Roller state judgment method and device, equipment, storage medium and product

Abstract

The invention provides a roller state judgment method and device, equipment, a storage medium and a product. The method comprises the following steps: acquiring operation data of a target roller, wherein the operation data comprises stress data in the operation process of the target roller; inputting the operation data into a preset fatigue life prediction model to obtain the expected residual life of the target roller; the fatigue life prediction model is obtained based on training of historical rolling data of a sample roller, and the historical rolling data comprises at least one of stress data, material data, process data and use data; extracting stress data in the operation data, and inputting the stress data into a preset damage model to obtain damage information of the target roller; and determining the damage state of the target roller based on the expected residual life and the damage information. And the target roller state is judged in combination with data in multiple aspects, so that the judgment accuracy of the roller state is improved, and the maintenance requirement of the target roller is determined more accurately.

Description

Technical Field

[0001] The present invention relates to the field of metal processing technology, and more particularly to a method, device, equipment, storage medium, and product for determining the state of a rolling mill roll. Background Art

[0002] In the metalworking industry, rolls are core components of production lines, and their performance directly determines the accuracy of rolled plate shape, surface quality, and production efficiency. Over long-term use, rolls are subjected to high-frequency, cyclical stresses, which gradually lead to fatigue damage on the surface and subsurface layers, such as pitting, crack propagation, and spalling. These damages not only affect the quality of rolled products but also reduce the operational stability of the equipment.

[0003] Conventional roll maintenance relies primarily on regular roll replacement and grinding, typically determined empirically based on maintenance intervals and grinding requirements. However, manual judgment of grinding requirements or roll replacement intervals can be inaccurate, potentially leading to over-grinding or premature roll replacement, resulting in wasted resources. Consequently, existing technologies lack the ability to assess and dynamically adjust roll fatigue in real time, limiting their effectiveness in industrial applications and inaccurately assessing roll condition. Summary of the Invention

[0004] The embodiments of the present application provide a method, device, equipment, storage medium and product for determining the state of a rolling mill, which can solve the technical problem of insufficient accuracy in determining the state of a rolling mill.

[0005] In a first aspect, the present application provides a method for determining a roll state, the method comprising:

[0006] Acquiring operating data of the target roller, the operating data including force data of the target roller during operation;

[0007] Inputting the operating data into a preset fatigue life prediction model to obtain the expected remaining life of the target roll; the fatigue life prediction model is trained based on historical rolling data of the sample roll, the historical rolling data including at least one of force data, material data, process data, and usage data;

[0008] Extract the force data from the operating data and input the force data into the preset damage model to obtain the damage information of the target roll;

[0009] The damage state of the target roll is determined based on the expected remaining life and the damage information.

[0010] In some possible implementations, the operating data is input into a preset fatigue life prediction model to obtain the expected remaining life of the target roll, including:

[0011] Inputting the operating data into the first data processing layer of the preset fatigue life prediction model to extract the time series characteristics of the operating data;

[0012] The time series characteristics are input into the second data processing layer of the preset fatigue life prediction model for regression prediction to obtain the expected remaining life of the target roll.

[0013] In some possible implementations, inputting the operating data into a first data processing layer of a preset fatigue life prediction model to extract time series features of the operating data includes:

[0014] Divide the running data into multiple subsequences according to the time sequence;

[0015] Inputting the multiple subsequences into a first data processing layer in a preset fatigue life prediction model to obtain time series features corresponding to the multiple subsequences; wherein the first data processing layer processes the multiple subsequences based on a self-attention mechanism;

[0016] The time series features of the running data are obtained according to the time series features corresponding to the multiple subsequences.

[0017] In some possible implementations, the time series features are input into the second data processing layer of a preset fatigue life prediction model for regression prediction to obtain the expected remaining life of the target roll, including:

[0018] Inputting the time series features into a second data processing layer in a preset fatigue life prediction model to obtain multiple regression prediction results; wherein the second data processing layer performs regression prediction based on multiple gradient boosting trees;

[0019] Based on multiple regression prediction results, the expected remaining life of the target roll is obtained.

[0020] In some possible implementations, before inputting the operating data into a preset fatigue life prediction model to obtain the expected remaining life of the target roll, the method further includes:

[0021] A training sample set and a verification sample set are selected from a plurality of training samples, each training sample including historical rolling data of a sample roll and remaining life information corresponding to the sample roll;

[0022] Construct an initial fatigue life prediction model based on the training sample set;

[0023] Input the verification sample set into the initial fatigue life prediction model to determine the remaining life prediction results corresponding to each sample roll in the verification sample set;

[0024] Based on the remaining life information and remaining life prediction results of each sample roll in the verification sample set, a loss function is constructed;

[0025] A new fatigue life prediction model is constructed based on the loss function, and a new loss function is constructed based on the validation sample set and the new fatigue life prediction model until the preset training stop condition is met to obtain the trained fatigue life prediction model.

[0026] In some possible implementations, extracting force data from the operating data and inputting the force data into a preset damage model to obtain damage information of the target roll includes:

[0027] Extract force data from operating data;

[0028] Input the force data into a preset damage model to determine the stress data of the target roll, wherein the damage model performs physical calculations based on the force data to obtain the corresponding stress data;

[0029] The initial position and extension direction of the damage are predicted based on the stress data, and the damage information of the target roll is obtained.

[0030] In some possible implementations, the damage initial location and extension direction are predicted based on stress data to obtain damage information of the target roll, including:

[0031] Based on the stress data, determine the stress concentration area on the target roll;

[0032] Determine whether each location point in the stress concentration area has reached the critical condition for damage formation;

[0033] According to the position point where the critical condition for damage formation is reached, the initial position and expansion direction of the damage are predicted as the damage information of the target roll.

