Intelligent cogging rolling method and system for hot-rolled H-shaped steel based on data driving

Abstract

The invention discloses a hot-rolled H-shaped steel intelligent blooming rolling method and system based on data driving. The method comprises the steps that parameters of an automatic control system in the rolling process are initialized, and a technological parameter database is established; collecting process parameters and preprocessing the process parameters; executing process parameter optimization calculation; inputting the preprocessed working condition data into a data-driven decision core, extracting high-dimensional nonlinear features by adopting a deep neural network, and outputting initial predicted values of process parameters; self-adaptive adjustment based on real-time feedback is completed through a reinforcement learning algorithm; meanwhile, an optimization algorithm is adopted to carry out global optimization on hyper-parameters, and a technological parameter control strategy is obtained; performing quality state evaluation and quality state difference analysis on the process parameters acquired in real time and the initial predicted values of the process parameters, and optimizing the process parameters; judging whether a production end condition is met or not, if not, returning to the step 2, and if yes, executing the step 6; and outputting the optimized process parameters.

Description

Technical Field

[0001] This invention relates to the field of industrial intelligence technology, and in particular to a data-driven intelligent billet rolling method and system for hot-rolled H-beams. Background Technology

[0002] H-beams are an economical and efficient structural steel profile with a more optimized cross-sectional area distribution and a more reasonable strength-to-weight ratio. They are named for their cross-section resembling the letter "H". Due to their cross-sectional shape, H-beams have significantly better section modulus, moment of inertia, and corresponding strength than ordinary I-beams of the same weight. H-beams are characterized by wide flanges, thin walls, and light weight. They also offer a wide range of heights, various specifications, and flexible applications, making them widely used in truss structures. They can significantly improve load-bearing capacity compared to ordinary I-beams while saving 10% to 40% of metal.

[0003] However, existing technologies still face numerous challenges in the hot rolling production of H-beams. Traditional rolling techniques suffer from strong reliance on experience and poor adaptability to dynamic conditions in multi-pass deformation distribution. Furthermore, the hot rolling process of H-beams involves intense plastic deformation, exhibiting both geometric and physical nonlinear characteristics, and its complex initial conditions make it difficult to describe using simplification, assumptions, experimental and empirical data, graphical methods, or models. In addition, most existing rolling force models are based on traditional empirical formulas or theoretical models. While these can provide rough guidance under certain conditions, they typically ignore many complex factors in actual production, such as variations in rolling process parameters and mill vibration, resulting in low prediction accuracy.

[0004] Furthermore, the production process of H-beams suffers from problems such as rigid or inflexible process parameters, resulting in poor adaptability to fluctuations in incoming materials. Quality control relies heavily on operator experience, leading to insufficient stability. Simultaneously, the production process lacks proactive forecasting and intervention capabilities, resembling "remedial" control. These issues result in low production efficiency, low yield, and difficulty in achieving stable "lights-out" production. To address the challenges of quality improvement, yield enhancement, and adaptability to fluctuations in incoming materials during the hot-rolled H-beam billet rolling process, existing technologies employ the following methods:

[0005] For example, CN119993342A discloses a rolling force prediction system for H-beams based on machine learning technology. This system includes a parameter calculation module that conforms to actual production conditions, a data preprocessing module, and a rolling force prediction module based on optimized random forest. This system can accurately predict the rolling force of H-beams and optimize the rolling process, avoiding the shortcomings of traditional methods that rely on human experience, while improving production efficiency and reducing scrap rates. However, the system still suffers from insufficient data quality and reliability in its data preprocessing module.

[0006] CN117519002A discloses an adaptive method and system for presetting rolling force in a cold continuous rolling digital twin production line. This system develops a cold continuous rolling digital twin production line to achieve dynamic real-time mapping of the production process and simulated rolling of key operational behaviors. Key process parameters are then input into the high-precision cold continuous rolling digital twin production line to simulate operational behavior and obtain predicted rolling force values ​​for the simulated process. Based on a metallurgical mechanism model, relevant parameters of the production process are screened to establish a rolling force deviation prediction model based on a TS fuzzy neural network, and deviations are corrected for the predicted rolling force. This system solves the problems of low setting accuracy and poor self-learning ability in traditional rolling force theoretical calculation models. However, this system still has problems with insufficient modeling accuracy and real-time performance in digital modeling. The existing technology has the following drawbacks:

[0007] 1. Traditional H-beam rolling technology faces problems such as strong reliance on experience and poor adaptability to dynamic working conditions in multi-pass deformation distribution, resulting in low production efficiency and low yield.

