids. This process is repeated until the centroid of each type does not change significantly after each iteration. In the data threshold calculated by the AI ​​algorithm, the centroid is the position with the same vector length as each data point, the number of classes is the predicted number, and the number of classes is the number of centroids. The centroid is the point with the shortest distance to all points in that class. In the AI ​​algorithm analysis steps: the AI ​​algorithm uses logistic regression. The input y to the logistic regression algorithm is defined as two classes. The specific principle of the algorithm is as follows:

[0102] import pandas as pd

[0103] import numpy as np

[0104] from sklearn.cluster import KMeans;

[0105] The algorithm is as follows:

[0106] y =[470, 509, 500, 511,435,489,483,483,482. 483, 504, 472,464,481, 501,581, 507,558, 569,467,38

[0107] km = KMeans(n_clusters-2)

[0108] km.fit(y)

[0109] y['fcst'] = km.predict(y)

[0110] The input y is defined as being divided into two categories (one normal and one abnormal). The field fcst records the predicted classification result. There are two categories, 0 and 1, but we do not know which category is considered normal. We make a reasonable assumption that the category with more people is normal and the category with fewer people is abnormal.

[0111] The anomaly detection and handling calculations are as follows:

[0112] labele = y[y.fcst==0]['fcst'].count()

[0113] label1= y[y.fcst==1]['fcst'].count()

[0114] if labell <= labele:

[0115] y['isAbnormal']=y['fcst']

[0116] else:

[0117] y.1oc[y['fcst'] == 0, 'isAbnormal']=1

[0118] y.1oc[y['fcst'] == 1, 'isAbnormal']=0

[0119] y.columns = ['data','fcst',' isAbnormal']

[0120] y[' isAbnormal']= y[ ' isAbnormal ']. astype(int)

[0121] y = y['data',' isAbnormal'll

[0122] The system records the number of items classified as 1 and the number of items classified as 0. The `isAbnormal` field records whether an item is abnormal, with 0 indicating normal and 1 indicating abnormal. The `fcst` field records the predicted classification result, which has only two possible values: 0 and 1. The two categories are defined as normal and abnormal, with more items in each category indicating normality and fewer items indicating abnormality. When performing analysis, the logistic regression algorithm uses two basic assumptions: first, that the data follows a Bernoulli distribution; and second, that the probability of a positive sample conforms to the output of the sigmoid function. The logistic regression algorithm continuously approaches the optimal solution by applying gradient descent to the sigmoid function. Gradient descent methods include stochastic gradient descent, batch gradient descent, and mini-batch gradient descent.

[0123] Batch gradient descent: It can obtain the global optimum, but the disadvantage is that it requires traversing all the data when updating each parameter, which will result in a large amount of computation and slow updates for each parameter.

[0124] Stochastic gradient descent updates parameters using only one sample point at a time. Therefore, the cost can fluctuate significantly.

[0125] Mini-batch gradient descent combines the advantages of SGD and batch GD, using n samples for each update. This reduces the number of parameter updates, resulting in more stable convergence. The final partitioning typically uses a threshold; predicted values ​​above this threshold belong to one class, and predicted values ​​below it belong to another. The threshold is adjusted based on the specific situation. A threshold of 0.5 is commonly used. In logistic regression, loss functions typically include 0-1 loss, squared loss, absolute value loss, logarithmic loss, hinge loss, and maximum likelihood function. Taking the logarithm of the maximum likelihood function is equivalent to the logarithmic loss function. In logistic regression, the logarithmic loss function is relatively fast for training and parameter solving. Furthermore, it is independent of the gradient of the sigmoid function itself, resulting in a consistently stable update speed.

[0126] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for solving tape-crossing anomalies at the entrance segment using AI technology, characterized in that, Includes the following steps: Camera installation steps: Install multiple cameras at the entrance section of the color coating unit and adjust the angle of the cameras so that the entrance of the color coating unit is located in the video area of ​​multiple cameras; Network setup steps: Connect multiple cameras to a network, then adjust the time of the multiple cameras to synchronize their time, and build a shared space and world coordinates based on the perspectives of the multiple cameras; Calibration steps: Sequence the multiple cameras, initialize each camera, and then calibrate the initialized cameras with world coordinates; Data collection steps: The camera captures images of materials passing through the inlet of the color coating unit and automatically generates speed, progress, and position parameters. These parameters are then synchronized with the backend, and the operating parameters of the color coating unit at the same time are also recorded and synchronized to the backend. Database setup steps: The backend establishes the database based on material parameters and color coating unit operating parameters; AI algorithm analysis steps: Use AI technology to filter and organize data in the database, and use algorithms to identify and analyze the correlation and patterns of abnormal states; Judgment steps: Based on the correlation and association patterns of abnormal states, determine the incoming material situation, automatically analyze and determine the probability of abnormal belt threading, and provide judgment and auxiliary decision-making suggestions; The camera tracks the conveyor belt process in real time, monitoring for steel strip jamming and stacking. It also records and stores the material speed, progress, position parameters, and unit operating parameters simultaneously in the database. In the AI ​​algorithm analysis step, the method for filtering and organizing the data in the database is as follows: The data threshold is calculated using AI algorithms: select some groups of data and randomly initialize the center point of each group of data; Calculate the distance from each data point to the center point, and group the data points closest to the center point into the same category; Calculate the center point in each class as the new center point, and repeat the above steps until the center of each class does not change much after each iteration. Based on the two types of points obtained, the centroids of the two types of points are recalculated, and all points are reassigned to one of the two new centroids. The above process is repeated until the centroid of each type does not change much after each iteration. In the data threshold calculated by the AI ​​algorithm, the centroid is the position with the same vector length as each data point. The number of classes is a known number, which is the number of centroids. The centroid is the point with the shortest distance to all points in that class. In the AI ​​algorithm analysis step, the AI ​​algorithm includes one of the following: linear regression algorithm, logistic regression algorithm, support vector machine algorithm, decision tree algorithm, random forest algorithm, and neural network algorithm.

2. The method for solving the abnormal tape-carrying at the entrance section using AI technology according to claim 1, characterized in that, In the AI ​​algorithm analysis steps: the AI ​​algorithm uses logistic regression, inputs a y to the logistic regression algorithm, defines two categories, and records the predicted classification results through the field fcst.

3. The method for solving the abnormal tape-carrying at the entrance section using AI technology according to claim 2, characterized in that, When performing analysis, the logistic regression algorithm adopts two basic assumptions. The first basic assumption is that the data follows a Bernoulli distribution; the second assumption is that the probability of a sample being positive conforms to the output of the sigmoid function.

4. The method for solving the abnormal tape-carrying at the entrance section using AI technology according to claim 3, characterized in that, The logistic regression algorithm continuously approaches the optimal solution by performing gradient descent on the sigmoid function. The gradient descent method includes stochastic gradient descent, batch gradient descent, and mini-batch gradient descent.

5. The method for solving the abnormal tape-carrying at the entrance section using AI technology according to claim 4, characterized in that, The loss function used in the logistic regression algorithm generally includes any one of the following: 0-1 loss function, squared loss function, absolute value loss function, log loss function, hinge loss function, and maximum likelihood function.

6. The method for solving the abnormal tape-carrying at the entrance section using AI technology according to claim 5, characterized in that, The classification result has only 0 and 1. The two categories defined are normal and abnormal. The more common categories are normal, and the fewer common categories are abnormal.

Images