Method and device for setting instability early warning value of cold rolling mill based on vibration energy

Abstract

The invention provides a setting method and device for an instability early-warning value of a cold rolling mill based on vibration energy. The method comprises the steps that actual operation parameters in the operation process of the cold rolling mill are obtained; corresponding rolled piece forming power is obtained based on the actual operation parameters; a vibration energy prediction model is called to process the actual operation parameters and the corresponding rolled piece forming power, and predicted vibration energy of the cold rolling mill is obtained; and a cold rolling mill instability early warning value is set based on the predicted vibration energy. According to the method, the rolled piece forming power is calculated through the actual operation parameters, the rolled piece forming power and the actual operation parameters serve as input of the vibration energy prediction model together, prediction of the vibration energy can be closer to the actual physical nature of the rolling process, and therefore the accuracy and reliability of the pre-warning value set based on the vibration energy are improved. In addition, enough vibration suppression regulation and control time can be reserved for operators, the loss and damage of production equipment are effectively avoided, and the maintenance cost of enterprises is reduced.

Description

Technical Field

[0001] This application relates to the field of cold rolling mill instability early warning technology, and in particular to a method and device for setting cold rolling mill instability early warning values ​​based on vibration energy. Background Technology

[0002] High-value-added cold-rolled sheet is a core basic material in the automotive, home appliance, and shipbuilding industries, and its quality and performance directly affect the technological upgrading needs of downstream industries. With the continuous improvement of product performance indicators in downstream industries, the control precision requirements for thin-gauge high-strength steel during the rolling process are also showing a significant upward trend. However, in actual production, when the rolling speed increases to a critical threshold, the rolling mill generates various unexplained "ghost vibrations," which not only accelerate the wear of key components such as rolls and bearings but also lead to fluctuations in strip thickness and periodic vibration marks on the roll surface. Such vibration problems have become a key technical bottleneck restricting the high-speed, stable, and intelligent development of sheet and strip materials, directly affecting production line operating efficiency and product iteration and upgrading capabilities.

[0003] Production process data and self-excited vibration theory analysis show that increasing rolling speed leads to vibration, with the vibration frequency exhibiting the mill's natural frequency and its harmonics. Reducing the speed restores a stable rolling state. With conventional detection devices in existing cold continuous rolling production lines, operators typically only detect and implement speed reduction measures after vibration intensifies and abnormal noise occurs. By this time, vibration marks on the rolls and strip surface have often appeared, potentially leading to major production accidents such as roll or strip breakage, endangering personal safety. Furthermore, the rolling process involves the coupling of multiple physical fields, including mechanical, fluid, and electrical systems, exhibiting typical nonlinear characteristics. Therefore, relying solely on manual experience to adjust process parameters is insufficient for precise vibration control, resulting in the mill's inability to fully utilize its production capacity.

[0004] To address the aforementioned problem of unstable vibration in cold rolling mills, several monitoring and early warning technologies have been proposed and applied.

[0005] Patent application CN115672994A discloses a method, device, medium, and electronic equipment for early warning of vibration in a cold continuous rolling mill. It determines whether a mill vibration early warning is triggered by calculating the signal-to-noise ratio of a pre-processed characteristic signal. Mill vibration is a process of vibration energy accumulation; once the energy threshold is exceeded, the amplitude increases rapidly within a very short time. Therefore, relying solely on the signal-to-noise ratio of the vibration signal for early warning of mill vibration can lead to untimely system response and may result in misjudgments and missed detections.

[0006] Patent application CN105436205A discloses a method for alarming and suppressing mill vibration. It establishes a mill vibration energy threshold using a vibration sensor installed above the mill stand and the rolling speed signal. When the energy exceeds the threshold, the vibration suppression device increases the thrust of the wedge plate, increasing the frictional resistance between the roll bearing housing and the stand column, thereby reducing the roll vibration energy. However, installing the vibration sensor above the stand results in insufficient energy of the collected vibration signal and loss of some characteristic frequencies, leading to delayed vibration suppression measures. Furthermore, the alarm line correction coefficient is calculated using linear interpolation, lacking theoretical basis, which can lead to an unreasonable setting of the vibration energy threshold.

[0007] Patent application CN114074118A discloses a method for predicting the rolling stability of a six-roll cold rolling mill. The method first calculates the dynamic rolling force and rolling force fluctuation during the vibration process, then establishes the vertical dynamic equations of the mill system, and uses the Newmark-Beta numerical method to obtain the vertical displacement of the rolls. Finally, it judges the stability of the rolling process based on the convergence of these equations. While this method can accurately determine rolling stability and calculate critical vibration velocities based on rolling theory and fluid lubrication theory, many assumptions in the theoretical model reduce prediction accuracy. Furthermore, this method requires calculating roll displacement curves at multiple rolling speeds to determine stability, resulting in long computation time and making it unsuitable for online monitoring.

[0008] Patent application CN118503911A discloses a time-series prediction method for rolling mill vibration based on multi-level network fusion. It first uses a one-dimensional CNN to extract long and short-period features from the data, then trains the short-period and long-period data using LSTM and TCN respectively, while introducing an attention mechanism to improve computational efficiency. Finally, it fuses the two feature sets for output. This method predicts rolling mill vibration solely based on data. Although it is computationally fast and meets online requirements, the model has poor interpretability, failing to explain the conditions under which rolling mill vibration occurs. Furthermore, it cannot provide reasonable justifications for parameter adjustments if used for parameter optimization. Summary of the Invention

[0009] In view of this, this application provides a method and apparatus for setting an instability warning value for a cold rolling mill based on vibration energy, in order to solve at least one of the problems mentioned above.

[0010] To achieve the above objectives, this application adopts the following approach:

[0011] According to a first aspect of this application, a method for setting an instability warning value for a cold rolling mill based on vibration energy is provided. The method includes: acquiring actual operating parameters during the operation of the cold rolling mill; acquiring the corresponding forming power of the rolled piece based on the actual operating parameters; calling a vibration energy prediction model to process the actual operating parameters and the corresponding forming power of the rolled piece to obtain the vibration energy of the cold rolling mill, wherein the vibration energy prediction model is trained using a training dataset consisting of historical operating parameters of the cold rolling mill and the corresponding forming power of the rolled piece; and setting an instability warning value for the cold rolling mill based on the vibration energy.

[0012] As an embodiment of this application, after obtaining the actual operating parameters during the operation of the cold rolling mill in the above method, the method further includes: matching the actual operating parameters obtained by different systems in time; calculating the Mahalanobis distance of each data point from the center point, and determining outliers based on the chi-square distribution; and filling in the missing values ​​and the outliers using third-order Lagrange polynomial interpolation.

[0013] As an embodiment of this application, the actual operating parameters and the historical operating parameters in the above method include: vibration parameters, process parameters, workpiece parameters and roll parameters.

[0014] As an embodiment of this application, the above method for matching the actual operating parameters obtained by different systems in time includes: aligning the process parameters and the vibration parameters in time, wherein the time alignment is based on the time when the rolled piece passes through the weld in the process parameters and the time when the acceleration signal abnormally increases in the vibration parameters.