[0008] 2. Most existing rolling force models are based on traditional empirical formulas or theoretical models, which usually ignore many complex factors in actual production, such as changes in rolling process parameters and mill vibration, resulting in low prediction accuracy and difficulty in meeting the requirements of high-precision production.

[0009] 3. In the current H-beam steel production process, process parameters are often fixed or have a rigid adjustment range, making them poorly adaptable to fluctuations in incoming materials and difficult to achieve stable production in a "lights-out factory" style.

[0010] 4. Existing technologies lack effective forward-looking prediction and intervention capabilities, and mainly rely on the experience of operators for quality control, resulting in insufficient stability of quality control and a high risk of batch quality accidents.

[0011] 5. The existing digital modeling and machine learning technologies are not fully applied in the H-beam rolling process, making it difficult to achieve intelligent monitoring and optimization of the entire process.

[0012] Therefore, there is an urgent need for an intelligent billet rolling method that can perceive in real time, make intelligent decisions, and execute precisely to address the many shortcomings of existing technologies. Summary of the Invention

[0013] In view of the above-mentioned problems in the existing technology, the purpose of this invention is to solve the above problems by providing a data-driven intelligent billet rolling method for hot-rolled H-beams.

[0014] Another objective of this invention is to provide a data-driven intelligent billet rolling system for hot-rolled H-beams.

[0015] To address the above problems, this invention employs the following technical solution: a data-driven intelligent billet rolling method for hot-rolled H-beams, the method comprising the following steps:

[0016] Step 1: Initialize the parameters of the automated control system for the rolling process and establish a process parameter database;

[0017] Step 2: Collect process parameters and preprocess the process parameters;

[0018] Step 3: Perform process parameter optimization calculations; input the preprocessed operating data into the data-driven decision core, use a deep neural network to extract high-dimensional nonlinear features and output the initial predicted values ​​of process parameters; complete adaptive adjustment based on real-time feedback through reinforcement learning algorithm; at the same time, use optimization algorithm to perform global optimization of hyperparameters and process parameter control strategy.

[0019] Step 4: Conduct a quality status assessment and quality status difference analysis on the real-time collected process parameters and the initial predicted values ​​of the process parameters, and optimize the process parameters.

[0020] Step 5: Determine whether the production end conditions have been met. If not, return to step 2; otherwise, proceed to step 6.

[0021] Step 6: Output the optimized process parameters.

[0022] Furthermore, step 1 includes:

[0023] Step 101: Establish a default value library for process parameters, wherein the process parameters are basic parameters including the reduction rate, rolling force, and speed for each pass;

[0024] Step 102: Set the adjustment range of process parameters, including the maximum allowable deviation value;

[0025] Step 103: Establish an adaptive model for process parameters, including the correlation rules between process parameters.

[0026] Furthermore, step 2 includes:

[0027] Step 201: Deploy a multi-sensor array at the inlet of the billet mill, between each stand of the rolling mill, and at the outlet;

[0028] Step 202: Initially position the billet and detect surface defects online using a machine vision unit to obtain initial position data and surface defect data of the billet;

[0029] Step 203: Use a laser scanner to measure the three-dimensional contour dimensions of the rolled piece in real time after each pass, and obtain the three-dimensional contour data of the rolled piece in each pass.

[0030] Step 204: Use a multi-band infrared thermal imager to collect the temperature field distribution of the entire length and width of the rolled piece, and obtain the temperature data of the rolled piece;

[0031] Step 205: Preprocess and standardize the collected data.

[0032] Furthermore, step 3 includes:

[0033] Step 301: Input the preprocessed data into the data-driven decision-making core;

[0034] Step 302: Use a deep neural network to extract features from the input data and predict process parameters;

[0035] Step 303: Optimize process parameters using reinforcement learning algorithms to enable the model to adapt to new production conditions;

[0036] Step 304: Use the Northern Eagle optimization algorithm to optimize the number of decision trees and the maximum depth of trees in the random forest model.

[0037] Furthermore, step 4 includes:

[0038] Step 401: Evaluate the quality status based on the real-time acquired process parameters and simulation calculation results;

[0039] Step 402: Filter outliers in the data using the Isolation Forest algorithm;

[0040] Step 403: Use a tree model to select features and filter out the features most relevant to rolling force prediction;

[0041] Step 404: Feed back the optimized process parameters to the slab rolling process for control.

