A method for automatically setting intermediate blank thickness by a neural network model

Abstract

The application discloses a method for automatically setting intermediate billet thickness through a neural network model, and the neural network model is established by collecting relevant process parameters in historical production, wherein the input layer of the model is operator type, steel type, steel plate width / thickness, intermediate billet length, slab thickness and the like, and the output layer is intermediate billet thickness deviation; in subsequent production, according to the process parameters of the current steel plate, the neural network model can automatically give a reference value of the appropriate intermediate billet thickness deviation parameter, and automatically set the reference value and transmit the reference value to the operation platform; if the operator does not change the reference value, the production setting is performed according to the reference value; if the operator intervenes and modifies the reference value, the production setting is performed according to the parameter value input by the operator for the last time. By using the method, the rolling production rhythm is effectively improved, the labor intensity of the operator is reduced, and the operation habit of the operator is also standardized.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of steel rolling production, and particularly relates to a method for automatically setting intermediate blank thickness through a neural network model. BACKGROUND

[0002] The controlled rolling and controlled cooling technology (TMCP) has been widely used in the production of medium plate, and has played a positive role in the production of pipeline steel, high-strength structural steel, offshore platform steel, and ship plate. The slab is discharged and rolled into an intermediate blank of a certain thickness and width by a roughing mill to meet the requirements of the subsequent finishing process. The specific thickness of the intermediate blank is related to the thickness of the finished strip and the requirements of the product organization and performance. The roughing rolling parameters are calculated according to the target intermediate blank thickness to set the roll gap, so that the intermediate blank size at the last pass of the roughing mill meets the requirements.

[0003] Generally speaking, the target thickness of the intermediate blank required by the process is generally a range rather than a single value. The thickness of the intermediate blank will affect the rolling parameters of the preceding and subsequent roughing and finishing, and in production, the operator will select the appropriate intermediate blank thickness within the target range according to the operation habits of the roughing and finishing processes, so that the entire rolling rhythm is more efficient. Within the controllable range, determining a most appropriate intermediate blank thickness deviation is beneficial to improving the rolling production efficiency. However, in actual production, the operator adjusts the deviation by experience, which greatly increases the requirements and workload of the operator. At present, the research on the best target size of the intermediate blank in the existing technology is mainly based on the performance of the steel plate, the state of the equipment, etc.

[0004] For example, a Chinese patent with publication number CN117259449 A discloses a roughing intermediate blank thickness control method and system, which mainly solves the problem of large deviation between the actual thickness of the intermediate blank at the last pass of roughing and the target value. The patent establishes a system to ensure that the actual intermediate blank thickness hits the target value. The system combines the wear state of the roll gap and the actual situation of the current rolling pass to real-time correct the next pass roll gap setting value, ensuring that the actual thickness of the intermediate blank hits the target value.

[0005] For another example, a Chinese patent with publication number CN114101344 A discloses an intermediate blank temperature thickness adjustment method during medium plate rolling. From the perspective of product quality, the patent provides an intermediate blank temperature thickness adjustment method for thick gauge steel plate rolling. According to the specific pre-rolling slab size and steel plate thickness, a more accurate intermediate blank temperature thickness upper and lower limit control range is calculated using an empirical formula, the intermediate blank temperature thickness control range is optimized, the problem of unstable steel plate performance caused by excessively wide intermediate blank thickness control range is avoided, especially the problem of low yield strength of finished steel plate, and finally the product yield is improved.

[0006] For example, the Chinese patent with publication number CN116475244 A discloses an intermediate blank thickness adjustment method and device. The method includes obtaining a current rolling steel degree setting value and a current intermediate blank thickness target value of the rolling steel; obtaining a quality inspection result; obtaining an intermediate blank result of the previous rolling steel that does not satisfy a first preset condition, or an intermediate blank thickness setting value of the previous rolling steel and the intermediate blank thickness of the current rolling steel; obtaining an intermediate blank thickness adjustment value; determining the intermediate blank thickness setting value of the next rolling steel according to a difference between the intermediate blank thickness adjustment value and the target value that does not satisfy a second preset condition; and adjusting the intermediate blank thickness of the next rolling steel. In this way, when the intermediate blank thickness or the working condition changes, the operator can adjust the thickness setting value according to the on-site situation, and the intermediate blank setting value is adjusted in time when the quality of the rolling steel on site changes, thereby avoiding batch quality defects of the product.