[0015] As an embodiment of this application, the method described above for obtaining the corresponding rolling power based on the actual operating parameters includes: constructing an exponential velocity field and a strain rate field that satisfy the volume invariance condition and the velocity boundary condition according to the actual operating parameters, and obtaining a functional expression for the rolling power; solving for the neutral angle that minimizes the power functional based on the functional expression for the rolling power, and obtaining the corresponding rolling power.

[0016] As an embodiment of this application, the forming power of the rolled piece in the above method includes: internal plastic deformation power, friction power, shear power and tension power.

[0017] As an embodiment of this application, the method described above, based on the functional expression of the forming power of the rolled piece, obtains the neutral angle that minimizes the power functional by: using the Newton-Raphson iteration method, with the functional expression of the forming power of the rolled piece as the objective function, to solve for the neutral angle that minimizes the objective function.

[0018] As an embodiment of this application, before calling the vibration energy prediction model to process the actual operating parameters and the corresponding rolling power in the above method, the method further includes: using the process parameters, the rolling parameters, the roll parameters, and the rolling power as initial input features; calculating the first maximum mutual information coefficient between the initial input features and the vibration energy; calculating the second maximum mutual information coefficient between the initial input features; and performing data dimensionality reduction based on the first maximum mutual information coefficient and the second maximum mutual information coefficient to obtain the final input features of the vibration energy prediction model.

[0019] As an embodiment of this application, the data dimensionality reduction based on the first maximum mutual information coefficient and the second maximum mutual information coefficient in the above method includes: removing initial input features whose first maximum mutual information coefficient is less than a first preset threshold, and removing initial input features whose second maximum mutual information coefficient is greater than a second preset threshold.

[0020] As an embodiment of this application, the vibration energy prediction model in the above method adopts the Att-BiLSTM network model.

[0021] As an embodiment of this application, the above method further includes: acquiring historical operating parameters of the cold rolling mill, the corresponding forming power of the rolled piece, and the corresponding historical target vibration energy, which together constitute a training dataset and a test dataset; training the vibration energy prediction model using the training dataset, and dynamically adjusting the learning rate using a learning rate preheating and cosine annealing strategy during the training process; continuously optimizing the parameters of the vibration energy prediction model through iterative training on the training dataset, and periodically evaluating the performance of the vibration energy prediction model using the test dataset, until the vibration energy prediction model can accurately predict vibration energy based on the input operating parameters and forming power of the rolled piece, wherein the parameters include at least the number of Att-BiLSTM network nodes, the number of hidden layers, the time step, and the retention rate.

[0022] As an embodiment of this application, the method described above for setting the cold rolling mill instability warning value based on the vibration energy includes: setting the cold rolling mill instability warning value based on the vibration energy and a preset percentile threshold.

[0023] According to a second aspect of this application, a device for setting an instability warning value for a cold rolling mill based on vibration energy is provided. The device includes: an operating parameter acquisition unit for acquiring actual operating parameters during the operation of the cold rolling mill; a forming power acquisition unit for acquiring the corresponding forming power of the rolled piece based on the actual operating parameters; a model invocation unit for invoking a vibration energy prediction model to process the actual operating parameters and the corresponding forming power of the rolled piece to obtain the predicted vibration energy of the cold rolling mill, wherein the vibration energy prediction model is trained using a training dataset consisting of historical operating parameters of the cold rolling mill and the corresponding forming power of the rolled piece; and a warning value setting unit for setting an instability warning value for the cold rolling mill based on the predicted vibration energy.

[0024] As an embodiment of this application, the above-mentioned apparatus further includes: a time matching unit, used to perform time matching on actual operating parameters obtained from different systems; an outlier determination unit, used to calculate the Mahalanobis distance of each data point from the center point and determine outliers based on the chi-square distribution; and a filling unit, used to fill in missing values ​​and outliers using third-order Lagrange polynomial interpolation.

[0025] As an embodiment of this application, the above-mentioned actual operating parameters and the historical operating parameters include: vibration parameters, process parameters, workpiece parameters and roll parameters.

[0026] As an embodiment of this application, the aforementioned timing matching unit is specifically used to: align the process parameters and the vibration parameters in time, wherein the time alignment is based on the time when the rolled piece passes through the weld in the process parameters and the time when the acceleration signal abnormally increases in the vibration parameters.

[0027] As an embodiment of this application, the forming power acquisition unit includes: a functional expression acquisition module, used to construct an exponential velocity field and a strain rate field that satisfy the volume invariance condition and velocity boundary condition based on the actual operating parameters, and obtain a functional expression for the forming power of the rolled piece; and a functional solution module, used to solve for the neutral angle that minimizes the power functional based on the functional expression for the forming power of the rolled piece, and obtain the corresponding forming power of the rolled piece.

[0028] As an embodiment of this application, the forming power of the rolled piece includes: internal plastic deformation power, friction power, shearing power and tension power.

[0029] As an embodiment of this application, the above-mentioned functional solving module is specifically used to: use the Newton-Raphson iteration method, with the functional expression of the forming power of the rolled piece as the objective function, to solve for the neutral angle that minimizes the objective function.

[0030] As an embodiment of this application, the above-mentioned device further includes a data dimensionality reduction unit, which is used to: use the process parameters, the workpiece parameters, the roll parameters and the workpiece forming power as initial input features; calculate the first maximum mutual information coefficient between the initial input features and the vibration energy; calculate the second maximum mutual information coefficient between the initial input features; and perform data dimensionality reduction based on the first maximum mutual information coefficient and the second maximum mutual information coefficient to obtain the final input features of the vibration energy prediction model.

[0031] As an embodiment of this application, the data dimensionality reduction unit performs data dimensionality reduction based on the first maximum mutual information coefficient and the second maximum mutual information coefficient, including: removing initial input features whose first maximum mutual information coefficient is less than a first preset threshold, and removing initial input features whose second maximum mutual information coefficient is greater than a second preset threshold.

[0032] As an embodiment of this application, the above-mentioned vibration energy prediction model adopts the Att-BiLSTM network model.

[0033] As an embodiment of this application, the above-mentioned device further includes a model training unit, used for: acquiring historical operating parameters of the cold rolling mill, the corresponding forming power of the rolled piece, and the corresponding historical target vibration energy, which together constitute a training dataset and a test dataset; training the vibration energy prediction model using the training dataset, and dynamically adjusting the learning rate during the training process using a learning rate preheating and cosine annealing strategy; continuously optimizing the parameters of the vibration energy prediction model through iterative training on the training dataset, and periodically evaluating the performance of the vibration energy prediction model using the test dataset, until the vibration energy prediction model can accurately predict vibration energy based on the input operating parameters and forming power of the rolled piece, wherein the parameters include at least the number of Att-BiLSTM network nodes, the number of hidden layers, the time step, and the retention rate.

[0034] As an embodiment of this application, the aforementioned warning value setting unit is specifically used to: set a cold rolling mill instability warning value based on the vibration energy and a preset percentile threshold.

[0035] According to a third aspect of this application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0036] According to a fourth aspect of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the above-described method.