[0034] In some possible implementations, obtaining the operating data of the target roll includes:

[0035] Acquire the raw data of the target roll through multiple sensors;

[0036] Eliminating outliers and / or removing noise from the original data to obtain first data;

[0037] The key characteristic parameters in the first data are extracted to obtain the operating data of the target roller within a preset time period.

[0038] In a second aspect, the present application provides a device for determining a roller state, the device comprising:

[0039] An acquisition module is used to acquire the operating data of the target roller, wherein the operating data includes the force data of the target roller during operation;

[0040] a calculation module, configured to input the operating data into a preset fatigue life prediction model to obtain the expected remaining life of the target roll; the fatigue life prediction model is trained based on historical rolling data of the sample roll, the historical rolling data including at least one of force data, material data, process data, and usage data;

[0041] The calculation module is also used to extract the force data from the operation data and input the force data into the preset damage model to obtain the damage information of the target roll;

[0042] The determination module is used to determine the damage state of the target roll based on the expected remaining life and the damage information.

[0043] In a third aspect, the present application provides a roller state determination device, which includes: a processor, and a memory storing computer program instructions; the processor reads and executes the computer program instructions to implement the roller state determination method described above.

[0044] In a fourth aspect, the present application provides a computer-readable storage medium having computer program instructions stored thereon. When the computer program instructions are executed by a processor, the roller state determination method described above is implemented.

[0045] In a fifth aspect, the present application provides a computer program product. When the instructions in the computer program product are executed by a processor of an electronic device, the electronic device executes the roller state determination method described above.

[0046] The roller status determination method, device, equipment, storage medium and product provided in the embodiments of the present application obtain the operating data of the target roller and input it into the fatigue life prediction model to predict the remaining life; at the same time, the damage information of the target roller is obtained based on the force data, and then the remaining life prediction and damage information results are integrated to determine the health status of the roller. The target roller status is determined by combining data from multiple aspects, thereby improving the accuracy of the roller status determination and more accurately determining the maintenance requirements of the target roller. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The present application can be better understood from the following description of specific embodiments of the present application in conjunction with the accompanying drawings, in which:

[0048] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings, in which the same or similar reference numerals represent the same or similar features.

[0049] Figure 1 1 is a flow chart of a method for determining a roll state provided by an embodiment of the present application;

[0050] Figure 2 This is a partial flow chart of a method for determining the state of a roll provided by another embodiment of the present application;

[0051] Figure 3 1 is a flow chart of a method for determining a roll state provided by another embodiment of the present application;

[0052] Figure 4 This is a schematic structural diagram of a roller state determination device provided by an embodiment of the present application;

[0053] Figure 5 It is a hardware structure diagram of the roller state determination device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0054] The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, rather than to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is merely to provide a better understanding of the present application by illustrating the examples of the present application.

[0055] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, the elements defined by the phrase "comprising..." do not exclude the presence of other identical elements in the process, method, article, or device comprising the elements.

[0056] Rollers are core components of steel rolling production lines, and their performance directly impacts the shape accuracy, surface quality, and overall production efficiency of the rolled strip. However, over long periods of service, rolls are subjected to high-frequency, cyclical stresses, leading to fatigue damage on their surfaces and subsurface layers. As fatigue damage accumulates, problems such as pitting, crack propagation, and spalling may occur on the roll surface, seriously impacting the quality of rolled products and equipment stability. To maintain roll performance, traditional methods rely primarily on regular roll replacement and grinding. Specifically, after a roll has been in service for a period of time, the grinding amount and replacement interval are empirically determined. However, this empirical maintenance approach has numerous limitations. First, the determination of grinding amount and replacement interval relies on manual judgment, resulting in low accuracy and potentially over- or under-grinding, which in turn affects the actual service life of the rolls. Second, due to the lack of real-time assessment of roll fatigue, some rolls may be replaced prematurely before reaching the damage threshold, resulting in unnecessary waste of resources. In addition, the existing roll replacement and grinding plans lack data-driven support and cannot be dynamically adjusted based on the roll service environment, load changes, and historical damage trends, making it difficult to meet the requirements of modern rolling production for efficient, precise, and economical management.

[0057] In order to solve the problems in the prior art, the embodiments of the present application provide a roll state determination method, device, equipment, storage medium and product. The roll state determination method provided in the embodiments of the present application is first introduced below.

[0058] Figure 1 FIG. 1 is a flow chart of a method for determining the state of a roll provided by an embodiment of the present application. Figure 1 As shown, the method may include the following steps: S101 to S104.

[0059] S101: Acquire operating data of a target roller, where the operating data includes force data of the target roller during operation.

[0060] The target roll is a key device used in the rolling process, and its operating data includes various working status information of the equipment.

[0061] Force data refers to the numerical values ​​and distribution of forces in different directions and types applied to the roll during operation. For example, it includes data on the mechanical behavior of the roll surface and interior under the influence of compression, torsion, shear, and tension.

[0062] In practice, operating data from the target roll is collected using a series of sensors. These sensors, embedded on or around the roll surface, monitor the roll's operating status in real time, recording various physical quantities and acquiring the target roll's operating data. For example, strain gauges can be installed on the rolling mill to measure the compressive and shear stresses acting on the roll. When rolling begins, the sensors record these force data in real time.

[0063] S102: Inputting the operating data into a preset fatigue life prediction model to obtain the expected remaining life of the target roll. The fatigue life prediction model is trained based on historical rolling data of the sample roll, where the historical rolling data includes at least one of force data, material data, process data, and usage data.

[0064] The fatigue life prediction model is a model established through training based on the historical operating data of the roll. Its purpose is to calculate the expected remaining life of the roll through the existing operating data.

[0065] In specific implementations, a pre-trained model is used to analyze the target roll based on input operating data. By analyzing the roll's performance under different operating conditions, the model can determine the degree of fatigue damage and predict the roll's remaining life. For example, the model uses operating data combined with mathematical formulas and / or machine learning algorithms to calculate the expected remaining life of the roll. If the roll has been subjected to a large amount of repeated loads during past rolling operations, the predicted remaining life may be short.