[0007] Due to the complexity of the influencing factors of the intermediate blank thickness, under the premise of meeting the process requirements, the intermediate blank thickness needs to be set in combination with the current rough rolling and finishing rolling rhythm and the operator's habit, and it is difficult to develop a related intermediate blank thickness self-adaptive mathematical model. Therefore, the matching of the current rough rolling and finishing rolling rhythm needs to rely on the operator to carefully and timely adjust the intermediate blank thickness deviation amount based on production experience, which greatly increases the ability requirement and workload of the operator. SUMMARY

[0008] The present application provides an automatic operation method for producing a medium plate based on a rolling control system. The method can automatically give an intermediate blank thickness deviation parameter and automatically set the parameter according to current steel plate information, effectively improve the rolling production rhythm, and reduce the labor intensity of the operator.

[0009] To achieve the above-mentioned purpose, the present application adopts the following technical scheme: a method for setting an intermediate blank thickness based on a neural network model, characterized in that it comprises the following steps.

[0010] S1: Collect historical data of steel plate specification parameters and corresponding intermediate blank thickness deviation parameters and establish a computer background database, and establish a neural network model based on the database; wherein the input layer of the neural network model is the operator characteristics, steel grade, steel plate width / thickness, intermediate blank length and slab thickness, and the output layer is the intermediate blank thickness deviation amount set manually by the operator.

[0011] S2: In the rough rolling production process, the current steel plate rolling parameters are collected. If the specification parameters of the current steel plate exist in the database, the neural network model automatically gives a reference value of the intermediate blank thickness deviation parameter, and step S3 is executed. If the specification parameters of the current steel plate do not exist in the database, the operator intervenes and manually inputs the intermediate blank thickness deviation parameter.

[0012] S3: automatically set the reference value of the intermediate blank thickness deviation parameter, and transmit it to the operation station in real time, if the operator does not change the reference value, the production setting is carried out according to the reference value, if the operator intervenes, the parameter value input by the operator last time is used for production setting.

[0013] Further, the neural network model continues the rolling parameters of each steel plate, and the neural network model is improved through self-learning.

[0014] Further, the neural network model is established as follows.

[0015] (1) Collecting the historical data of the related process, establishing a large database for neural network training, and the specific process parameters include: operator type, steel grade, steel plate width / thickness, intermediate blank length, slab thickness, and manual setting of intermediate blank thickness deviation amount by the operator.

[0016] (2) 90% of the data groups in the database are set as the training group, and 10% of the data groups are set as the experimental group.

[0017] (3) Without manually setting the initial weight value and bias term, the neural network self-learning model is established through the training group data, wherein the activation function is Sigmoid function, and the hidden layer is set to two layers.

[0018] (4) The accuracy and calculation speed of the neural network model are verified through the experimental group, and the error between the model output deviation amount and the actual deviation amount of the experimental group is less than 10%, which is the satisfaction condition. If the accuracy and calculation speed of the neural network model meet the requirements of online production, the model is successfully established. If the accuracy and calculation speed of the neural network model cannot meet the requirements of online production, the weight value of the input layer parameter can be manually set according to the production experience, and the bias term of the intermediate blank length and the actual situation of the steel plate width / thickness can be set according to the characteristics of the activation function, and the model training is re-performed, and after the training is completed, step (4) is repeated.

[0019] Further, the setting logic of the bias term and the weight value is determined by the operator according to the actual production experience data and the specific parameters of the database.

[0020] The beneficial effects of the present application are: the present application adopts a regression analysis method based on big data, establishes a machine learning model based on a historical production database, gives a suitable reference value of the intermediate blank thickness deviation according to the current process parameters, and automatically sets it. Effectively improve the rolling production rhythm, reduce the labor intensity of the operator, and at the same time, it can also standardize the operation habit of the operator. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1The intermediate blank thickness deviation neural network model established in the present application.

[0022] Figure 2 The neural network model correction method example diagram.