[0037] According to a fifth aspect of this application, a computer program product is provided, comprising a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0038] As can be seen from the above technical solution, the method and device for setting instability warning values ​​for cold rolling mills based on vibration energy provided in this application calculates the forming power of the rolled piece using actual operating parameters, and uses this power, along with the actual operating parameters, as input to the vibration energy prediction model. The forming power of the rolled piece is a key physical quantity affecting the stability of the rolling process; incorporating it into the calculation makes the vibration energy prediction more closely reflect the actual physical nature of the rolling process, thereby improving the accuracy and reliability of the warning value set based on this vibration energy. Furthermore, by accurately predicting vibration energy and setting instability warning values ​​accordingly, this application allows operators sufficient time for vibration suppression and control, effectively avoiding wear and damage to production equipment and reducing enterprise maintenance costs. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0040] Figure 1 This is a schematic flowchart of a method for setting an instability warning value for a cold rolling mill based on vibration energy, provided in an embodiment of this application.

[0041] Figure 2 This is a schematic diagram of the data preprocessing process provided in the embodiments of this application;

[0042] Figure 3 This is a schematic diagram of the process for obtaining the forming power of a rolled piece according to an embodiment of this application;

[0043] Figure 4 This is a schematic diagram of the horizontal velocity distribution in the deformation zone provided in an embodiment of this application;

[0044] Figure 5 This is a schematic diagram of the data dimensionality reduction process provided in the embodiments of this application;

[0045] Figure 6 This is a schematic diagram of the maximum mutual information coefficient between each initially selected input feature and vibration energy provided in the embodiments of this application;

[0046] Figure 7 This is a schematic diagram of the maximum mutual information coefficient among the initially selected input features provided in the embodiments of this application;

[0047] Figure 8 This is a performance comparison chart of the model before and after feature dimensionality reduction provided in the embodiments of this application;

[0048] Figure 9 This is a flowchart of the training process for the vibration energy prediction model provided in the embodiments of this application;

[0049] Figure 10 This is a schematic diagram illustrating the impact of the number of nodes on model performance, as provided in the embodiments of this application.

[0050] Figure 11 This is a schematic diagram illustrating the impact of the number of hidden layers on model performance, provided in an embodiment of this application.

[0051] Figure 12 This is a schematic diagram illustrating the impact of time step on model performance provided in the embodiments of this application;

[0052] Figure 13 This is a schematic diagram illustrating the impact of retention rate on model performance provided in the embodiments of this application;

[0053] Figure 14 This is a schematic diagram showing the comparison and analysis of predicted and measured values ​​of RF provided in the embodiments of this application;

[0054] Figure 15 This is a schematic diagram illustrating the comparison and analysis of ANN predicted values ​​and measured values ​​provided in the embodiments of this application;

[0055] Figure 16 This is a schematic diagram showing the comparison and analysis of GA-SVR predicted values ​​and measured values ​​provided in the embodiments of this application;

[0056] Figure 17 This is a schematic diagram illustrating the comparison and analysis of RNN predicted values ​​and measured values ​​provided in the embodiments of this application;

[0057] Figure 18 This is a schematic diagram illustrating the comparison and analysis between Att-BiLSTM predicted values ​​and measured values ​​provided in the embodiments of this application;

[0058] Figure 19 This is a schematic diagram comparing the performance of five models provided in the embodiments of this application;

[0059] Figure 20 This is a schematic diagram of a device for setting an instability warning value for a cold rolling mill based on vibration energy, provided in an embodiment of this application.

[0060] Figure 21 This is a schematic diagram of the structure of a device for setting an instability warning value for a cold rolling mill based on vibration energy, provided in another embodiment of this application;

[0061] Figure 22 This is a schematic diagram of the forming power acquisition unit provided in an embodiment of this application;

[0062] Figure 23 This is a schematic diagram of the structure of a device for setting an instability warning value for a cold rolling mill based on vibration energy, provided in another embodiment of this application;

[0063] Figure 24 This is a schematic diagram of the electronic device provided in the embodiments of this application. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the embodiments of this application will be further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments and descriptions of this application are used to explain this application, but are not intended to limit this application.

[0065] like Figure 1 The diagram shown is a flowchart illustrating a method for setting an instability warning value for a cold rolling mill based on vibration energy, according to an embodiment of this application. The method includes the following steps:

[0066] Step S101: Obtain the actual operating parameters during the operation of the cold rolling mill.

[0067] This step is the data acquisition phase, which requires comprehensively collecting various operating parameters of the cold rolling mill during operation. These operating parameters can be obtained from the cold rolling production line or indirectly through relevant calculations.

[0068] Step S102: Obtain the corresponding rolling power based on the actual operating parameters.

[0069] The objective is to calculate the power of the rolled piece during the forming process based on the actual operating parameters collected in step S101. In this embodiment, the rolling process can be analyzed from an energy perspective by establishing a theoretical model based on a rolling mechanism, and then the forming power of the rolled piece, which is closely related to vibration, can be calculated.

[0070] Step S103: Call the vibration energy prediction model to process the actual operating parameters and the corresponding rolling power to obtain the predicted vibration energy of the cold rolling mill. The vibration energy prediction model is trained from a training dataset consisting of the historical operating parameters of the cold rolling mill and the corresponding rolling power.

[0071] This step involves using a machine learning model for prediction. The real-time acquired, preprocessed, and feature-filtered actual operating parameters and the calculated forming power of the rolled piece are input into the trained Att-BiLSTM model, which will output the predicted vibration energy value of the cold rolling mill.

[0072] Step S104: Set the instability warning value of the cold rolling mill based on the predicted vibration energy.

[0073] In this embodiment, a percentile threshold can be set, and then the cold rolling mill instability warning value can be set based on the predicted vibration energy and the set percentile threshold. For example, if the percentile threshold is 80%, the cold rolling mill instability warning value is set to 80% of the predicted vibration energy. When the predicted vibration energy changes, the instability warning value will also change accordingly, but the set percentile threshold will not change.

[0074] As can be seen from the above technical solution, the method for setting the instability warning value of a cold rolling mill based on vibration energy provided in this application calculates the forming power of the rolled piece using actual operating parameters and uses it together with the actual operating parameters as input to the vibration energy prediction model. This allows the prediction of vibration energy to be closer to the actual physical nature of the rolling process, thereby improving the accuracy and reliability of the warning value set based on the vibration energy. In addition, by accurately predicting vibration energy and setting the instability warning value accordingly, this application can provide operators with sufficient time for vibration suppression and control, effectively avoiding wear and damage to production equipment and reducing enterprise maintenance costs.

[0075] In one embodiment of this application, such as Figure 2 As shown, after step S101, the method further includes:

[0076] Step S201: Match the actual operating parameters obtained from different systems in real time.

[0077] The actual operating parameters in step S201 above may include: vibration parameters, process parameters, workpiece parameters, and roll parameters. Among them,

[0078] Vibration parameters include: vibration acceleration signal of the work roll;

[0079] Process parameters include: rolling force, motor torque, rolling speed, workpiece inlet / outlet speed, tensile stress at the front / rear edge of the workpiece, emulsion flow rate and concentration, actual roll gap value and its rate of change, and deformation zone length;

[0080] The parameters of the rolled product include: the grade of the rolled product, the thickness and width of the incoming rolled product, and the thickness at the inlet / outlet of the rolled product;

[0081] Roll parameters include: roll radius, work roll roughness, and rolling length.