[0066] S103: extracting force data from the operating data, and inputting the force data into a preset damage model to obtain damage information of the target roll.

[0067] A damage model analyzes roll damage based on information such as force data and operating conditions. By inputting real-time force data, the model can determine the extent of roll damage under current operating conditions.

[0068] In practice, real-time force data is extracted from the roll's operating data and fed into a damage model. Based on the material's external force data, the damage model calculates the likely type and extent of damage to the roll. For example, the model can estimate the likelihood of surface crack formation based on the accumulation of compressive stress. Suppose the roll is subjected to high compressive and shear forces during the rolling process. The damage model analysis might reveal slight wear on the roll surface and the development of incipient cracks within the roll.

[0069] S104: Determine the damage state of the target roll based on the expected remaining life and the damage information.

[0070] The damage status describes the degree of physical damage sustained by the target roll during current use. It includes not only the expected remaining lifespan but also the potential damage that may occur during long-term operation. By integrating these factors, it is possible to determine whether the roll is in a dangerous state and requires downtime for maintenance.

[0071] In practice, the system combines the expected remaining life with the output of the damage model to assess the overall damage state of the roll. For example, if the remaining life of a roll is short and the damage information indicates severe surface cracks or wear, the system will determine that the roll is in a "severely damaged" state and recommend repair or replacement.

[0072] The roll condition determination method provided in the embodiments of the present application obtains the operating data of a target roll and inputs it into a fatigue life prediction model to predict the remaining life. Simultaneously, damage information of the target roll is obtained based on the force data. The remaining life prediction and damage information are then integrated to determine the health status of the roll. This combined use of multiple data points improves the accuracy of roll condition determination and more accurately determines the maintenance needs of the target roll.

[0073] In order to obtain a more accurate expected remaining life, in some embodiments, S102 may include the following steps: S1021 to S1022.

[0074] S1021: Input the operating data into the first data processing layer of the preset fatigue life prediction model to extract the time series characteristics of the operating data.

[0075] In a specific implementation, after the operating data is input into a preset fatigue life prediction model, it is processed through the first data processing layer in the fatigue life prediction model to extract the time series features in the operating data.

[0076] S1022: Inputting the time series characteristics into the second data processing layer of the preset fatigue life prediction model for regression prediction to obtain the expected remaining life of the target roll.

[0077] In practice, once the time series features are extracted, they are passed to the second data processing layer of the pre-set fatigue life prediction model. This layer performs a series of calculations on the input time series features and ultimately generates a numerical output, which represents the expected remaining life of the roll.

[0078] The roll condition determination method provided in this embodiment accurately extracts meaningful features from complex time series data and analyzes these features using an appropriate regression algorithm to obtain a final prediction result. The first data processing layer extracts time series features, and then the second data processing layer uses these time series features to obtain a more accurate estimated remaining life.

[0079] In order to accurately extract the timing features, in some implementations, S1021 may include the following steps: S10211 to S10213.

[0080] S10211: Divide the running data into multiple subsequences according to the time sequence.

[0081] In the implementation, the input operating data is first segmented. Sensors collect operating data from the target roller within a certain time range, and this data is stored in a long time series in chronological order. To more effectively extract temporal features from the data, the software segments the entire dataset into multiple smaller subsequences based on time windows. For example, if the roller operating data is collected at 1000 time points, and the length of each subsequence is set to 100 time points, the data will ultimately be segmented into 10 subsequences, each containing data from 100 consecutive time points.

[0082] S10212: Input the multiple subsequences into a first data processing layer in a preset fatigue life prediction model to obtain time series features corresponding to the multiple subsequences. The first data processing layer processes the multiple subsequences based on a self-attention mechanism.

[0083] The first data processing layer uses a neural network model based on the self-attention mechanism, a technique that can capture the relationships between different parts of a sequence.

[0084] In practice, for each subsequence, the first data processing layer calculates its internal temporal features and assigns different weights to the relationships between each time point. Using a self-attention mechanism, it focuses on the most important time points in the input sequence while ignoring less important portions. For example, if the time data of a subsequence reflects changes in roller speed, the self-attention mechanism will calculate correlations across different time points to determine which speed changes are most critical for predicting remaining life. For example, abnormal vibration at a certain moment may be closely correlated with the fatigue life of the roller, while other time points may have no significant impact on the prediction.

[0085] S10213: Obtaining time series features of the operating data according to the time series features corresponding to the multiple subsequences.

[0086] In the specific implementation, the time series features of the entire operating data are obtained based on the time series features extracted from all subsequences. The features extracted from each subsequence will be used to characterize the global pattern of the data, thereby forming an overall time series feature set. This integrated time series feature will be used for further analysis and regression prediction of the subsequent fatigue life prediction model. Specifically, the time series features of each subsequence can be merged into an overall time series feature through aggregation functions (such as average, weighted sum, etc.). For example, the average feature value of each subsequence can be calculated, or each subsequence can be weighted and summarized according to its importance (calculated by the self-attention mechanism). In this way, a comprehensive time series feature can be obtained to help more accurately predict the fatigue life of the target roller.

[0087] The roll state determination method provided in this embodiment of the application uses efficient time series data segmentation and feature extraction to ensure that the software can extract key features from complex operating data, thereby providing strong support for subsequent fatigue life prediction. Time series feature extraction is performed through a self-attention mechanism to accurately extract time series features.

[0088] In order to accurately obtain the expected remaining life, in some embodiments, S1022 may include the following steps: S10221 to S10222.

[0089] S10221: Inputting the time series features into a second data processing layer of a preset fatigue life prediction model to obtain multiple regression prediction results. The second data processing layer performs regression prediction based on multiple gradient boosting trees.

