Method, device and equipment for predicting service life of rolling mill and medium

Abstract

The invention provides a rolling mill service life prediction method and device, equipment and a medium, and the method comprises the steps: receiving first operation data, sent by at least one sensor, of a first rolling mill; data preprocessing is carried out on the first operation data to obtain target operation data, and the data preprocessing comprises at least one of data cleaning processing, data normalization processing and feature extraction processing; and inputting the target operation data into a first model to obtain a life prediction result corresponding to the target operation data, the first model being a trained second model, and the second model being a deep learning model or a reinforcement learning model. In the invention, the first model is a trained deep learning model or reinforcement learning model, and can capture a complex nonlinear relationship, so that the service life of the rolling mill is predicted through the first model, and the accuracy of predicting the service life of the rolling mill can be improved without depending on manual prediction.

Description

Technical Field

[0001] The present invention belongs to the technical field of life prediction, and in particular relates to a method, device, equipment and medium for predicting the life of a rolling mill. Background Art

[0002] As equipment for realizing the metal rolling process, rolling mills play a vital role in the steel and other industries. Their stability and reliability are directly related to production efficiency and product quality, so their life prediction is more important. In related technologies, the remaining life of rolling mills is mainly obtained based on statistical models or expert experience analysis. However, statistical models are usually difficult to capture nonlinear relationships, resulting in poor accuracy in the prediction results of the remaining life of rolling mills; rolling mill life prediction methods based on expert experience analysis rely too much on manual judgment, resulting in strong subjectivity. It can be seen that how to improve the accuracy of rolling mill life prediction without relying on manual prediction is a problem to be solved. Summary of the invention

[0003] In view of the above-mentioned shortcomings of the prior art, an object of the present invention is to provide a method, device, equipment and medium for predicting the life of a rolling mill, so as to solve the above-mentioned problems.

[0004] In a first aspect, the present invention provides a method for predicting the life of a rolling mill, the method comprising:

[0005] receiving first operation data of a first rolling mill sent by at least one sensor;

[0006] Performing data preprocessing on the first operation data to obtain target operation data, wherein the data preprocessing includes at least one of the following: data cleaning processing, data normalization processing, and feature extraction processing;

[0007] The target operating data is input into a first model to obtain a first life prediction result corresponding to the target operating data, wherein the first model is a trained second model, and the second model is a deep learning model or a reinforcement learning model.

[0008] Optionally, the data preprocessing includes feature extraction processing, and the feature extraction processing includes time-frequency feature extraction processing based on wavelet transform.

[0009] Optionally, the data preprocessing includes data cleaning processing, data normalization processing and feature extraction processing, and the data preprocessing of the first operation data includes:

[0010] Performing data cleaning processing on the first operation data to obtain second operation data, wherein the data cleaning processing includes at least one of outlier removal processing, missing value compensation processing, and denoising processing;

[0011] performing data normalization processing on the second operation data to obtain third operation data;

[0012] Performing time-frequency feature extraction processing on the third operation data based on wavelet transform to obtain fourth operation data;

[0013] The fourth operation data is subjected to main feature extraction processing based on principal component analysis technology PCA to obtain the target operation data.

[0014] Optionally, the method further comprises:

[0015] When receiving the label corresponding to the target operation data, adding a first sample to a first sample set, the first sample consisting of the target operation data and the label corresponding to the target operation data, the label being an artificially determined life prediction result;

[0016] When the sample features in the first sample set meet preset requirements, the parameters of the first model are updated by an online learning method based on the first sample set to obtain a third model.

[0017] Optionally, the first rolling mill is a rolling mill operating in a first operating condition, and the method further comprises:

[0018] Acquire a second sample set, wherein the second sample set includes a plurality of different second samples, each of which includes fifth operating data that has undergone the data preprocessing and a corresponding label, the fifth operating data is historical operating data of the rolling mill operating in the second working condition, and the label is an artificially determined life prediction result;

[0019] Based on the second sample set, the parameters of the first model are updated by using a transfer learning method to obtain a fourth model, and the fourth model is used to determine the life prediction result of the rolling mill working under the second working condition.

[0020] Optionally, the first rolling mill is a rolling mill operating in a first working condition, and the first model is obtained by the following steps:

[0021] Constructing a third sample set, wherein the third sample set includes a plurality of different third samples, each of the third samples includes sixth operating data that has been preprocessed and a corresponding label, the sixth operating data includes historical operating data of the rolling mill operating under the first operating condition, and the label is an artificially determined life prediction result;

[0022] Training the second model using the third sample set to obtain a fifth model;

[0023] Constructing a fourth sample set, wherein the fourth sample set includes a plurality of different fourth samples, each of which includes seventh operating data that has been preprocessed and a corresponding label, and the seventh operating data includes historical operating data of the rolling mill operating in the first working condition;

[0024] The fifth model is verified by using the fourth sample set, and when the verification passes, the fifth model is determined to be the first model.

[0025] Optionally, each of the at least one sensor is used to collect operating data of the first rolling mill at a preset frequency, and the at least one sensor includes a temperature sensor and a pressure sensor.

[0026] In a second aspect, the present invention also provides a device for predicting the life of a rolling mill, comprising:

[0027] A receiving module, used for receiving first operation data of a first rolling mill sent by at least one sensor;

[0028] A data preprocessing module, used to perform data preprocessing on the first operation data to obtain target operation data, wherein the data preprocessing includes at least one of the following: data cleaning processing, data normalization processing and feature extraction processing;

[0029] A prediction module is used to input the target operating data into a first model to obtain a first life prediction result corresponding to the target operating data, wherein the first model is a trained second model, and the second model is a deep learning model or a reinforcement learning model.

[0030] Optionally, the data preprocessing includes feature extraction processing, and the feature extraction processing includes time-frequency feature extraction processing based on wavelet transform.