[0082] Of the above operating parameters, the actual roll gap change rate can be obtained by differentiating the actual roll gap value, and all other parameters can be obtained from the cold rolling production line.

[0083] On a cold rolling production line, various parameters may be acquired by different sensors or data acquisition systems. These systems may have slight differences or asynchrony in their data acquisition frequencies and timestamps. Directly using this temporally misaligned data for analysis and modeling will result in input data that does not match actual operating conditions, thus affecting the accuracy of the model. Therefore, time-matching is a crucial step in ensuring that all input data remains consistent across the time dimension.

[0084] In one embodiment of this application, process parameter acquisition and vibration parameter acquisition belong to two separate systems, and the time difference between the two systems will cause a time lag between the data. The inventors discovered that when the workpiece passes through the weld, the vibration acceleration signal of the work roll increases significantly, far exceeding the acceleration signal during stable rolling. Therefore, this embodiment uses the moment when the workpiece passes through the weld in the process parameters and the moment when the acceleration signal abnormally increases in the vibration parameters as the basis for time alignment, and aligns the process parameters and the vibration parameters in time.

[0085] Specifically, this embodiment identifies and matches the occurrence times of the common event "weld passage" recorded in each of the two systems. Using these two moments as reference points, the time axes of the entire process parameter data stream and vibration parameter data stream are aligned. For example, the inherent average time difference between the two systems can be calculated, and then the data from one of the systems can be shifted over time to eliminate this lag.

[0086] This time-alignment method utilizes a well-defined and easily detectable physical event during the rolling process, resulting in high reliability. It solves the common synchronization problems encountered in multi-system data fusion by leveraging information inherent in existing data, without requiring additional synchronization equipment or complex clock synchronization protocols.

[0087] Step S202: Calculate the Mahalanobis distance of each data point from the center point, and determine outliers based on the chi-square distribution.

[0088] The actual collected data may contain outliers caused by sensor malfunctions, signal interference, unexpected events, and other reasons. These outliers deviate significantly from the normal distribution range of the data. If used directly for model training or analysis, they will distort the model's parameters and reduce its generalization ability and prediction accuracy. Therefore, it is necessary to effectively identify and handle these outliers.

[0089] Mahalanobis distance is a distance metric that takes into account the covariance of the data. It effectively measures the distance between a data point and the center of the data distribution, and is unaffected by the data scale. Simply put, it tells you how far a point deviates from the centroid of the data group, while also considering the correlations between different dimensions of the data.

[0090] When data approximately follows a multivariate normal distribution, the squared Mahalanobis distance approximately follows a chi-square distribution. Using this property, a confidence level (e.g., 95% or 99%) can be set, and a corresponding threshold can be found based on the chi-square distribution. The Mahalanobis distance for each data point is calculated; if the Mahalanobis distance of a data point exceeds the preset chi-square distribution threshold, then that data point is considered an outlier. Compared to simple standard deviation-based methods, Mahalanobis distance handles outliers in multivariate data better because it considers the correlation between variables.

[0091] Step S203: Fill in the missing values ​​and outliers using third-order Lagrange polynomial interpolation.

[0092] The collected operational parameter data may contain not only outliers but also missing values, meaning that data for certain time points or parameters were not collected. Furthermore, the outliers identified in step S202 are typically considered as data to be processed. To ensure the integrity of the dataset, prevent excessive information loss, and enable the data to be successfully input into subsequent models, these missing values ​​and identified outliers need to be appropriately imputed.

[0093] Lagrange polynomial interpolation is a classic polynomial interpolation method. Given n+1 data points (x0, y0), (x1, y1), ..., (x... n Lagrange interpolation can construct a unique polynomial of order no more than n, such that the polynomial passes through all these data points.

[0094] Specifically, in this embodiment, the calculation formula is as follows:

[0095]

[0096] Where, x i For the calculated substitution and filler values, x i-1 and x i-2 x represents the first two values ​​of outliers and missing values. i+1 and x i+2 These are the last two values ​​for outliers and missing values.

[0097] In another embodiment of this application, such as Figure 3 As shown, the step S102 above, which involves obtaining the corresponding rolling power based on actual operating parameters, may further include:

[0098] Step S301: Construct an exponential velocity field and strain rate field that satisfy the volume invariance condition and velocity boundary condition based on the actual operating parameters, and obtain the functional expression of the forming power of the rolled piece.

[0099] This step establishes a mathematical model for the metal flow within the rolling deformation zone. The constant volume condition is a fundamental assumption of metal plastic processing, meaning that the metal maintains a constant volume (incompressibility) during plastic deformation. This condition constrains the construction of the velocity field. Velocity boundary conditions define the motion state of the material at the boundaries of the deformation zone. Examples include the velocity of the workpiece entering the roll, the velocity of the workpiece leaving the roll, and the velocity relationship between the workpiece and the contact surface with the roll.

[0100] like Figure 4 The diagram shown is a schematic representation of the horizontal velocity distribution in the deformation zone provided in this embodiment, where α b α n and α f These represent the contact angles at arbitrary vertical interfaces in the backward slip zone, the neutral surface of the workpiece, and the forward slip zone, respectively, in rad; θ is the bite angle, in rad; R is the roll radius, in m; x is the distance from the workpiece inlet, in m; y is the distance from the rolling line, in m; v x h0 represents the horizontal velocity distribution of the rolled piece, in m / s; h1 represents the thickness of the rolled piece at the inlet, in m; h2 represents the thickness of the rolled piece at the outlet, in m; h3 represents the horizontal velocity distribution of the rolled piece at the outlet, in m. n V represents the thickness of the neutral surface of the rolled piece, in meters (m). r is the rolling speed, in m / s; l is the length of the deformation zone.

[0101] by Figure 4 Based on the schematic diagram of the horizontal velocity distribution in the deformation zone, this embodiment constructs an exponential velocity field and a strain rate field according to the following assumptions:

[0102] (1) From the entrance to the exit of the deformation zone, the horizontal velocity of the rolled piece surface increases exponentially.

[0103] (2) The horizontal velocity of the rolled piece is uniformly distributed on the vertical sections at the entrance, neutral surface and exit of the deformation zone, while the horizontal velocity of the rolled piece is exponentially distributed on other vertical sections.

[0104] (3) In the back slip zone, the horizontal acceleration of the workpiece gradually decreases from the surface to the core; in the front slip zone, the horizontal acceleration of the workpiece gradually increases from the surface to the core.

[0105] In this embodiment, the constructed exponential velocity field is shown in equation (2) below:

[0106]

[0107] In the formula:

[0108]

[0109] U = v0h0b = v1h1b = v n hn b = v r cosa n b(h1+2R-2Rcosa n (5)

[0110] Where x is the distance from the entry point of the workpiece, in meters; y is the distance from the rolling line, in meters; v x The horizontal velocity distribution of the rolled piece, in m / s; v y The vertical velocity distribution of the rolled piece is expressed in m / s; U is the volumetric flow rate of the rolled piece per second, expressed in m³ / s. 3 h x h0 is the thickness of any vertical section of the rolled piece, in meters; h1 is the thickness at the entry point of the rolled piece, in meters; h2 is the thickness at the exit point of the rolled piece, in meters; h3 is the thickness at the exit point of the rolled piece, in meters. n α represents the thickness of the neutral surface of the rolled piece, in meters (m). n The neutral angle, measured in rad; v n y is the horizontal speed of the roll at the neutral plane, in m / s; b is the width of the workpiece, in m; R is the roll radius, in m; v r The rolling speed is expressed in m / s.