[0023] Figure 3 The neural network model data set example.

[0024] Figure 4 The Sigmoid activation function curve.

[0025] Figure 5 The model input layer data conversion example.

[0026] Figure 6 The interface display when the operator inputs parameters.

[0027] Figure 7 The interface display when the reference value is automatically loaded. DETAILED DESCRIPTION

[0028] In order to make the technical solutions and advantages clearer, the technical solutions will be described clearly and completely in combination with specific process data and corresponding drawings, and the advantages of the present application will be embodied.

[0029] The present application is a method for setting intermediate blank thickness based on a neural network model, comprising the following steps.

[0030] S1: Collect the steel plate specification parameters and the corresponding intermediate blank thickness deviation parameter historical data and establish a computer background database, and establish a neural network model based on the database; wherein the input layer of the neural network model is the operator characteristics, steel grade, steel plate width / thickness, intermediate blank length and slab thickness, and the output layer is the intermediate blank thickness deviation amount manually set by the operator.

[0031] S2: In the rough rolling production process, the current steel plate rolling parameters are collected, if the specification parameters of the current steel plate exist in the database, the neural network model automatically gives the reference value of the intermediate blank thickness deviation parameter, and step S3 is executed. If the specification parameters of the current steel plate do not exist in the database, the operator intervenes and manually inputs the intermediate blank thickness deviation parameter.

[0032] S3: The reference value of the intermediate blank thickness deviation parameter is automatically set and transmitted to the operation platform in real time, if the operator does not change the reference value, the production setting is performed according to the reference value; if the operator intervenes and modifies the reference value, the production setting is performed according to the last parameter value input by the operator.

[0033] Specifically, the technical scheme of the present application is to collect the intermediate blank thickness deviation values set under different processes in historical production, and to establish a neural network model according to the related process parameter types, the parameters involved including operator habit, steel grade, steel plate width / thickness, intermediate blank length, slab thickness, etc. The basic design rule of the background neural network model is: taking the operator type, steel grade, target width range of the rolled steel plate, target thickness range of the rolled steel plate, and intermediate blank length as the index conditions of the input, and the output data including the intermediate blank thickness deviation offset.

[0034] First step: Collect the historical data of the related process, and establish a large database for neural network training, the specific process parameters including: operator type, steel grade, steel plate width / thickness, intermediate blank length, slab thickness, and intermediate blank thickness deviation amount set manually by the operator. Among them, the operators are divided into 4 categories according to their operation habits, and parameters 1, 2, 3, and 4 are used to represent them respectively.

[0035] Second step: Use random sampling to set 90% of the data groups in the database as the training group, and 10% of the data groups as the experimental group.

[0036] Third step: Without manually setting the initial weight value and bias term of the input layer, a neural network self-learning model is established through the training group data (the activation function is Sigmoid function), and the hidden layer is 2 layers.

[0037] Sigmoid activation function: ;

[0038] This function can map a real number to the interval (0, 1), and is commonly used as the activation function of the input layer in neural networks. It works well when the input value characteristics differ greatly or the absolute value is close to 0.

[0039] Fourth step: Verify its accuracy and calculation speed through the experimental group. The error between the model output deviation and the actual deviation of the experimental group is less than 10% as the satisfaction condition, and the verification result is shown in Table 1.

[0040] Table 1: Error statistics of the experimental group

[0041]

[0042] According to Table 1, the number of groups with an error less than 10% is only 52.5%. The model needs to be corrected. The specific operation is: manually set the weight before retraining the model, according to the production experience, reduce the steel grade and steel plate width weight value to 0.3. And set the bias term for the steel plate thickness / width and the intermediate blank length. The reason for choosing the intermediate blank length and the steel plate width as the bias term is that when the absolute value in the input layer is greater than 5, the "gradient disappearance" phenomenon is easy to occur after the Sigmoid function activation (for exampleFigure 4 ).

[0043] The input layer bias term formula is:

[0044]

[0045] wherein, is the average value of the data, , is the maximum / minimum value of the data. After the input layer parameters are processed by the above formula bias, the value range is determined to be [-5, 5], which is beneficial to the model to distinguish the special nature of the input layer (such as Figure 5 ) during self-learning.