[0031] Optionally, the data preprocessing includes data cleaning processing, data normalization processing and feature extraction processing, and the data preprocessing module is specifically used to:

[0032] Performing data cleaning processing on the first operation data to obtain second operation data, wherein the data cleaning processing includes at least one of outlier removal processing, missing value compensation processing, and denoising processing;

[0033] performing data normalization processing on the second operation data to obtain third operation data;

[0034] Performing time-frequency feature extraction processing on the third operation data based on wavelet transform to obtain fourth operation data;

[0035] The fourth operation data is subjected to main feature extraction processing based on principal component analysis technology PCA to obtain the target operation data.

[0036] Optionally, the device further comprises a first updating module, configured to

[0037] When receiving the label corresponding to the target operation data, adding a first sample to a first sample set, the first sample consisting of the target operation data and the label corresponding to the target operation data, the label being an artificially determined life prediction result;

[0038] When the sample features in the first sample set meet preset requirements, the parameters of the first model are updated by an online learning method based on the first sample set to obtain a third model.

[0039] Optionally, the first rolling mill is a rolling mill operating in a first operating condition, and the device further comprises a second updating module, which is used to:

[0040] Acquire a second sample set, wherein the second sample set includes a plurality of different second samples, each of which includes fifth operating data that has undergone the data preprocessing and a corresponding label, the fifth operating data is historical operating data of the rolling mill operating in the second working condition, and the label is an artificially determined life prediction result;

[0041] Based on the second sample set, the parameters of the first model are updated by using a transfer learning method to obtain a fourth model, and the fourth model is used to determine the life prediction result of the rolling mill working under the second working condition.

[0042] Optionally, the first rolling mill is a rolling mill operating in a first working condition, and the first model is obtained by the following steps:

[0043] Constructing a third sample set, wherein the third sample set includes a plurality of different third samples, each of the third samples includes sixth operating data that has been preprocessed and a corresponding label, the sixth operating data includes historical operating data of the rolling mill operating under the first operating condition, and the label is an artificially determined life prediction result;

[0044] Training the second model using the third sample set to obtain a fifth model;

[0045] Constructing a fourth sample set, wherein the fourth sample set includes a plurality of different fourth samples, each of which includes seventh operating data that has been preprocessed and a corresponding label, and the seventh operating data includes historical operating data of the rolling mill operating in the first working condition;

[0046] The fifth model is verified by using the fourth sample set, and when the verification passes, the fifth model is determined to be the first model.

[0047] Optionally, each of the at least one sensor is used to collect operating data of the first rolling mill at a preset frequency, and the at least one sensor includes a temperature sensor and a pressure sensor.

[0048] In a third aspect, the present invention further provides an electronic device, the electronic device comprising:

[0049] one or more processors;

[0050] A storage device is used to store one or more programs, and when the one or more programs are executed by the one or more processors, the electronic device implements the method for predicting the life of a rolling mill.

[0051] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor of a computer, the computer executes the method for predicting the life of a rolling mill.

[0052] Beneficial effects of the technical solution: In the technical solution, by preprocessing the first operating data of the first rolling mill to obtain the target operating data, the quality and availability of the data are effectively improved, and a reliable input is provided for the first model; by inputting the target operating data into the first model, the corresponding life prediction result can be obtained. In the technical solution, the first model is a trained deep learning model or reinforcement learning model, which can capture complex nonlinear relationships, so that the life of the rolling mill can be predicted through the first model, which can improve the accuracy of the life prediction of the rolling mill without relying on manual prediction.

[0053] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The drawings herein are incorporated into the specification and constitute a part of the specification, showing embodiments consistent with the present application, and together with the specification, are used to explain the principles of the present application. Obviously, the drawings described below are only some embodiments of the present application, and for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0055] Figure 1 is a flow chart of a method for predicting the life of a rolling mill shown in an exemplary embodiment of the present application;

[0056] Figure 2is a structural diagram of a prediction system for rolling mill life shown in an exemplary embodiment of the present application;

[0057] Figure 3 is a prediction accuracy improvement curve diagram shown in an exemplary embodiment of the present application;

[0058] Figure 4 is an adaptability improvement curve diagram shown in an exemplary embodiment of the present application;

[0059] Figure 5 is a structural diagram of a prediction system for rolling mill life shown in an exemplary embodiment of the present application; DETAILED DESCRIPTION

[0060] The following will describe the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention, not for limiting the scope of protection of the present invention.

[0061] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and thus the drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0062] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it is obvious to those skilled in the art that the embodiments of the present invention can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present invention difficult to understand.

[0063] See also Figure 1 , Figure 1 FIG. 1 is a flow chart of a method for predicting the life of a rolling mill, shown in an exemplary embodiment of the present application. Figure 1 As shown, the method at least includes steps S110 to S130, which are described in detail as follows:

[0064] S110, receiving first operation data of a first rolling mill sent by at least one sensor;

[0065] S120, performing data preprocessing on the first operation data to obtain target operation data, wherein the data preprocessing includes at least one of the following: data cleaning processing, data normalization processing, and feature extraction processing;

[0066] S130. Input the target operating data into a first model to obtain a life prediction result corresponding to the target operating data, wherein the first model is a trained second model, and the second model is a deep learning model or a reinforcement learning model.

[0067] In step S110, the at least one sensor is installed on the first rolling mill to detect the operating status of the first rolling mill.

[0068] In some embodiments, each of the at least one sensor is used to collect the operating data of the first rolling mill at a preset frequency. The preset frequency can be adjusted according to the working conditions, operating load and state of the rolling mill. For example, when the first rolling mill is a rolling mill of a cold rolling steel mill, the preset frequency is the first frequency; when the first rolling mill is a rolling mill of a hot rolling steel mill, the preset frequency is the second frequency, which is higher than the first frequency. As an example, the range of the preset frequency can be set to 1Hz to 100Hz. In order to improve the monitoring effect, each of the above sensors can also be set to have a high-precision and anti-interference design.