[0111] The corresponding strain rate field is constructed as shown in equation (6):

[0112]

[0113] in, The x-direction linear strain rate is expressed in seconds. The linear strain rate in the y-direction is expressed in seconds ( / s). The shear strain rate in the xy plane is expressed in seconds ( / s).

[0114] Next, this embodiment obtains the functional expression of the forming power of the rolled piece based on the exponential velocity field and the strain rate field. The forming power of the rolled piece includes: internal plastic deformation power, friction power, shear power and tension power.

[0115] In this embodiment, the expression for the internal plastic deformation power is as follows (7):

[0116]

[0117] In the formula:

[0118]

[0119] in, σ represents the internal plastic deformation power; s h represents the deformation resistance of the rolled piece, in MPa. m The average thickness of the rolled piece in the deformation zone is expressed in meters (m).

[0120] In this embodiment, the friction power expression is as follows (10):

[0121]

[0122] In the formula:

[0123]

[0124]

[0125] in, denoted as friction power; m as friction coefficient; and k as shear strength of the rolled piece, in MPa.

[0126] In this embodiment, the obtained shear power expression is as follows (16):

[0127]

[0128] in, v is the shear power; y | x＝0 v represents the discontinuity in the rolling speed at the inlet section, in m / s. y | x＝l This represents the discontinuity in the speed of the rolled piece at the exit section, expressed in m / s.

[0129] In this embodiment, the tension power expression is as follows (17):

[0130]

[0131] in, For tension power; σ b σ represents the tension of the rolled piece in MPa. f V represents the tension of the workpiece before rolling, in MPa. x | x＝0 v is the horizontal velocity of the workpiece at the inlet section, in m / s. x | x＝l The horizontal velocity of the rolled piece at the exit section is expressed in m / s.

[0132] Step S302: Based on the functional expression of the forming power of the rolled piece, the neutral angle that minimizes the power functional is obtained, and the corresponding forming power of the rolled piece is obtained.

[0133] In one embodiment of this application, this step can employ the Newton-Raphson iteration method, using the functional expression of the forming power of the rolled piece as the objective function, and then solving for the neutral angle that minimizes the objective function.

[0134] In this embodiment, the partial derivative of the power functional with respect to the neutral angle αn is taken and equal to 0. Then, the neutral angle that minimizes the power functional is solved by Newton's iteration method. The expression is shown in the following equation (18):

[0135]

[0136] Finally, the internal plastic deformation power, friction power, shear power, and tension power at this neutral angle are calculated.

[0137] The advantage of this method is that it provides a way to estimate the forming power of the rolled piece without relying on direct measurement, but rather based on physical principles and mathematical derivation. This calculated forming power will then serve as an important input feature for the vibration energy prediction model in step S103.

[0138] In another embodiment of this application, such as Figure 5 As shown, before calling the vibration energy prediction model to process the actual operating parameters and the corresponding rolling power in step S103 above, the method further includes:

[0139] Step S501: Use the process parameters, the workpiece parameters, the roll parameters, and the workpiece forming power as initial input features.

[0140] For example, in this embodiment, the initial input features include 13 dimensions of process parameters, 4 dimensions of workpiece parameters, 3 dimensions of roll parameters, and the calculated internal plastic deformation power, shear power, and tension power, totaling 24 dimensions, as shown in Table 1. Among them, the deformation resistance is determined by the workpiece grade.

[0141] Table 1

[0142]

[0143]

[0144] Step S502: Calculate the first maximum mutual information coefficient between the initially selected input features and the vibration energy.

[0145] This step aims to evaluate the correlation or predictive ability between each initially selected input feature and the final predicted target vibration energy, which can be calculated from the aforementioned vibration parameters, i.e., vibration acceleration. This application seeks to select the features most closely related to the vibration energy.

[0146] For each initially selected input feature obtained in step S501, calculate its MIC value with the target variable vibration energy. For example, calculate the MIC between rolling force and vibration energy, the MIC between rolling forming power and vibration energy, and so on.

[0147] The second maximum mutual information coefficient obtained should refer to the set of MIC values ​​between each initially selected input feature and the target variable vibration energy obtained in this calculation process.

[0148] Step S503: Calculate the first maximum mutual information coefficient between the initially selected input features.

[0149] This is to assess the interdependence or redundancy among the initial input features. If two input features are highly correlated, they may carry similar information; retaining both may increase model complexity without significantly improving performance, and could even introduce collinearity. The Maximum Information Coefficient (MIC) is a method for measuring the strength of the association between two variables, capturing both linear and nonlinear relationships well. Its value range is typically between 0 and 1, with larger values ​​indicating a stronger association.

[0150] For the initial input feature set obtained in step S501, calculate the MIC value between any two different features. For example, calculate the MIC between rolling force and rolling speed, the MIC between rolling speed and motor torque, and so on.

[0151] The first maximum mutual information coefficient obtained refers to the set of MIC values ​​between features obtained in this calculation process.

[0152] Step S504: Perform data dimensionality reduction based on the first maximum mutual information coefficient and the second maximum mutual information coefficient to obtain the final input features of the vibration energy prediction model.

[0153] This step is based on the MIC values ​​calculated in the previous two steps. It performs feature selection to remove unimportant or redundant features, resulting in a more concise and efficient feature subset.

[0154] In one embodiment of this application, this step may further include: removing initial input features whose first maximum mutual information coefficient is less than a first preset threshold, and removing initial input features whose second maximum mutual information coefficient is greater than a second preset threshold.

[0155] The initial selection of input features, which excludes those with a maximum mutual information coefficient less than a first preset threshold, aims to eliminate features with low correlation to vibration energy and insufficient predictive ability. Only features that exhibit a sufficiently strong correlation with vibration energy are retained for subsequent modeling. This helps reduce noise and improve the model's predictive efficiency and generalization ability.

[0156] The purpose of removing initial input features whose second maximum mutual information coefficient exceeds a second preset threshold is to reduce multicollinearity and information redundancy among input features. If two or more input features are highly correlated, they actually provide similar information. Retaining all these redundant features unnecessarily increases model complexity, potentially leading to model instability, overfitting, and reduced interpretability. By removing one or more of these highly correlated features, a more concise and robust feature set can be obtained.

[0157] Taking the data in Table 1 above as an example, according to the MIC calculation method, the MIC between each initially selected input feature and the vibration energy is first calculated, and the results are as follows: Figure 6 As shown, features with a MIC less than 0.06 are then removed; next, the MIC between each initially selected input feature is calculated, and the results are as follows. Figure 7 As shown, features with a MIC greater than 0.9 are then removed.