[0090] In the implementation, time series features are extracted by the first data processing layer and passed as input to the second data processing layer. The second data processing layer performs regression prediction based on multiple gradient boosting trees. Gradient boosting trees are an ensemble learning algorithm that gradually reduces the overall model error by building a series of decision tree models. Each tree independently learns the features of the input data and outputs a prediction result.

[0091] S10222: Obtain the expected remaining life of the target roll based on multiple regression prediction results.

[0092] In practice, the outputs of all trees are weighted averaged (or weighted summed) to produce the final prediction. Specifically, each regression result is assigned a weight, with predictions with higher weights having a greater impact on the final prediction. Ultimately, all predictions are weighted and summed to produce the final predicted remaining life of the roll.

[0093] The roller state determination method provided in the embodiment of the present application obtains a more stable and accurate remaining life prediction by integrating multiple regression results, thereby accurately obtaining the expected remaining life.

[0094] In order to obtain an accurate fatigue life prediction model, in some embodiments, reference may be made to Figure 2 Before S102, the following steps may also be included: S201 to S205.

[0095] S201: Filtering out a training sample set and a verification sample set from a plurality of training samples, where each training sample includes historical rolling data of a sample roll and remaining life information corresponding to the sample roll.

[0096] In the implementation, training and validation sets are selected from all available data. This data includes historical rolling data for the rolls and information about the actual remaining life of each roll. For example, suppose there are 100 roll samples, 70 of which are used as the training set and 30 as the validation set. The input features of the training samples may include roll vibration, load, and usage time, while the output is the remaining life of each roll.

[0097] S202: Constructing an initial fatigue life prediction model based on the training sample set.

[0098] In practice, a preliminary fatigue life prediction model is constructed using training data. The training process uses a selected algorithm to learn how to link historical data with actual lifespan based on the input data in the training sample. Through continuous iterative optimization, the initial fatigue life prediction model is obtained.

[0099] S203: Input the verification sample set into the initial fatigue life prediction model to determine the remaining life prediction result corresponding to each sample roll in the verification sample set.

[0100] In practice, the sample data from the validation set is fed into the trained model to obtain predictions for each validation sample. The validation set allows us to evaluate the model's performance on unseen data, thereby measuring its generalization ability and accuracy.

[0101] S204: Constructing a loss function based on the remaining life information and remaining life prediction results of each sample roller in the verification sample set.

[0102] In practice, a loss function is defined based on the error between the actual remaining life information in the validation set and the model's predictions. This function measures the difference between the predicted and true values. Specifically, the error is calculated for each sample in the validation set, and then the total loss function value is calculated based on these errors. Smaller loss function values ​​indicate stronger predictive capabilities of the model.

[0103] S205: Constructing a new fatigue life prediction model based on the loss function, and constructing a new loss function based on the validation sample set and the new fatigue life prediction model, until a preset training stop condition is met, thereby obtaining a trained fatigue life prediction model.

[0104] In the specific implementation, a new fatigue life prediction model is constructed based on the loss function. Specifically, the calculation is performed iteratively, and the fatigue life prediction model is continuously trained through the verification sample set until the preset training stop condition is met, thereby obtaining the trained fatigue life prediction model.

[0105] The roller state determination method provided in the embodiment of the present application trains the model by screening out a training sample set and a verification sample set from multiple training samples, constructs a new fatigue life prediction model based on the loss function, and constructs a new loss function based on the verification sample set and the new fatigue life prediction model until the preset training stop condition is met, thereby obtaining an accurate fatigue life prediction model.

[0106] In order to accurately locate the damage of the target roll, in some embodiments, S103 may include the following steps: S1031 to S1033.

[0107] S1031: Extracting force data from the operating data.

[0108] Force data is data that describes the various external forces (such as compression, shear, tension, etc.) that the roller bears during operation. These external forces are key factors affecting the fatigue life and damage evolution of the roller.

[0109] In a specific implementation, the working data of the roller is collected by the sensor to obtain the operation data, and then the force data in the operation data is extracted.

[0110] S1032: Inputting the force data into a preset damage model to determine stress data of the target roller, wherein the damage model performs physical calculation based on the force data to obtain corresponding stress data.

[0111] In practice, stress data is calculated using physics formulas based on the mechanical properties of the material and the operating conditions of the roll. Specific damage models include finite element analysis, fracture mechanics, and cyclic loading models. These models, fed with force data, simulate the stress state of the roll material and generate stress values ​​at various locations and time points.

[0112] S1033: Predicting the initial position and expansion direction of the damage based on the stress data to obtain damage information of the target roll.

[0113] In practice, the initial location and direction of damage are determined by stress concentration areas. Stress concentration areas are regions within the material with relatively high stress, making them more susceptible to cracks and damage. Stress data is analyzed to identify areas with high stress values ​​as potential damage locations. The damage extension direction is then predicted accordingly, yielding damage information for the target roll.

[0114] The roller state determination method provided in the embodiment of the present application obtains actual force data and calculates stress distribution using a physical model, thereby accurately locating the damage of the target roller based on physical force calculation.

[0115] In order to ensure the accuracy and applicability of the prediction results, in some embodiments, S1033 may include the following steps: S10331 to S10333.

[0116] S10331: Determine the stress concentration area on the target roll based on the stress data.

[0117] In practice, stress data is used for spatial analysis, where stress values ​​are calculated for each region and compared with those of surrounding areas. If the stress value in a region is significantly higher than that of its neighboring regions, it is marked as a stress concentration area. Typically, the software defines a threshold value, and when the stress at a point exceeds this threshold, it is considered a stress concentration area.

[0118] S10332: Determine whether each position point in the stress concentration area has reached the critical condition for damage formation.