[0069] In some embodiments, each of the at least one sensor may be connected to a central monitoring system, and the central monitoring system may detect each sensor, thereby facilitating ensuring that each sensor can collect data at the preset frequency, and the data detected by each sensor may be transmitted to the central monitoring system via an industrial Ethernet or wireless network connection. As an example, Modbus or EtherCAT industrial communication protocols may be used for data transmission to ensure the real-time, stability, and accuracy of data transmission.

[0070] In an exemplary embodiment, the at least one sensor includes a temperature sensor and a pressure sensor. The temperature sensor can be set on the rolling mill bearings and gears to monitor the temperature; the pressure sensor can be set in the pressure system of the rolling mill to detect the pressure. Among them, the at least one sensor can also include an acceleration sensor, a current sensor, a vibration sensor, a strain gauge sensor, etc., which can be determined according to the monitoring requirements.

[0071] In S120, data preprocessing is performed on the first operation data, and the data preprocessing includes at least one of the following: data cleaning processing, data normalization processing, and feature extraction processing. By performing the above data preprocessing on the input data of the first model, the quality of the input data can be improved.

[0072] In some embodiments, the data preprocessing includes feature extraction processing, and the feature extraction processing is used to extract features of the operating data, such as extracting mean features, variance features, and frequency domain features. In other embodiments, the feature extraction processing includes time-frequency feature extraction processing based on wavelet transform. Through wavelet transform, the time domain features and frequency features corresponding to the operating data can be extracted, that is, the multi-scale features of the operating data can be extracted. The first model is used to analyze the multi-scale features, which is conducive to improving the accuracy of life prediction.

[0073] In some embodiments, data preprocessing includes data cleaning, data normalization and feature extraction. The above step S120 includes:

[0074] Performing data cleaning processing on the first operation data to obtain second operation data, wherein the data cleaning processing includes at least one of outlier removal processing, missing value compensation processing, and denoising processing;

[0075] performing data normalization processing on the second operation data to obtain third operation data;

[0076] Performing time-frequency feature extraction processing on the third operation data based on wavelet transform to obtain fourth operation data;

[0077] The fourth operation data is subjected to main feature extraction processing based on principal component analysis technology PCA to obtain the target operation data.

[0078] In this embodiment, the data cleaning process of the first operating data may include at least one of outlier removal, missing value compensation and denoising. In order to improve the quality of the input data of the first model, the first operating data may be simultaneously subjected to outlier removal, missing value compensation and denoising. Among them, outlier removal includes removing invalid data points, missing value processing may adopt interpolation and filling methods, and denoising may adopt high-pass filtering or low-pass filtering methods to remove noise in sensor signals. As an example, a Kalman filter (Kalman Filter) algorithm may be used to remove noise. During the data cleaning process, the Kalman filter algorithm adaptively adjusts the filter parameters according to the dynamic changes of the sensor noise level to ensure that high-precision valid data can still be extracted in a noisy operating environment. At the same time, the state variables are recursively estimated to eliminate short-term fluctuations and enhance the tracking capability of the rolling mill operating state. The Kalman filter algorithm recursively corrects the state estimate, and its core formula is:

[0079]

[0080] in, is a priori estimate, K k is the Kalman gain, zk is the measured value of the sensor, H k is the measurement matrix of the sensor.

[0081] The above normalization process can use MinMax or ZScore to normalize the second operation data, and linearly map the second operation data according to the interval [0, 1] to obtain the third operation data. Taking MinMax as an example, its normalization formula is:

[0082]

[0083] Among them, x is the original data, x min and x max are the minimum and maximum values ​​of the data respectively.

[0084] In the step of extracting the time-frequency features of the third operation data based on wavelet transform to obtain the fourth operation data, the wavelet transform can be decomposed based on the db4 wavelet basis function to extract the multi-scale time-frequency features of the operation state. The relevant formula of the wavelet transform is:

[0085]

[0086] Among them, ψ is the mother wavelet function, a and b are the scale and displacement parameters respectively, and x(t) represents the signal to be feature extracted.

[0087] After extracting the time-frequency features to obtain the fourth operating data, PCA can be used to reduce the dimension of the fourth operating data. PCA can map high-dimensional data to a low-dimensional space by calculating the covariance matrix and performing eigenvalue decomposition on it. The formula is:

[0088] X new =XW

[0089] Among them, X new is the data after dimensionality reduction, and W is the feature vector matrix. By reducing the dimensionality of the fourth running data and extracting its main features, it is beneficial to reduce redundant information and improve the processing speed of subsequent models.

[0090] In the embodiment of the present application, the above-mentioned data preprocessing method is used to perform data cleaning, data normalization and feature extraction on the data, so as to obtain high-quality input data.

[0091] It is worth mentioning that the above-mentioned data preprocessing method is also used to process the historical operation data of the rolling mill during the model training stage. The above-mentioned data preprocessing method is used to process the operation data during the training process, which effectively improves the quality and availability of the data and provides reliable input for subsequent model training.

[0092] In S130, the target operating data that has undergone data preprocessing is input into the first model to obtain a life prediction result corresponding to the target operating data. It is worth mentioning that the target life prediction result is related to the setting of the model output of the first model. As an example, the life prediction result may be a specific remaining life value or a remaining life level. Specifically, during the training process of the first model, the label corresponding to the operating data in its training sample may be a specific remaining life value, and the output of the first model obtained through the sample training is a specific life value. In order to improve the efficiency of life prediction, it can be divided into multiple levels according to the range of the remaining life value, and different levels correspond to different remaining life value ranges. The label corresponding to the operating data in the training sample is determined to be the remaining life level corresponding to the operating data, and the output of the first model obtained through the sample training is the remaining life level.