[0158] In the experiment, this embodiment trained the model 30 times before and after feature selection, and the performance comparison is as follows. Figure 8 As shown, feature filtering can not only reduce the dimensionality of input features and thus reduce computation time, but also improve the correlation coefficient R and determination coefficient R of the model. 2 This also helps to better avoid the impact of random initialization of model parameters on the stability of calculation results.

[0159] In another embodiment of this application, the vibration energy prediction model in step S103 above employs an Att-BiLSTM network model. The method described above in this application may further include a training process for the vibration energy prediction model, such as... Figure 9 The process includes the following steps:

[0160] Step S901: Obtain the historical operating parameters of the cold rolling mill, the corresponding rolling power, and the corresponding historical target vibration energy, which together constitute the training dataset and the test dataset.

[0161] Historical operating parameters are data recorded during the actual operation of the cold rolling mill, covering process parameters, workpiece parameters, roll parameters, etc. (as mentioned in step S101). These are the input features of the model. Based on the historical operating parameters, the workpiece forming power is calculated using the theoretical calculation methods in steps S301 and S302. The corresponding historical target vibration energy is the vibration energy value that actually occurred, was measured, or calculated under the same time period or operating conditions as the aforementioned historical operating parameters and workpiece forming power. This is the target variable (label) that the model needs to learn and predict.

[0162] The training dataset is used to train the model, allowing it to learn the mapping relationship between input features and target vibrational energy. The test dataset is used to evaluate the performance of the trained model on unseen data, testing the model's generalization ability. Typically, the training and test sets do not overlap.

[0163] Step S902: Train the vibration energy prediction model using the training dataset, and dynamically adjust the learning rate during the training process using a learning rate warm-up and cosine annealing strategy.

[0164] By adjusting the internal parameters (weights and biases) of the Att-BiLSTM model using the training dataset, it is able to learn patterns for predicting vibrational energy from the input features.

[0165] The input features (filtered operating parameters and rolling power) and the corresponding target vibration energy from the training dataset are fed into the Att-BiLSTM model. The model performs forward propagation based on its internal structure (Attention mechanism, bidirectional LSTM layers, etc.) to obtain a predicted vibration energy. By comparing the predicted value with the actual target vibration energy, a loss function (e.g., mean squared error, MSE) is calculated. Then, the model weights are adjusted using a backpropagation algorithm and an optimizer (e.g., Adam, SGD) to minimize the loss function.

[0166] The dynamic learning rate is a key hyperparameter controlling the magnitude of model parameter updates. A suitable learning rate strategy can help the model converge to the optimal solution faster and more stably. In this embodiment, a learning rate warm-up and cosine annealing strategy is used to dynamically adjust the learning rate.

[0167] Step S903: Continuously optimize the parameters of the vibration energy prediction model through iterative training on the training dataset, and periodically evaluate the performance of the vibration energy prediction model using the test dataset until the vibration energy prediction model can accurately predict vibration energy based on the input operating parameters and rolling power. The parameters include at least the number of Att-BiLSTM network nodes, the number of hidden layers, the time step, and the retention rate.

[0168] Through repeated training and validation, the optimal structure and parameter configuration of the Att-BiLSTM model are found, ensuring that the model achieves the expected prediction accuracy. The model is trained on the training dataset in multiple epochs. In each epoch, the model processes the entire training dataset or a batch of data and updates its weights. This process is continuous, and the model parameters are optimized after each iteration, gradually reducing the loss function.

[0169] During training, instead of waiting until all iterations are complete to evaluate the model, its performance is evaluated periodically (e.g., every few epochs) on the test dataset. Evaluation metrics may include mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R²). 2 (e.g., ) are used to measure the accuracy of model predictions.

[0170] If a model performs well on the training set but performs poorly or worse on the test set, it indicates that the model may be overfitting. Early stopping strategies can be implemented accordingly. The training process continues until a certain stopping condition is met, such as reaching a preset maximum number of iterations, the model's performance on the test set no longer improving or improving very little, or the model's prediction accuracy on the test set reaching a preset target.

[0171] In this embodiment, a grid search method can be used to optimize parameters and determine the number of Att-BiLSTM network nodes, the number of hidden layers, the time step, and the retention rate. Specifically, this embodiment selects four hyperparameters: the number of Att-BiLSTM network nodes, the number of hidden layers, the time step, and the retention rate.

[0172] In this embodiment, the number of nodes in the Att-BiLSTM network ranges from [50, 60, 70, 80, 90], the number of hidden layers ranges from [1, 2, 3], the time step ranges from [5, 10, 20, 25, 50], and the retention rate ranges from [0.55, 0.65, 0.75, 0.85, 0.95]. The average performance of the model is obtained by calculating each set of model parameters 30 times, and the results are as follows: Figures 10 to 13 As shown in the figure. The calculation results show that a more complex network structure, a larger time step, and a higher retention rate do not necessarily lead to better model performance. After optimization using the grid search method, the optimal hyperparameters were finally determined to be 70 nodes, 2 hidden layers, a time step of 20, and a retention rate of 0.75.

[0173] The following section further explains the beneficial effects of using the Att-BiLSTM network model in this application:

[0174] In this embodiment, actual production data from 12 coils of strip steel (12,000 data points in total) under different rolling conditions were acquired, including 4 coils (4,000 data points) containing vibration processes. One coil (2,000 data points) containing vibration processes and one coil (2,000 data points) containing only steady-state rolling processes were selected as the test set to evaluate the algorithm's prediction performance. The remaining strip steel data were used as the training set. The model hyperparameters were determined as described above: 70 nodes, 2 hidden layers, 20 time steps, and a retention rate of 0.75. The learning rate reached its maximum value of 0.015 in the 100th iteration, with a total of 1500 iterations. The vibration energy prediction results of five models—RF, ANN, GA-SVR, RNN, and Att-BiLSTM—were then compared.

[0175] The number of predictors in the RF model was determined to be 200 through iterative calculations. The kernel function for the SVR was RBF, and the parameters were optimized using GA. The initial population size and maximum number of generations for GA were 50 and 200, respectively. The optimal parameters for the SVR were found to be: C = 0.34, gamma = 63.56, and epsilon = 0.08. The parameter settings for the ANN and RNN were the same as those for the Att-BiLSTM model. The comparative analysis results of the vibration energy predictions and measured values ​​for the five models are as follows: Figures 14 to 18 As shown.

[0176] contrast Figures 14-18 As shown in (a), ANN performs the worst overall; RF and GA-SVR show improved performance, but their performance is not ideal during the energy decay stage (rolling time of 20-30s); RNN and Att-BiLSTM have higher prediction accuracy during the energy decay stage, with Att-BiLSTM showing the best performance. (Comparison) Figures 14-18 The scatter plot and histogram of the error distribution in (b) show that the error distributions of the five models are basically in line with a normal distribution; the error distributions of ANN and GA-SVR are relatively scattered, with more scatter points having an absolute error exceeding ±0.5; the errors of RF, RNN and Att-BiLSTM are smaller and more concentrated, with Att-BiLSTM having the smallest error and a higher frequency and smaller variance near 0, indicating that the prediction accuracy of Att-BiLSTM is higher than that of the other four models.