[0046] Then, the training set is trained using the adjusted weight and bias processing, and the accuracy of the corrected neural network model is verified using the experimental group. The error is less than 10% when the model output deviation is less than 10% of the actual deviation of the experimental group. The verification results are shown in Table 2. The error is less than 10% of the number of groups, accounting for 78.9%, which basically meets the production requirements.

[0047] Table 2: Error statistics of the experimental group

[0048]

[0049] In addition, in order to ensure that the neural network model in the background can be discovered and stopped in time when an abnormal situation occurs, the reference value given by the neural network is sent to the operation platform at the same time, and the operator can know and adjust in time. If the intermediate blank handover thickness function input box or button on the screen is the number directly input by the operator or the switch directly selected, including the value modified on the basis of the reference value, the display remains the same as the original WinCC, as shown in Figure 6 .

[0050] If the input box or button on the screen is the reference value or switch given by the L2 background program of the automation system, the input box or button on the WinCC screen will increase a yellow box to prompt, as shown in Figure 7 , to inform the operator that this data is not the original input value, but the reference value automatically adjusted by the background, please pay attention to confirmation. When the operator modifies the value of this input box again, the yellow box disappears.

[0051] If the operator does not adjust the value of this input box, the yellow box remains, and the value is automatically updated by the L2 background until the corresponding reference value function button is closed or the next piece of steel does not have a corresponding reference value (for example, after changing the specification, there is no parameter recommendation script for the new specification), then the yellow box disappears, and the value in the input box returns to the last manually input value.

[0052] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method of setting an intermediate slab thickness based on a neural network model, characterized by, It comprises the following steps: S1: Collecting the historical data of the steel plate specification parameters and the corresponding intermediate blank thickness deviation parameters and establishing a computer background database, and establishing a neural network model based on the database; wherein the input layer of the neural network model is the operator characteristics, steel grade, steel plate width / thickness, intermediate blank length and slab thickness, and the output layer is the intermediate blank thickness deviation amount set manually by the operator; S2: In the rough rolling production process, the current steel plate rolling parameters are collected, if the specification parameters of the current steel plate exist in the database, the neural network model automatically gives the reference value of the intermediate blank thickness deviation parameter, and step S3 is executed; If the specification parameters of the current steel plate do not exist in the database, the operator intervenes, and the intermediate blank thickness deviation parameter is manually inputted; S3: The reference value of the intermediate blank thickness deviation parameter is automatically set and transmitted to the operation platform in real time, if the operator does not change the reference value, the production setting is made according to the reference value; if the operator intervenes and modifies the reference value, the production setting is made according to the last parameter value inputted by the operator; The steps of establishing the neural network model are as follows: (1) Collecting the historical data of the related process and establishing a large database for neural network training, the specific process parameters include: operator type, steel grade, steel plate width / thickness, intermediate blank length, slab thickness, and intermediate blank thickness deviation amount set manually by the operator; (2) Setting 90% of the data groups in the database as the training group and 10% of the data groups as the experimental group; (3) Without manually setting the initial weight value and bias term, the neural network self-learning model is established through the training group data, wherein the activation function is Sigmoid function and the hidden layer is set to two layers; (4) The accuracy and calculation speed of the neural network model are verified through the experimental group, and the error between the model output deviation amount and the actual deviation amount of the experimental group is less than 10% as the satisfaction condition; If the accuracy and calculation speed of the neural network model meet the requirements of online production, the model is successfully established; If the accuracy and calculation speed of the neural network model cannot meet the requirements of online production, the weight value of the input layer parameter can be manually set according to the production experience, and the bias term of the intermediate blank length and steel plate width / thickness can be set according to the characteristics of the activation function, and the model training is re-performed, and after the training is completed, step (4) is repeated.

2. The method of setting an intermediate slab thickness based on a neural network model according to claim 1, wherein, The neural network model will continue the rolling parameters of each steel plate to improve the neural network model through self-learning.

3. The method of setting an intermediate slab thickness based on a neural network model according to claim 1, wherein, The setting logic of the bias term and the weight value is determined by the operator according to the actual production experience data and the specific parameters of the database.

Images