[0093] The first model is the trained second model, and the second model is an adaptive learning model, specifically a deep learning model or a reinforcement learning model. Both the deep learning model and the reinforcement learning model can effectively capture the complex nonlinear relationship in the operation process of the rolling mill, and the above setting is conducive to improving the accuracy of life prediction. As an example, the deep learning model can be a long short-term memory network (Long Short-Term Memory, LSTM) or a gated recurrent unit neural network (Gated Recurrent Unit, GRU), etc.; reinforcement learning modules such as Qlearning or DQN, etc. LSTM is a time recurrent neural network that introduces a gating mechanism (including an input gate, a forget gate, and an output gate) to regulate the flow of information, so that the network can better capture long-term dependencies. GRU also regulates the flow of information by introducing a gating mechanism (but it is simpler than LSTM, with only two main gates: an update gate and a reset gate), which has fewer parameters and stronger modeling capabilities. Both LSTM and GRU are suitable for processing complex changes in the operation state of the rolling mill. It is worth mentioning that the present application is not limited to the use of the above two neural network structures. The appropriate neural network can be selected according to the actual prediction accuracy and computing resource requirements. For example, the Gradient Boosting Decision Tree (GBT) can also be used. GBT predicts life by constructing multiple decision trees and using gradient descent to optimize the loss function.

[0094] Taking the above deep learning model as an LSTM model as an example, the core formula of LSTM includes:

[0095] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0096] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0097] -C t =f t ×C t-1 +i t ×tanh(W C ·[h t-i ,x t ]+b C )

[0098] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0099] h t =o t ×tanh(C t )

[0100] Among them, f, i t .o t are the activation functions of the forget gate, input gate, and output gate, respectively, and C t is the unit state, h t is the hidden state. The above formulas describe how LSTM t and the state of the previous moment (including the hidden state h t-1 and cell state C t-1 To update the current state (including cell state C t and the hidden state h t ) and output the prediction result at the current moment.

[0101] It is worth mentioning that when using the above-mentioned first model to predict the model life, it is necessary to first train the second model and adjust the parameters of the second model to obtain the above-mentioned first model. Specifically, the first rolling mill is a rolling mill working in the first working condition, and the first model is obtained by the following steps:

[0102] Constructing a third sample set, wherein the third sample set includes a plurality of different third samples, each of the third samples includes sixth operating data that has been preprocessed and a corresponding label, the sixth operating data includes historical operating data of the rolling mill operating under the first operating condition, and the label is an artificially determined life prediction result;

[0103] Training the second model using the third sample set to obtain a fifth model;

[0104] Constructing a fourth sample set, wherein the fourth sample set includes a plurality of different fourth samples, each of which includes seventh operating data that has been preprocessed and a corresponding label, and the seventh operating data includes historical operating data of the rolling mill operating in the first working condition;

[0105] The fifth model is verified by using the fourth sample set, and when the verification passes, the fifth model is determined to be the first model.

[0106] The first model is used to predict the life of the rolling mill working in the first working condition. To ensure the accuracy of the model prediction, it is necessary to collect historical operating data of the rolling mill working in the first working condition to train and verify the second model.

[0107] The third sample set is used to train the model, and the fourth sample set is used to verify the trained model. Each third sample in the third sample set includes the sixth operating data of data preprocessing. The content of each third sample is different. Each third sample can be set to include operating data corresponding to different acquisition time periods and / or operating data corresponding to different rolling mills. Similarly, each fourth sample in the fourth sample set includes the seventh operating data of data preprocessing. The content of each fourth sample is different. Each fourth sample can be set to include operating data corresponding to different acquisition time periods and / or operating data corresponding to different rolling mills. In addition, the data in the third sample set used for training is different from the data in the fourth sample set used for verification. After the model is trained, model verification is also required to ensure the generalization ability of the model, that is, the model can also show good performance when facing unseen data. Among them, the verification method for the above model can be set according to demand, for example, it can be cross-validation.

[0108] In an embodiment of the present application, historical operating data of a rolling mill operating in a first operating condition is collected, and after data preprocessing is performed (the specific implementation method of the data preprocessing can refer to the above embodiment), corresponding labels are added to obtain a third sample set and a fourth sample set, respectively; the first model obtained by training and verifying the second model using the third sample set and the fourth sample set can more accurately predict the life of the rolling mill under the first operating condition.

[0109] It should be noted that the specific method for training the second model can be determined based on the type of model and the sample size.

[0110] Taking GBT as an example, the second model builds multiple decision tree models and optimizes the next tree based on the error of the previous tree. The gradient descent algorithm of the loss function is used to update the GBT model parameters, and its formula is:

[0111] F m (x) = F m-1 (x)+γh m (x)

[0112] Among them, F m (x) is the prediction result of the mth tree, γ is the learning rate, h m (x) is the base learner of the mth tree.

[0113] Taking the second model as an LSTM model as an example, the Adam optimizer can be used to train the LSTM, the initial learning rate is set to 0.001, and the early stopping mechanism is used to prevent overfitting.

[0114] When the number of training samples is insufficient, the model parameters can be updated using the above training method, and then the Monte Carlo simulation algorithm can be used to perform random sampling analysis on different operating scenarios to generate multiple possible operating scenarios, and the Bayesian update method can be used to dynamically adjust the model parameters, as follows:

[0115] Uncertainty modeling of complex scenarios is achieved through Monte Carlo simulation. Monte Carlo simulation generates a large number of possible operating scenarios by randomly sampling input variables, covering various potential input conditions or environmental changes, providing the model with more extensive training data, thereby improving its adaptability to unknown scenarios. The number of operating scenarios generated by the Monte Carlo simulation is dynamically adjusted according to the prediction accuracy requirements, and the default value is 1000 random samplings.