[0177] To more comprehensively evaluate model performance, this embodiment also compares the MAE, RMSE, and MAPE of five models, and the results are as follows: Figure 19 As shown, ANN and GA-SVR have the highest errors, while RNN falls between RF and Att-BiLSTM. Att-BiLSTM has a slightly higher MAE than RF, but its RMSE and MAPE are both lower than RF. Therefore, Att-BiLSTM has better overall performance.

[0178] Table 2 shows a comparison of the training and prediction times for the five models mentioned above. The training and prediction times for RF and GA-SVR are longer than those for ANN, RNN, and Att-BiLSTM. Since the response time of field-based online control systems is often in the millisecond range, RF and GA-SVR are not practically applicable.

[0179] Table 2

[0180] RF ANN GA-SVR RNN Att-BiLSTM Training time / s 858.436 10.196 <![CDATA[2.432×10 4 ]]> 41.512 83.858 Prediction time / s 34.898 <![CDATA[1.198×10 -3 ]]> 13.608 <![CDATA[2.399×10 -3 ]]> <![CDATA[3.541×10 -3 ]]>

[0181] Based on the error analysis of the above models, the cold rolling mill vibration energy prediction method proposed in this invention, which combines the coupled mechanism model and the Att-BiLSTM model, has good overall performance.

[0182] like Figure 20 The diagram shows a schematic representation of a device for setting an instability warning value for a cold rolling mill based on vibration energy, according to an embodiment of this application. The device includes: an operating parameter acquisition unit 2010, a forming power acquisition unit 2020, a model recall unit 2030, and a warning value setting unit 2040, which are connected sequentially.

[0183] The operating parameter acquisition unit 2010 is used to acquire the actual operating parameters during the operation of the cold rolling mill.

[0184] The forming power acquisition unit 2020 is used to acquire the corresponding forming power of the rolled piece based on the actual operating parameters.

[0185] The model calling unit 2030 is used to call the vibration energy prediction model to process the actual operating parameters and the corresponding rolling power to obtain the predicted vibration energy of the cold rolling mill. The vibration energy prediction model is trained from a training dataset consisting of the historical operating parameters of the cold rolling mill and the corresponding rolling power.

[0186] The warning value setting unit 2040 is used to set the cold rolling mill instability warning value based on the predicted vibration energy.

[0187] As one embodiment of this application, such as Figure 21 As shown, the above-mentioned device also includes:

[0188] The timing matching unit 2050 is used to time-match the actual operating parameters obtained from different systems.

[0189] The outlier detection unit 2060 is used to calculate the Mahalanobis distance of each data point from the center point and to determine outliers based on the chi-square distribution.

[0190] The filling unit 2070 is used to fill in missing values ​​and outliers using third-order Lagrange polynomial interpolation.

[0191] As an embodiment of this application, the above-mentioned actual operating parameters and the historical operating parameters include: vibration parameters, process parameters, workpiece parameters and roll parameters.

[0192] As an embodiment of this application, the aforementioned timing matching unit 2050 is specifically used to: align the process parameters and the vibration parameters in time, wherein the time alignment is based on the time when the rolled piece passes through the weld in the process parameters and the time when the acceleration signal abnormally increases in the vibration parameters.

[0193] As one embodiment of this application, such as Figure 22 As shown, the forming power acquisition unit 2020 includes:

[0194] The functional expression acquisition module 2021 is used to construct an exponential velocity field and a strain rate field that satisfy the volume invariance condition and velocity boundary condition based on the actual operating parameters, and obtain the functional expression of the forming power of the rolled piece.

[0195] The functional solver module 2022 is used to solve for the neutral angle that minimizes the power functional based on the functional expression of the rolling forming power, and to obtain the corresponding rolling forming power.

[0196] As an embodiment of this application, the forming power of the rolled piece includes: internal plastic deformation power, friction power, shearing power and tension power.

[0197] As an embodiment of this application, the functional solving module 2022 is specifically used to: use the Newton iteration method, with the functional expression of the forming power of the rolled piece as the objective function, to solve for the neutral angle that minimizes the objective function.

[0198] As one embodiment of this application, such as Figure 23 As shown, the above-mentioned device also includes a data dimensionality reduction unit 2080, which is used to: take the process parameters, the workpiece parameters, the roll parameters and the workpiece forming power as initial input features; calculate the first maximum mutual information coefficient between the initial input features and the vibration energy; calculate the second maximum mutual information coefficient between the initial input features; and perform data dimensionality reduction based on the first maximum mutual information coefficient and the second maximum mutual information coefficient to obtain the final input features of the vibration energy prediction model.

[0199] As an embodiment of this application, the data dimensionality reduction unit 2080 performs data dimensionality reduction based on the first maximum mutual information coefficient and the second maximum mutual information coefficient, including: removing initial input features whose first maximum mutual information coefficient is less than a first preset threshold, and removing initial input features whose second maximum mutual information coefficient is greater than a second preset threshold.

[0200] As an embodiment of this application, the above-mentioned vibration energy prediction model adopts the Att-BiLSTM network model.

[0201] As an embodiment of this application, the above-mentioned device further includes a model training unit, used for: acquiring historical operating parameters of the cold rolling mill, the corresponding forming power of the rolled piece, and the corresponding historical target vibration energy, which together constitute a training dataset and a test dataset; training the vibration energy prediction model using the training dataset, and dynamically adjusting the learning rate during the training process using a learning rate preheating and cosine annealing strategy; continuously optimizing the parameters of the vibration energy prediction model through iterative training on the training dataset, and periodically evaluating the performance of the vibration energy prediction model using the test dataset, until the vibration energy prediction model can accurately predict vibration energy based on the input operating parameters and forming power of the rolled piece, wherein the parameters include at least the number of Att-BiLSTM network nodes, the number of hidden layers, the time step, and the retention rate.

[0202] As an embodiment of this application, the aforementioned warning value setting unit 2040 is specifically used to: set a cold rolling mill instability warning value based on the vibration energy and a preset percentile threshold.

[0203] As can be seen from the above technical solution, the device for setting instability warning values ​​for cold rolling mills based on vibration energy provided in this application calculates the forming power of the rolled piece using actual operating parameters, and uses this power, along with the actual operating parameters, as input to the vibration energy prediction model. The forming power of the rolled piece is a key physical quantity affecting the stability of the rolling process; incorporating it into the calculation makes the vibration energy prediction more closely reflect the actual physical nature of the rolling process, thereby improving the accuracy and reliability of the warning value set based on this vibration energy. Furthermore, by accurately predicting vibration energy and setting instability warning values ​​accordingly, this application allows operators sufficient time for vibration suppression and control, effectively avoiding wear and damage to production equipment and reducing enterprise maintenance costs.

[0204] Figure 24 This is a schematic diagram of the electronic device provided in the embodiments of this application. Figure 24 The illustrated electronic device is a general-purpose data processing apparatus, comprising a general-purpose computer hardware structure, including at least a processor 801 and a memory 802. The processor 801 and memory 802 are connected via a bus 803. The memory 802 is adapted to store one or more instructions or programs executable by the processor 801. These instructions or programs are executed by the processor 801 to implement the steps in the aforementioned method for setting the instability warning value of a cold rolling mill based on vibration energy.