[0119] In practice, the stress value at each stress concentration point is compared with the material's theoretical damage threshold. This threshold is typically determined by factors such as the material's fatigue strength and yield strength. When the stress exceeds these thresholds, damage is likely to occur at that location. This allows us to determine whether the current stress value is sufficient to cause permanent damage to the material.

[0120] S10333: Based on the position point where the critical condition for damage formation is reached, the initial position and expansion direction of the damage are predicted as damage information of the target roll.

[0121] In the specific implementation, once a certain point in the stress concentration area reaches the critical condition, the initial position and expansion direction of the damage are predicted based on mechanical theory and crack propagation model as the damage information of the target roll.

[0122] The roller state determination method provided in the embodiment of the present application determines the stress concentration area on the target roller based on stress data, and then judges whether each position point in the stress concentration area reaches the critical conditions for damage formation. Then, based on the position points that reach the critical conditions for damage formation, the initial position and expansion direction of the damage are predicted as the damage information of the target roller, thereby ensuring the accuracy and applicability of the prediction results.

[0123] In order to accurately obtain the roller operation data, in some embodiments, refer to Figure 3 , S101, may include the following steps: S301 to S303.

[0124] S301: Acquire raw data of a target roller through multiple sensors.

[0125] In the specific implementation, by installing various sensors (such as stress sensors, temperature sensors, vibration sensors, pressure sensors, etc.) on the rolling equipment, various data of the target rolling mill in operation can be collected in real time to obtain the original data of the target rolling mill.

[0126] S302: Eliminate outliers and / or remove noise from the original data to obtain first data.

[0127] In practice, data denoising typically uses filtering methods, such as low-pass filters and Kalman filters. Outlier detection compares collected data with historical data to identify data that significantly deviates from the normal range. These outliers may be caused by sensor failures, changes in the external environment, and so on. Outliers are removed and / or denoised based on the actual situation to obtain the first data.

[0128] S303: Extract key characteristic parameters from the first data to obtain operating data of the target roller within a preset time period.

[0129] In a specific implementation, characteristic parameters related to damage judgment are extracted from the first data, thereby obtaining the operating data of the target roller within a preset time period.

[0130] The roller status determination method provided in the embodiment of the present application obtains the original data of the target roller through multiple sensors, and then eliminates outliers and / or denoises the original data to obtain first data, so as to extract key characteristic parameters in the first data and obtain the operating data of the target roller within a preset time period, thereby efficiently and accurately obtaining the roller operation data, and providing a reliable basis for subsequent damage detection and maintenance decisions.

[0131] In some implementations, historical data from the roll's service life is collected in advance to construct a comprehensive database covering multi-dimensional information. This accurately characterizes the roll's stress conditions and fatigue damage evolution under different operating conditions, providing reliable data support for subsequent modeling. The rolling load data includes rolling force, contact pressure, rolling force per unit width, and rolling speed; the roll material property data includes hardness distribution, microstructure, elastic modulus, and fatigue strength; the rolling process parameter data includes rolling temperature, lubrication conditions, cooling method, and rolling tension; and the roll replacement and grinding records include grinding volume, roll replacement time, grinding frequency, and service life.

[0132] To ensure data accuracy and validity, the data collection process can incorporate a variety of sensors and online monitoring systems, such as pressure sensors, vibration sensors, temperature sensors, and ultrasonic testing equipment, to monitor the real-time status of the rolls during service. Historical data can then be archived and analyzed. During data processing, the raw data undergoes preprocessing, including denoising, outlier removal, and smoothing, to reduce the influence of external interference. Feature engineering methods are also employed to deeply mine the data and extract key characteristic parameters, such as maximum contact stress, maximum shear stress, residual stress distribution, and temperature gradient, to enhance the model's predictive capabilities. Furthermore, to enhance the data's time series characteristics, time series analysis can be performed to construct long-term trends, short-term fluctuations, and periodicity, thereby more accurately describing the cumulative evolution of roll damage. Comprehensive data collection and processing provide high-quality data input for subsequent roll damage status and fatigue life prediction models, ensuring accurate and stable predictions. This can then optimize roll maintenance strategies, improve roll service life, and enhance rolling production stability.

[0133] Based on data collection, a roll damage state and fatigue life prediction model was constructed to accurately characterize the damage evolution and remaining life of the roll under different service conditions. The model includes a transformer processing layer and an XGBoost processing layer. The transformer processing layer extracts time series features, which are then processed by the XGBoost processing layer to predict the remaining life. Deep learning methods are used to mine the complex nonlinear relationships in historical data, while ensemble learning is used to improve the stability and generalization of predictions, thereby achieving high-precision fatigue life prediction and damage location assessment.

[0134] Specifically, define the input dataset where x i =[x i1 ,x i2 ,...,x im ] represents the characteristic vector of the i-th sample, including rolling load, roll material characteristics, rolling process parameters, grinding records, etc., y i The corresponding target variables include fatigue life, damage depth, etc., N is the total number of samples, and m is the feature dimension.

[0135] Since the evolution of roller damage is time-dependent, in order to better capture its long-term trends and short-term fluctuations, the Transformer structure is used to model time series data. The feature weights of different time steps are calculated through the self-attention mechanism, and the calculation formula is as follows:

[0136]

[0137] Among them, Q, K, and V represent the query matrix, key matrix, and value matrix respectively, d k is the dimension of the key matrix. The Transformer architecture uses a multi-head self-attention mechanism to enhance the model's ability to focus on key factors affecting the damage evolution process.

[0138] In the fatigue life prediction model, the Transformer processing layer is mainly used to extract time series features, while XGBoost uses the Gradient Boosting Tree (GBT) to perform regression prediction of fatigue life and damage depth. XGBoost uses an additive model to build a strong predictor. Its basic formula is as follows:

[0139]

[0140] in, is the loss function, Ω(f t ) is a regularization term used to control model complexity to prevent overfitting.