[0116] The Bayesian update method is a statistical method that dynamically adjusts model parameters based on new observation data. Its core is to combine the impact of new data with existing prior knowledge to update the posterior distribution of the model. During the training process, the Bayesian update gradually corrects the parameters of the model based on the samples generated by the Monte Carlo simulation to make it closer to the actual operating characteristics. 1000 training samples are generated by random sampling. The Bayesian update fine-tunes the model parameters based on each training sample, and updates the model parameters based on the maximum a posteriori probability estimate to dynamically adapt to changes in real-time data. 1000 sets of model parameters determined based on 1000 training samples are obtained, and then the final model parameters can be determined based on the 1000 sets of model parameters (the average or standard deviation of each parameter, etc.), that is, the first model is determined. The core formula of the above Bayesian update is:

[0117]

[0118] P(θ|D) is the posterior probability of the model parameter θ under the observed data D, P(D|θ) is the likelihood function, P(θ) is the prior probability, and P(D) is the marginal likelihood.

[0119] The above embodiment adopts the Bayesian update method to dynamically adjust the model parameters, and combines the Monte Carlo simulation algorithm to perform random sampling analysis on different operating scenarios. It can dynamically adjust the parameters so that the model is continuously optimized over time, and can cover a variety of operating scenarios, thereby improving the model's adaptability and generalization ability to complex environments.

[0120] Optionally, the method further comprises:

[0121] When receiving the label corresponding to the target operation data, adding a first sample to a first sample set, the first sample consisting of the target operation data and the label corresponding to the target operation data, the label being an artificially determined life prediction result;

[0122] When the sample features in the first sample set meet preset requirements, the parameters of the first model are updated by an online learning method based on the first sample set to obtain a third model.

[0123] In this implementation, online learning is an incremental learning method, where the model gradually accepts single or small batch samples and updates model parameters based on the information of the new samples without retraining the entire model. When the sample features in the first sample set meet the preset requirements, online learning is triggered. The above preset requirements may be, for example, that the number of samples reaches the preset requirements, or that the model prediction accuracy reaches the preset requirements for triggering online learning, etc.

[0124] It is worth mentioning that the online learning method can be based on algorithms such as the recursive least squares algorithm or the gradient descent algorithm. Taking the gradient descent algorithm as an example, its core formula is:

[0125]

[0126] Among them, θ t is the model parameter, η is the learning rate, is the gradient of the loss function with respect to the parameters.

[0127] The learning rate in the above formula is a hyperparameter, which can be set to a fixed value, or adjusted according to the training situation of the model, or the hyperparameter optimization method based on the genetic algorithm can automatically search for the optimal hyperparameter combination during the model update process to improve the adaptability and generalization ability of the model.

[0128] The embodiment of the present application adopts an online learning method to continuously optimize the model according to actual operating conditions and prediction results, which is conducive to more accurate life prediction and management.

[0129] Optionally, the first rolling mill is a rolling mill operating in a first operating condition, and the method further comprises:

[0130] Acquire a second sample set, wherein the second sample set includes a plurality of different second samples, each of which includes fifth operating data that has undergone the data preprocessing and a corresponding label, the fifth operating data is historical operating data of the rolling mill operating in the second working condition, and the label is an artificially determined life prediction result;

[0131] Based on the second sample set, the parameters of the first model are updated by using a transfer learning method to obtain a fourth model, and the fourth model is used to determine the life prediction result of the rolling mill working under the second working condition.

[0132] In this implementation, transfer learning is a method that uses the knowledge of an existing model in a certain field or task (source field / task) to improve the learning efficiency and effect in the target field / task. When it is necessary to predict the life of a rolling mill under a new working condition, some parameters of the existing first model can be used to collect data under the new working condition to form a new training sample, and another part of the parameters of the first model can be updated through the new training sample.

[0133] See also Figure 2 The present application also provides a system for predicting the life of a rolling mill. Figure 2 The system shown includes a data acquisition module 210 , a data preprocessing module 220 , a model training module 230 , a prediction module 240 , and a model updating module 250 .

[0134] The data acquisition module 210 is connected to the data preprocessing module 220 and is used to collect the rolling mill operation data in real time through sensors and monitoring systems, including parameters such as temperature, pressure, vibration and load;

[0135] The data preprocessing module 220 is connected to the model training module 230 and is used to preprocess the collected data, including cleaning, normalization and feature extraction, etc., to remove noise and outliers and generate high-quality input data;

[0136] A model training module 230, connected to the prediction module 240, is used to train a second model using historical data and preprocessed real-time data;

[0137] The prediction module 240 is connected to the model update module 250 and is used to make a real-time prediction of the life of the rolling mill based on the trained first model, and output the prediction result and the remaining life estimate;

[0138] The model updating module 250 is connected to the prediction module 240 and is used to continuously update and optimize the prediction model according to the actual operation situation and prediction results, including online learning update and transfer learning update.

[0139] In this embodiment, the data acquisition module 210, the data preprocessing module 220, the model training module 230, the prediction module 240 and the model updating module 250 can be integrated into a unified system through a software integration platform (such as LabVIEW, SCADA system) to ensure that each module works in coordination.

[0140] Data acquisition module 210

[0141] Function: Real-time collection of mill operation data. Hardware: High-precision sensors (temperature sensors, pressure sensors, accelerometers, strain gauges, etc.). Software: Data acquisition software (such as LabVIEW, SCADA system). Parameters: temperature, pressure, vibration, load, etc. Implementation: Install sensors at key locations, monitor and record data in real time through the data acquisition system to ensure the real-time and accuracy of the data.

[0142] Data preprocessing module 220

[0143] Functions: data cleaning, normalization, feature extraction, etc.; Algorithms: missing value processing (such as interpolation, filling), noise removal (such as filtering), normalization (MinMax, ZScore) and feature extraction (PCA, time-frequency analysis), etc.; Software: Python, MATLAB; Implementation: Write data preprocessing scripts to clean the collected data, remove noise and outliers, perform normalization, and extract key features to generate high-quality input data.