[0205] The processor 801 described above can be a standalone microprocessor or a collection of one or more microprocessors. Thus, the processor 801 executes commands stored in the memory 802, thereby performing the method flow described in the embodiments of this application to process data and control other devices. The bus 803 connects the aforementioned components together, and also connects these components to the display controller 804, the display device, and the input / output (I / O) device 805. The input / output (I / O) device 805 can be a mouse, keyboard, modem, network interface, touch input device, motion-sensing input device, printer, and other devices known in the art. Typically, the input / output (I / O) device 805 is connected to the system via an input / output (I / O) controller 806.

[0206] The memory 802 can store software components, such as an operating system, a communication module, an interaction module, and application programs. Each of the modules and application programs described above corresponds to a set of executable program instructions that perform one or more functions and the methods described in the embodiments of the invention.

[0207] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described method for setting the instability warning value of a cold rolling mill based on vibration energy.

[0208] This application also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the above-described method for setting the instability warning value of a cold rolling mill based on vibration energy.

[0209] Preferred embodiments of this application have been described above with reference to the accompanying drawings. Many features and advantages of these embodiments are apparent from this detailed description, and therefore the claims are intended to cover all such features and advantages that fall within the true spirit and scope of these embodiments. Furthermore, since many modifications and alterations will readily occur to those skilled in the art, the embodiments of this application are not intended to be limited to the precise structures and operations illustrated and described, but rather to encompass all suitable modifications and equivalents falling within their scope.

[0210] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0211] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0212] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0213] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0214] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above descriptions are merely specific embodiments of this application and are not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for setting an instability early warning value for a cold rolling mill based on vibration energy, characterized in that, The method includes: Obtain the actual operating parameters during the operation of the cold rolling mill; The corresponding rolling power is obtained based on the actual operating parameters. The vibration energy prediction model is called to process the actual operating parameters and the corresponding rolling power to obtain the predicted vibration energy of the cold rolling mill. The vibration energy prediction model is trained from a training dataset consisting of the historical operating parameters of the cold rolling mill and the corresponding rolling power. The instability warning value of the cold rolling mill is set based on the predicted vibration energy.

2. The method for setting the instability early warning value of a cold rolling mill based on vibration energy as described in claim 1, characterized in that, After obtaining the actual operating parameters during the operation of the cold rolling mill, the method further includes: Match the actual operating parameters obtained from different systems in real time; Calculate the Mahalanobis distance of each data point from the center point, and identify outliers based on the chi-square distribution; Third-order Lagrange polynomial interpolation was used to fill in the missing values ​​and the outliers.

3. The method for setting the instability early warning value of a cold rolling mill based on vibration energy as described in claim 2, characterized in that, The actual operating parameters and the historical operating parameters include: vibration parameters, process parameters, workpiece parameters, and roll parameters.

4. The method for setting the instability early warning value of a cold rolling mill based on vibration energy as described in claim 3, characterized in that, The step of matching the actual operating parameters obtained from different systems in real time includes: The process parameters and the vibration parameters are time-aligned, wherein the time alignment is based on the moment when the rolled piece passes through the weld in the process parameters and the moment when the acceleration signal abnormally increases in the vibration parameters.

5. The method for setting the instability early warning value of a cold rolling mill based on vibration energy as described in claim 1, characterized in that, The process of obtaining the corresponding rolling power based on the actual operating parameters includes: Based on the actual operating parameters, an exponential velocity field and a strain rate field that satisfy the conditions of constant volume and velocity boundary conditions are constructed, and the functional expression of the forming power of the rolled piece is obtained. Based on the functional expression of the forming power of the rolled piece, the neutral angle that minimizes the power functional is obtained, and the corresponding forming power of the rolled piece is obtained.

6. The method for setting the instability early warning value of a cold rolling mill based on vibration energy as described in claim 5, characterized in that, The forming power of the rolled piece includes: internal plastic deformation power, friction power, shearing power, and tension power.

7. The method for setting the instability early warning value of a cold rolling mill based on vibration energy as described in claim 5, characterized in that, The neutral angles that minimize the power functional based on the power functional expression of the rolled product include: Using the Newton-Raphson iteration method, with the functional expression of the forming power of the rolled piece as the objective function, the neutral angle that minimizes the objective function is found.

8. The method for setting the instability early warning value of a cold rolling mill based on vibration energy as described in claim 3, characterized in that, Before calling the vibration energy prediction model to process the actual operating parameters and the corresponding rolling power, the method further includes: The process parameters, the workpiece parameters, the roll parameters, and the workpiece forming power are used as initial input features. Calculate the first maximum mutual information coefficient between the initially selected input features and the vibration energy; Calculate the second maximum mutual information coefficient between the initially selected input features; Data dimensionality reduction is performed based on the first maximum mutual information coefficient and the second maximum mutual information coefficient to obtain the final input features of the vibration energy prediction model.

9. The method for setting the instability early warning value of a cold rolling mill based on vibration energy as described in claim 8, characterized in that, The data dimensionality reduction based on the first maximum mutual information coefficient and the second maximum mutual information coefficient includes: Initially selected input features whose first maximum mutual information coefficient is less than a first preset threshold are removed, and initial selected input features whose second maximum mutual information coefficient is greater than a second preset threshold are also removed.

10. The method for setting the instability early warning value of a cold rolling mill based on vibration energy as described in claim 1, characterized in that, The vibration energy prediction model adopts the Att-BiLSTM network model.

11. The method for setting the instability early warning value of a cold rolling mill based on vibration energy as described in claim 10, characterized in that, The method further includes: The historical operating parameters of the cold rolling mill, the corresponding forming power of the rolled piece, and the corresponding historical target vibration energy are obtained and used to form the training dataset and the test dataset. The vibration energy prediction model is trained using the training dataset, and the learning rate is dynamically adjusted during the training process using a learning rate warm-up and cosine annealing strategy. The parameters of the vibration energy prediction model are continuously optimized through iterative training on the training dataset, and the performance of the vibration energy prediction model is periodically evaluated using the test dataset until the vibration energy prediction model can accurately predict vibration energy based on the input operating parameters and the forming power of the rolled piece. The parameters of the vibration energy prediction model include at least the number of Att-BiLSTM network nodes, the number of hidden layers, the time step, and the retention rate.

12. The method for setting the instability early warning value of a cold rolling mill based on vibration energy as described in claim 1, characterized in that, The method of setting the cold rolling mill instability early warning value based on the vibration energy includes: The instability warning value of the cold rolling mill is set based on the vibration energy and the preset percentile threshold.

13. A device for setting an early warning value for instability in a cold rolling mill based on vibration energy, characterized in that, The device includes: The operating parameter acquisition unit is used to acquire the actual operating parameters during the operation of the cold rolling mill; A forming power acquisition unit is used to acquire the corresponding forming power of the rolled piece based on the actual operating parameters. The model calling unit is used to call the vibration energy prediction model to process the actual operating parameters and the corresponding rolling power to obtain the vibration energy of the cold rolling mill. The vibration energy prediction model is trained from a training dataset consisting of the historical operating parameters of the cold rolling mill and the corresponding rolling power. The warning value setting unit is used to set the instability warning value of the cold rolling mill based on the vibration energy.

14. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-12.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-12.

16. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 12.

Images