[0141] During model training, Transformer is used to extract time series features, whose output serves as input features for XGBoost, enhancing the model's ability to depict the cumulative trend of roll damage. During training, an optimizer is used to adjust Transformer parameters, and Bayesian optimization is combined to fine-tune XGBoost hyperparameters such as the learning rate, tree depth, and regularization parameter to achieve optimal prediction results.

[0142] This solution, based on a roll damage state and fatigue life prediction model, comprehensively analyzes the load characteristics, stress distribution, and historical damage accumulation experienced by the roll during its service life. It then combines key factors such as the roll's material parameters, operating temperature, rolling pressure, and friction coefficient to calculate the roll's remaining fatigue life under current operating conditions. The model dynamically tracks the evolution of roll fatigue damage, accurately assesses its operational status, and predicts life degradation trends under future operating conditions.

[0143] On this basis, the fatigue damage position and damage depth of the roller are further predicted. By analyzing the three-dimensional stress distribution on the surface and subsurface of the roller, the radial stress σ r , hoop stress σ θ and axial stress σ z , whose distribution can be calculated by the Lame Equation:

[0144]

[0145]

[0146] σ z =ν(σ r +σ θ )

[0147] Where P is the rolling pressure, R o is the outer diameter of the roller, R i is the inner diameter of the roller, r is the radial position of the calculation point, and ν is the Poisson's ratio.

[0148] Then, the maximum shear stress τ is calculated from the three principal stresses max :

[0149]

[0150] Among them, σ max and σ min are the maximum and minimum principal stresses.

[0151] Furthermore, combined with the damage accumulation model, the linear fatigue accumulation criterion assumes that at stress level S i Under this condition, the fatigue life of the material is N i , the actual number of cycles is n i , then the fatigue damage D is calculated as:

[0152]

[0153] When D≥1, the material reaches fatigue failure state.

[0154] where N i Calculated by the Basquin equation:

[0155]

[0156] Among them, the crack growth rate coefficient C and the crack growth index m are determined by experiments, S i is the cyclic stress amplitude.

[0157] Furthermore, after fatigue damage accumulates, the crack will further expand, and its expansion rate follows the Paris-Erdogan formula, and a crack growth dynamics model is constructed.

[0158] The Paris-Erdogan formula is specifically:

[0159]

[0160] Where: a is the crack length, da / dN is the crack growth per cycle; ΔK is the stress intensity factor range: The crack growth rate coefficient C and crack growth exponent m are Paris-Erdogan constants determined experimentally; Y is the geometric correction factor (related to the crack morphology); Δσ is the cyclic stress amplitude.

[0161] Afterwards, the potential initiation location of the crack is inferred. The crack preferentially initiates in the area of ​​maximum stress concentration. The stress concentration coefficient K is calculated by combining the Neuber formula. t :

[0162]

[0163] Where: a is the defect size, r is the radius around the defect.

[0164] Finally, the final failure depth of the crack is inferred, and the crack extends to the critical crack depth a c The critical crack depth is determined by the Griffith fracture criterion:

[0165]

[0166] Among them, K c is the fracture toughness of the material, and σ is the maximum principal stress.

[0167] This approach comprehensively analyzes the three-dimensional stress distribution on the roll surface and subsurface to calculate the maximum shear stress path, and then uses the Miner linear fatigue accumulation criterion to assess the degree of fatigue damage. The crack growth kinetics model, based on the Paris-Erdogan equation, fully considers the crack initiation location, growth rate, and ultimate failure depth. The Neuber equation is also used to calculate the stress concentration factor to determine the preferred crack initiation area. The critical crack depth is assessed using the Griffith fracture criterion to ensure the accuracy and applicability of the prediction results.

[0168] Based on the prediction results, the maintenance strategy of the rolls is optimized, and the roll replacement cycle and grinding amount are reasonably formulated to maximize the utilization of the roll life and ensure efficient and stable operation of the rolling production.

[0169] Through real-time calculations using a fatigue life prediction model, the health of the rolls is dynamically assessed. Combined with historical load data, rolling process parameters, and material properties, the roll damage accumulation patterns under different operating conditions are determined. Through in-depth analysis of the fatigue damage evolution process, critical damage areas and their development trends are identified, and damage thresholds are scientifically set to ensure timely maintenance of the rolls before damage reaches acceptable limits. Based on the roll fatigue damage accumulation, a data-driven optimization approach is used to dynamically adjust the roll replacement cycle and regrinding strategy. By accurately predicting the damage rate, stress concentration areas, and crack propagation paths during the roll's service life, the amount of regrinding is optimized to ensure that each regrinding effectively removes the damaged layer while avoiding shortening the roll life due to excessive regrinding. A tiered optimization strategy is established for roll conditions under different materials, rolling processes, and rolling loads, enabling maintenance plans to adapt to different production environments and improving the accuracy of roll replacement and regrinding.

[0170] The data-driven roll damage detection and fatigue life prediction method can not only effectively reduce the frequency of roll replacement and reduce roll cost expenditure, but also ensure the stability of product quality during the rolling process and improve production efficiency.

[0171] Based on the roll state determination method provided in the above embodiment, the present application also provides a specific implementation of a roll state determination device. Please refer to the following embodiment.

[0172] See first Figure 4 The roller state determination device 400 provided in the embodiment of the present application includes the following modules:

[0173] The acquisition module 401 is used to acquire the operating data of the target roller, where the operating data includes the force data of the target roller during operation.

[0174] Calculation module 402 is configured to input the operating data into a preset fatigue life prediction model to determine the expected remaining life of the target roll. The fatigue life prediction model is trained based on historical rolling data of a sample roll. The historical rolling data includes at least one of force data, material data, process data, and usage data.