[0144] Model training module 230

[0145] Function: training life prediction model; Algorithm: deep learning (LSTM or GRU), reinforcement learning (Qlearning or DQN); Framework: TensorFlow, PyTorch; Data: preprocessed historical data and preprocessed real-time data; Implementation: using historical data and real-time data, using adaptive learning algorithm to train life prediction model, select appropriate neural network structure and hyperparameters, and verify model performance through cross-validation and test set.

[0146] Prediction module 240

[0147] Function: Real-time prediction of rolling mill life; Algorithm: Based on the trained life prediction model; Implementation: Input the preprocessed real-time data into the trained model, perform life prediction, output the prediction results, and visualize the prediction results.

[0148] Model update module 250

[0149] Function: Update and optimize the prediction model; Algorithm: Online learning, transfer learning; Implementation: According to the actual operation situation and prediction results, the model parameters are continuously updated through online learning, and the model is quickly adapted to new environments and new working conditions through transfer learning to improve the accuracy and adaptability of the model.

[0150] In some embodiments, in order to demonstrate the effect of the present application, a prediction accuracy improvement curve and an adaptability improvement curve can be drawn. The prediction accuracy improvement curve is used to compare the prediction accuracy of the life prediction results of the present application with the prediction accuracy of the life prediction results in the prior art; the adaptability improvement curve is used to compare the adaptability of the prediction method in the present application with the adaptability of the prediction method in the prior art. The following code can be used to implement it:

[0151] ```Python

[0152] import matplotlib.pyplot as plt

[0153] import numpy as np

[0154] #data

[0155] time = np.arange(1,25)

[0156] old_prediction_accuracy=np.random.normal(70,5,size=time.shape)

[0157] new_prediction_accuracy=np.random.normal(90,2,size=time.shape)

[0158] #Draw the prediction accuracy improvement curve

[0159] plt.figure(figsize=(10,6))

[0160] plt.plot(time,old_prediction_accuracy,label='Old PredictionAccuracy',marker='o')

[0161] plt.plot(time,new_prediction_accuracy,label='New PredictionAccuracy',marker='o')

[0162] plt.xlabel('Time(hours)')

[0163] plt.ylabel('Prediction Accuracy(%)')

[0164] plt.title('Prediction Accuracy Improvement')

[0165] plt.legend()

[0166] plt.grid(True)

[0167] plt.show()

[0168] # Adaptive improvement curve data

[0169] conditions = np.arange(1, 11)

[0170] old_adaptability_score = np.random.normal(5, 1.5, size = conditions.shape)

[0171] new_adaptability_score = np.random.normal(8, 1, size = conditions.shape)

[0172] # Plot the adaptive improvement curve

[0173] plt.figure(figsize=(10, 6))

[0174] plt.plot(conditions, old_adaptability_score, label='Old AdaptabilityScore', marker='o')

[0175] plt.plot(conditions, new_adaptability_score, label='New AdaptabilityScore', marker='o')

[0176] plt.xlabel('Operating Conditions')

[0177] plt.ylabel('Adaptability Score(010)')

[0178] plt.title('Adaptability Improvement')

[0179] plt.legend()

[0180] plt.grid(True)

[0181] plt.show()

[0182] In order to more clearly understand the technical solutions of the embodiments of the present application, two specific embodiments are provided below to illustrate the present application.

[0183] Example 1: Prediction of rolling mill life in a hot rolling mill

[0184] 1. Background

[0185] In hot rolling mills, rolling mills are subjected to high temperature, high pressure and high load working environment. Life prediction is crucial to ensure production safety and reduce downtime.

[0186] 2. Implementation Method

[0187] Data acquisition module 210: Hardware, temperature sensors, pressure sensors, vibration sensors and strain gauges are installed; temperature sensors are installed on rolling mill bearings and gears to monitor temperature. Pressure sensors are installed in hydraulic systems and cooling systems to monitor pressure; vibration sensors are installed on rolling mill bearings and motors to monitor vibration; strain gauges are installed on key structural components to monitor stress; software: LabVIEW system is used to collect and record data in real time;

[0188] Data preprocessing module 220: data cleaning, processing missing values ​​in sensor data and removing noise; interpolation method to fill missing values; low-pass filtering to remove noise in vibration signals; data normalization: using ZScore normalization method to process temperature, pressure, vibration and stress data; feature extraction: performing PCA dimensionality reduction and extracting main features; performing time-frequency analysis and extracting time domain and frequency domain features.

[0189] Model training module 230: Deep learning model: Use LSTM model to process time series data and capture nonlinear relationships in data; or, reinforcement learning model: build a reinforcement learning environment and use DQN algorithm to optimize prediction strategy. Training and validation: Use historical operation data to train the model and validate the model performance through cross-validation and test set.

[0190] Prediction module 240: Real-time prediction, inputting the pre-processed real-time data into the trained model, predicting the life of the rolling mill, and outputting the prediction results and remaining life estimation.

[0191] Model update module 250: Online learning, real-time update of LSTM model parameters to ensure that the model adapts to dynamic changes; transfer learning, using transfer learning methods to quickly adapt to new production environments and working conditions.

[0192] The above modules are integrated through the LabVIEW system to ensure data flow and module collaboration. The above system prediction can achieve accurate prediction of the life of the rolling mill, reduce downtime and improve production efficiency.

[0193] Example 2: Prediction of rolling mill life in a cold rolling mill

[0194] 1. Background

[0195] In a cold rolling mill, the working environment of the rolling mill is relatively low temperature, but it is still subjected to large mechanical stress and vibration. Life prediction is very important to ensure production continuity and reduce maintenance costs.

[0196] 2. Implementation process

[0197] Data acquisition module 210: Hardware, install temperature sensors, pressure sensors, vibration sensors and strain gauges. Temperature sensors: Installed on rolling mill bearings and gears to monitor temperature. Pressure sensors: Installed in hydraulic systems to monitor pressure. Vibration sensors: Installed on rolling mill bearings and motors to monitor vibration. Strain gauges: Installed on key structural components to monitor stress. Software: Use SCADA system to collect and record data in real time.