[0175] The calculation module 402 is further used to extract force data from the operating data and input the force data into a preset damage model to obtain damage information of the target roll.

[0176] The determination module 403 is configured to determine the damage state of the target roll based on the expected remaining life and the damage information.

[0177] As an implementation of the present application, the calculation module 402 includes:

[0178] The extraction unit is used to input the operating data into the first data processing layer in the preset fatigue life prediction model to extract the time series characteristics of the operating data.

[0179] The determination unit is used to input the time series characteristics into the second data processing layer of the preset fatigue life prediction model for regression prediction to obtain the expected remaining life of the target roller.

[0180] As an implementation of the present application, the extraction unit includes:

[0181] The segmentation sub-unit is used to segment the running data into multiple sub-sequences according to the time sequence.

[0182] The processing subunit is configured to input the multiple subsequences into a first data processing layer of a preset fatigue life prediction model to obtain time series features corresponding to the multiple subsequences. The first data processing layer processes the multiple subsequences based on a self-attention mechanism.

[0183] The determination subunit is used to obtain the time series characteristics of the operation data according to the time series characteristics corresponding to the multiple subsequences.

[0184] As an implementation of the present application, the determining unit includes:

[0185] The processing subunit is configured to input the time series features into a second data processing layer of a preset fatigue life prediction model to obtain multiple regression prediction results. The second data processing layer performs regression prediction based on multiple gradient boosting trees.

[0186] The determination subunit is used to obtain the expected remaining life of the target roll according to multiple regression prediction results.

[0187] As an implementation of the present application, the roller state determination device 400 further includes:

[0188] The screening module is used to screen out a training sample set and a verification sample set from a plurality of training samples, each training sample including historical rolling data of the sample roll and the remaining life information corresponding to the sample roll.

[0189] A building module is used to build an initial fatigue life prediction model based on a training sample set.

[0190] The determination module is used to input the verification sample set into the initial fatigue life prediction model to determine the remaining life prediction result corresponding to each sample roller in the verification sample set.

[0191] The construction module is also used to construct a loss function based on the remaining life information and remaining life prediction results of each sample roller in the verification sample set.

[0192] The construction module is also used to construct a new fatigue life prediction model based on the loss function, and to construct a new loss function based on the verification sample set and the new fatigue life prediction model, until the preset training stop condition is met to obtain the trained fatigue life prediction model.

[0193] As an implementation of the present application, the calculation module 402 includes:

[0194] The extraction unit is used to extract force data from the operating data.

[0195] The determination unit is configured to input the force data into a preset damage model to determine stress data of the target roller, wherein the damage model performs physical calculation based on the force data to obtain corresponding stress data.

[0196] The prediction unit is used to predict the initial position and expansion direction of the damage based on the stress data to obtain the damage information of the target roller.

[0197] As an implementation of the present application, the prediction unit includes:

[0198] The determination subunit is used to determine the stress concentration area on the target roll based on the stress data.

[0199] The judgment subunit is used to judge whether each position point in the stress concentration area has reached the critical condition for damage formation.

[0200] The prediction subunit is used to predict the initial position and expansion direction of the damage according to the position point where the critical condition for damage formation is reached, so as to serve as the damage information of the target roller.

[0201] As an implementation of the present application, the acquisition module 401 includes:

[0202] The acquisition unit is used to acquire the original data of the target roller through multiple sensors.

[0203] The processing unit is used to remove outliers and / or remove noise from the original data to obtain first data.

[0204] The extraction unit is used to extract key characteristic parameters from the first data to obtain the operating data of the target roller within a preset time period.

[0205] The various modules in the roller state determination device provided in the embodiment of the present application can implement the various steps in the above-mentioned roller state determination method and achieve corresponding effects. For the sake of concise description, they will not be repeated here.

[0206] Figure 5 A schematic structural diagram of the roll state determination hardware provided in an embodiment of the present application is shown.

[0207] The roller state determination device may include a processor 501 and a memory 502 storing computer program instructions.

[0208] Specifically, the processor 501 may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0209] The memory 502 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 502 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 502 may include removable or non-removable (or fixed) media. Where appropriate, the memory 502 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 502 is a non-volatile solid-state memory.

[0210] The memory may include read-only memory (ROM), random access memory (RAM), magnetic disk storage media devices, optical storage media devices, flash memory devices, electrical, optical or other physical / tangible memory storage devices. Therefore, generally, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the roll state determination method according to any one embodiment of the present disclosure.

[0211] The processor 501 reads and executes computer program instructions stored in the memory 502 to implement any one of the roll state determination methods in the above embodiments.

[0212] In one example, the roller state determination device may further include a communication interface 503 and a bus 510. Figure 5 As shown, the processor 501, the memory 502, and the communication interface 503 are connected via a bus 510 and communicate with each other.

[0213] The communication interface 503 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0214] Bus 510 includes hardware, software or both, and the components of online data flow metering equipment are coupled to each other. For example, but not limitation, bus can include accelerated graphics port (AGP) or other graphics bus, enhanced industry standard architecture (EISA) bus, front side bus (FSB), hypertransport (HT) interconnection, industry standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnection (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 510 can include one or more buses. Although the present application embodiment describes and shows specific bus, the application considers any suitable bus or interconnection.

[0215] In addition, in conjunction with the roll state determination method in the above embodiments, embodiments of the present application may provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when the computer program instructions are executed by a processor, any of the roll state determination methods in the above embodiments is implemented.

[0216] An embodiment of the present application further provides a computer program product, including a computer program, which, when processed and executed, implements any one of the methods for determining the state of a rolling mill in the above embodiments.

[0217] It should be understood that the present application is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, a detailed description of known methods is omitted here. In the above embodiments, several specific steps are described and illustrated as examples. However, the method process of the present application is not limited to the specific steps described and illustrated. Those skilled in the art can make various changes, modifications, and additions, or change the order of the steps after understanding the spirit of the present application.