[0198] Data preprocessing module 220: Data cleaning, processing missing values ​​in sensor data and removing noise; interpolation method to fill missing values; high-pass filtering to remove low-frequency noise in vibration signals; data normalization, using MinMax normalization method to process temperature, pressure, vibration and stress data; feature extraction: PCA dimensionality reduction to extract main features; time-frequency analysis to extract time domain and frequency domain features.

[0199] Model training module 230, deep learning model, using GRU model, processing time series data, capturing nonlinear relationships in data; or, reinforcement learning model: building a reinforcement learning environment, using Qlearning algorithm to optimize prediction strategy. Training and validation: using historical operation data for model training, and validating model performance through cross validation and test set.

[0200] Prediction module 240: inputs the preprocessed real-time data into the trained GRU model to predict the life of the rolling mill, and outputs the prediction results and the remaining life estimation.

[0201] Model update module 250: Online learning, real-time update of GRU model parameters to ensure that the model adapts to dynamic changes; transfer learning: using transfer learning methods to quickly adapt to new production environments and working conditions.

[0202] The above modules are integrated through the SCADA system to ensure data flow and module collaboration. The above system can achieve accurate prediction of mill life, reduce downtime and improve production efficiency.

[0203] The present application also uses the existing technology to predict the life of hot-rolled steel mills and cold-rolled steel mills, and determines the adaptability of the existing model, so as to compare the prediction accuracy of the present application with the prediction accuracy of the existing technology, and to compare the adaptability of the present application with the adaptability of the existing technology. For details, please refer to Figure 3 and Figure 4 . Figure 3 The vertical axis represents the prediction accuracy percentage, the horizontal axis represents time, the upper curve is the prediction accuracy curve chart obtained by using the method or system of the present application, and the lower curve is the prediction accuracy curve chart obtained by using the method or system of the prior art. It can be seen that the prediction accuracy of the present application is much higher than that of the prior art. Figure 4 The middle vertical axis represents the adaptability score, the horizontal axis represents the operating conditions, the upper curve is the adaptability score curve obtained by using the method or system of the present application, and the lower curve is the adaptability score curve obtained by using the method or system of the prior art. It can be seen that the adaptability of the present application is also higher than that of the prior art.

[0204] pass Figure 3 and Figure 4 It can be concluded that the present application significantly improves the prediction accuracy, and the prediction accuracy is relatively stable; and the present application shows higher adaptability under different working conditions, with higher scores and smaller fluctuations.

[0205] It is worth mentioning that the rolling mill life prediction system in the embodiment of the present application belongs to the same concept as the rolling mill life prediction method mentioned above, and its specific implementation method can refer to the above embodiment, and will not be described in detail to avoid repetition.

[0206] In the embodiments of the present application, through modular design and efficient data preprocessing, it is possible to make full use of real-time data and historical data to improve data utilization and prediction effects; through a unified prediction model, the influence of human subjective factors is reduced, and the standardization and consistency of the prediction method are improved; the first model adopts deep learning and reinforcement learning algorithms, which can effectively capture the complex nonlinear relationships in the operation process of the rolling mill and improve the prediction accuracy; the first model can be updated and optimized through online learning and transfer learning, adapt to the dynamic changes in the operating environment and conditions of the rolling mill, and improve the accuracy and reliability of the prediction.

[0207] See also Figure 5The present application also provides a device for predicting the life of a rolling mill, including:

[0208] A receiving module 510, configured to receive first operating data of a first rolling mill sent by at least one sensor;

[0209] A data preprocessing module 520 is used to perform data preprocessing on the first operation data to obtain target operation data, wherein the data preprocessing includes at least one of the following: data cleaning processing, data normalization processing, and feature extraction processing;

[0210] The prediction module 530 is used to input the target operating data into the first model to obtain a first life prediction result corresponding to the target operating data, wherein the first model is a trained second model, and the second model is a deep learning model or a reinforcement learning model.

[0211] Optionally, the data preprocessing includes feature extraction processing, and the feature extraction processing includes time-frequency feature extraction processing based on wavelet transform.

[0212] Optionally, the data preprocessing includes data cleaning processing, data normalization processing and feature extraction processing. The data preprocessing module 520 is specifically used to:

[0213] Performing data cleaning processing on the first operation data to obtain second operation data, wherein the data cleaning processing includes at least one of outlier removal processing, missing value compensation processing, and denoising processing;

[0214] performing data normalization processing on the second operation data to obtain third operation data;

[0215] Performing time-frequency feature extraction processing on the third operation data based on wavelet transform to obtain fourth operation data;

[0216] The fourth operation data is subjected to main feature extraction processing based on principal component analysis technology PCA to obtain the target operation data.

[0217] Optionally, the device further comprises a first updating module, configured to

[0218] When receiving the label corresponding to the target operation data, adding a first sample to a first sample set, the first sample consisting of the target operation data and the label corresponding to the target operation data, the label being an artificially determined life prediction result;

[0219] When the sample features in the first sample set meet preset requirements, the parameters of the first model are updated by an online learning method based on the first sample set to obtain a third model.

[0220] Optionally, the first rolling mill is a rolling mill operating in a first operating condition, and the device further comprises a second updating module, which is used to:

[0221] Acquire a second sample set, wherein the second sample set includes a plurality of different second samples, each of which includes fifth operating data that has undergone the data preprocessing and a corresponding label, the fifth operating data is historical operating data of the rolling mill operating in the second working condition, and the label is an artificially determined life prediction result;

[0222] Based on the second sample set, the parameters of the first model are updated by using a transfer learning method to obtain a fourth model. The third model is used to determine the life prediction result of the rolling mill operating under the second working condition.