[0218] The functional blocks shown in the above-described block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of the present application are programs or code segments that are used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link by a data signal carried in a carrier wave. "Machine-readable medium" can include any medium that can store or transmit information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROMs, flash memories, erasable ROMs (EROMs), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0219] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps. In other words, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0220] Aspects of the present disclosure have been described above with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the implementation of the function / action specified in one or more boxes of the flowchart and / or block diagram. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor or a field programmable logic circuit. It is also understood that each box in the block diagram and / or flowchart and the combination of the boxes in the block diagram and / or flowchart can also be implemented by dedicated hardware that performs the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0221] The above description is only a specific embodiment of the present application. Those skilled in the art will clearly understand that for the convenience and brevity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the scope of protection of the present application.

Claims

1. A method for determining the state of a roll, characterized in that: The method comprises: Acquiring operating data of a target roller, wherein the operating data includes force data of the target roller during operation; Inputting the operating data into a preset fatigue life prediction model to obtain the expected remaining life of the target roll; the fatigue life prediction model is trained based on historical rolling data of a sample roll, the historical rolling data including at least one of force data, material data, process data, and usage data; extracting force data from the operating data, and inputting the force data into a preset damage model to obtain damage information of the target roll; A damage state of the target roll is determined based on the expected remaining life and the damage information.

2. The method for determining the roll state according to claim 1, wherein: Inputting the operating data into a preset fatigue life prediction model to obtain the expected remaining life of the target roll includes: Inputting the operating data into a first data processing layer in a preset fatigue life prediction model to extract time series features of the operating data; The time series characteristics are input into the second data processing layer of the preset fatigue life prediction model for regression prediction to obtain the expected remaining life of the target roller.

3. The method for determining the roll state according to claim 2, wherein: Inputting the operating data into a first data processing layer of a preset fatigue life prediction model to extract time series features of the operating data includes: Dividing the operating data into multiple subsequences according to time sequence; Inputting the plurality of subsequences into a first data processing layer in a preset fatigue life prediction model to obtain time series features corresponding to the plurality of subsequences; wherein the first data processing layer processes the plurality of subsequences based on a self-attention mechanism; The time series features of the operation data are obtained according to the time series features corresponding to the multiple subsequences.

4. The method for determining the state of a roll according to claim 2, wherein: The step of inputting the time series characteristics into a second data processing layer of a preset fatigue life prediction model for regression prediction to obtain the expected remaining life of the target roll includes: Inputting the time series features into a second data processing layer in a preset fatigue life prediction model to obtain multiple regression prediction results; wherein the second data processing layer performs regression prediction based on multiple gradient boosting trees; The expected remaining life of the target roll is obtained based on the multiple regression prediction results.

5. The method for determining the state of a roll according to claim 1, wherein: Before inputting the operating data into a preset fatigue life prediction model to obtain the expected remaining life of the target roll, the method further includes: Screening out a training sample set and a verification sample set from a plurality of training samples, each of the training samples including historical rolling data of a sample roll and remaining life information corresponding to the sample roll; constructing an initial fatigue life prediction model based on the training sample set; Inputting the verification sample set into the initial fatigue life prediction model to determine the remaining life prediction result corresponding to each sample roll in the verification sample set; constructing a loss function based on the remaining life information and remaining life prediction results of each sample roller in the verification sample set; A new fatigue life prediction model is constructed based on the loss function, and a new loss function is constructed based on the verification sample set and the new fatigue life prediction model until a preset training stop condition is met to obtain a trained fatigue life prediction model.

6. The method for determining the state of a roll according to claim 1, wherein: The extracting force data from the operating data and inputting the force data into a preset damage model to obtain damage information of the target roll includes: extracting force data from the operating data; Inputting the force data into a preset damage model to determine the stress data of the target roller, wherein the damage model performs physical calculation based on the force data to obtain corresponding stress data; The initial position and expansion direction of the damage are predicted based on the stress data to obtain damage information of the target roll.

7. The method for determining the state of a roll according to claim 6, wherein: The method of predicting the initial position and extension direction of the damage based on the stress data to obtain the damage information of the target roller includes: determining a stress concentration area on the target roll based on the stress data; Determining whether each position point in the stress concentration area reaches a critical condition for damage formation; According to the position point where the critical condition for damage formation is reached, the initial position and expansion direction of the damage are predicted as the damage information of the target roll.

8. The method for determining the state of a roll according to any one of claims 1 to 7, characterized in that: The obtaining of the operating data of the target roller includes: Acquire the raw data of the target roll through multiple sensors; Eliminating outliers and / or removing noise from the original data to obtain first data; The key characteristic parameters in the first data are extracted to obtain the operation data of the target roller within a preset time period.

9. A device for determining the state of a roll, characterized in that: The device comprises: An acquisition module, configured to acquire operating data of a target roller, wherein the operating data includes force data of the target roller during operation; a calculation module, configured to input the operating data into a preset fatigue life prediction model to obtain an expected remaining life of the target roll; the fatigue life prediction model is trained based on historical rolling data of a sample roll, the historical rolling data including at least one of force data, material data, process data, and usage data; The calculation module is further used to extract force data from the operating data and input the force data into a preset damage model to obtain damage information of the target roller; A determination module is used to determine the damage state of the target roller based on the expected remaining life and the damage information.

10. A roller state determination device, characterized in that: The device includes: a processor and a memory storing computer program instructions; the processor reads and executes the computer program instructions to implement the roller state determination method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer program instructions, and when the computer program instructions are executed by a processor, the method for determining the state of a roller according to any one of claims 1 to 8 is implemented.

12. A computer program product, characterized in that When the instructions in the computer program product are executed by a processor of an electronic device, the electronic device executes the roller state determination method according to any one of claims 1 to 8.

Images