[0223] Optionally, the first rolling mill is a rolling mill operating in a first working condition, and the first model is obtained by the following steps:

[0224] Constructing a third sample set, wherein the third sample set includes a plurality of different third samples, each of the third samples includes sixth operating data that has been preprocessed and a corresponding label, the sixth operating data includes historical operating data of the rolling mill operating under the first operating condition, and the label is an artificially determined life prediction result;

[0225] Training the second model using the third sample set to obtain a fifth model;

[0226] Constructing a fourth sample set, wherein the fourth sample set includes a plurality of different fourth samples, each of which includes seventh operating data that has been preprocessed and a corresponding label, and the seventh operating data includes historical operating data of the rolling mill operating in the first working condition;

[0227] The fifth model is verified by using the fourth sample set, and when the verification passes, the fifth model is determined to be the first model.

[0228] Optionally, each of the at least one sensor is used to collect operating data of the first rolling mill at a preset frequency, and the at least one sensor includes a temperature sensor and a pressure sensor.

[0229] It should be noted that the prediction device for the life of a rolling mill provided in the above embodiment and the prediction method for the life of a rolling mill provided in the above embodiment belong to the same concept, wherein the specific manner in which each module and unit performs operations has been described in detail in the method embodiment, and will not be repeated here. In practical applications, the #device provided in the above embodiment can allocate the above functions to different functional modules as needed, that is, divide the internal structure of the device into different functional modules to complete all or part of the functions described above, and this is not limited here.

[0230] An embodiment of the present application also provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the electronic device implements the rolling mill life prediction method provided in the above-mentioned embodiments.

[0231] Another aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor of a computer, causes the computer to execute the above-mentioned method for predicting the life of a rolling mill. The computer-readable storage medium may be included in the electronic device described in the above embodiment, or may exist independently without being assembled into the electronic device.

[0232] Another aspect of the present application also provides a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. A processor of a computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the method provided in each of the above embodiments.

[0233] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by a person of ordinary skill in the art without departing from the spirit and technical ideas disclosed by the present invention shall still be covered by the claims of the present invention.

Claims

1. A method for predicting the life of a rolling mill, characterized in that: include: receiving first operation data of a first rolling mill sent by at least one sensor; Performing data preprocessing on the first operation data to obtain target operation data, wherein the data preprocessing includes at least one of the following: data cleaning processing, data normalization processing, and feature extraction processing; The target operating data is input into a first model to obtain a life prediction result corresponding to the target operating data, wherein the first model is a trained second model, and the second model is a deep learning model or a reinforcement learning model.

2. The method according to claim 1, characterized in that The data preprocessing includes feature extraction processing, and the feature extraction processing includes time-frequency feature extraction processing based on wavelet transform.

3. The method according to claim 1, characterized in that The data preprocessing includes data cleaning processing, data normalization processing and feature extraction processing, and the data preprocessing of the first operation data includes: Performing data cleaning processing on the first operation data to obtain second operation data, wherein the data cleaning processing includes at least one of outlier removal processing, missing value compensation processing, and denoising processing; performing data normalization processing on the second operation data to obtain third operation data; Performing time-frequency feature extraction processing on the third operation data based on wavelet transform to obtain fourth operation data; The fourth operation data is subjected to main feature extraction processing based on principal component analysis technology PCA to obtain the target operation data.

4. The method according to claim 1, characterized in that The method further comprises: When receiving the label corresponding to the target operation data, adding a first sample to a first sample set, the first sample consisting of the target operation data and the label corresponding to the target operation data, the label being an artificially determined life prediction result; When the sample features in the first sample set meet preset requirements, the parameters of the first model are updated by an online learning method based on the first sample set to obtain a third model.

5. The method according to claim 1, characterized in that The first rolling mill is a rolling mill operating in a first operating condition, and the method further comprises: Acquire a second sample set, wherein the second sample set includes a plurality of different second samples, each of which includes fifth operating data that has undergone the data preprocessing and a corresponding label, the fifth operating data is historical operating data of the rolling mill operating in the second working condition, and the label is an artificially determined life prediction result; Based on the second sample set, the parameters of the first model are updated by using a transfer learning method to obtain a fourth model, and the fourth model is used to determine the life prediction result of the rolling mill working under the second working condition.

6. The method according to any one of claims 1 to 5, characterized in that The first rolling mill is a rolling mill operating in a first working condition, and the first model is obtained by the following steps: Constructing a third sample set, wherein the third sample set includes a plurality of different third samples, each of the third samples includes sixth operating data that has been preprocessed and a corresponding label, the sixth operating data includes historical operating data of the rolling mill operating under the first operating condition, and the label is an artificially determined life prediction result; Training the second model using the third sample set to obtain a fifth model; Constructing a fourth sample set, wherein the fourth sample set includes a plurality of different fourth samples, each of which includes seventh operating data that has undergone the data preprocessing and a corresponding label, and the seventh operating data includes historical operating data of the rolling mill operating in the first operating condition; The fifth model is verified by using the fourth sample set, and when the verification passes, the fifth model is determined to be the first model.

7. The method according to any one of claims 1 to 5, characterized in that Each of the at least one sensor is used to collect operating data of the first rolling mill at a preset frequency, and the at least one sensor includes a temperature sensor and a pressure sensor.

8. A device for predicting the life of a rolling mill, characterized in that: include: A receiving module, used for receiving first operation data of a first rolling mill sent by at least one sensor; A data preprocessing module, used to perform data preprocessing on the first operation data to obtain target operation data, wherein the data preprocessing includes at least one of the following: data cleaning processing, data normalization processing and feature extraction processing; A prediction module is used to input the target operating data into a first model to obtain a first life prediction result corresponding to the target operating data, wherein the first model is a trained second model, and the second model is a deep learning model or a reinforcement learning model.

9. A device, characterized in that: include: one or more processors and memory, A computer program is stored in the memory, and when the one or more processors execute the computer program, the device is caused to perform the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, which, when executed by one or more processors, causes the device to perform the method according to any one of claims 1 to 7.

Images