[0042] Furthermore, step 6 includes:

[0043] Step 601: Export the optimized process parameters in Excel spreadsheet format;

[0044] Step 602: Visualize the process parameters;

[0045] Step 603: Generate a production suggestion report.

[0046] This application also provides a system for implementing the data-driven intelligent billet rolling method for hot-rolled H-beams, the system comprising:

[0047] The execution layer receives optimized process parameter setting instructions and converts them into adaptive control signals for the rolling process through a fuzzy PID controller; it drives the adaptive actuator to adjust the execution parameters; and it monitors and dynamically corrects the position and attitude of the rolled piece in real time based on machine vision or laser detection feedback.

[0048] The output module is used to visualize and generate reports on the key raw data of real-time control signals, actuator status, and process stability indicators generated by the output execution layer during the control process.

[0049] The sensing layer is used to collect process status data of the rolling process in real time through a multi-sensor array;

[0050] The data preprocessing module is used to convert the received sensing layer process status data into structured data by removing equipment noise and electromagnetic interference through signal filtering, eliminating sensor dimension differences and numerical range deviations through standardization and normalization.

[0051] The decision-making level uses a deep neural network to extract high-dimensional nonlinear features from the structured data processed by the data preprocessing module and makes initial predictions of process parameters and quality indicators.

[0052] By using reinforcement learning algorithms, process parameters are dynamically optimized and adaptively adjusted based on real-time production feedback.

[0053] The optimization algorithm is used to globally optimize the hyperparameters of the key model, and the key variables are selected by combining the feature selection method, and the optimized process parameter control strategy is output.

[0054] Compared with the prior art, the beneficial technical effects of the present invention are as follows:

[0055] 1. The system described in this application, by constructing an intelligent rolling system consisting of a three-layer structure of "perception-decision-execution," achieves real-time monitoring and intelligent management of the entire process, effectively solving the problems of strong reliance on experience and poor dynamic adaptability in traditional technologies, and significantly improving production efficiency and yield. Simultaneously, it achieves stable production in a "lights-out factory" manner, overcoming the uncertainties caused by human factors in traditional production processes.

[0056] 2. The method described in this application employs a multi-sensor array, including machine vision units, laser scanners, and multi-band infrared thermal imagers, to achieve real-time monitoring and dynamic analysis of the entire billet process, effectively improving the controllability and stability of the production process and ensuring the stability of product quality.

[0057] 3. The hybrid model based on deep neural networks and reinforcement learning provided in this application establishes a theoretical system for dynamic optimization of process parameters, overcomes the limitations of traditional empirical formulas or theoretical models that ignore complex actual factors, and significantly improves the accuracy and reliability of process parameter prediction.

[0058] 4. This application achieves forward-looking prediction and intervention in the production process through a digital twin-driven quality prediction and correction module, effectively avoiding defects caused by uneven temperature, dimensional deviations, etc., and reducing the defect rate of folding cracks, peeling ears, etc. by more than 50%.

[0059] 5. This application employs an adaptive actuator and an intelligent centering and guidance system, which enables precise control and automation of the rolling process, avoids errors caused by manual operation, and further improves the stability and reliability of the production process. Attached Figure Description

[0060] Figure 1 This is a system structure diagram shown in the embodiments of this application;

[0061] Figure 2 This is a flowchart of the method described in the embodiments of this application. Detailed Implementation

[0062] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "vertical," "horizontal," and "inner," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. Furthermore, unless otherwise explicitly specified and limited, the terms "installed," "connected," and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication between two elements. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0064] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0065] Example 1

[0066] like Figure 1 , Figure 2 As shown, this application provides a data-driven intelligent billet rolling method for hot-rolled H-beams, including the following steps:

[0067] Step 1: Initialize system parameters and establish a process parameter database.

[0068] Step 101: Establish a default value library for process parameters, including basic parameters such as reduction rate, rolling force, and speed for each pass. Specifically, the reduction rate for the first pass is set to 28%, the rolling force to 22.6 kN, and the speed to 1.8 m / s; the reduction rates for the second to fourth passes are 25%, 28%, and 30%, respectively, with rolling forces of 41.6 kN, 58.8 kN, and 54.4 kN, respectively, and the speed is maintained at 2.0 m / s; the reduction rate for the fifth pass is 30%, the rolling force is 38.4 kN, and the speed is 2.4 m / s.

[0069] Step 102: Set the adjustment range of process parameters, including the maximum allowable deviation. The reduction rate adjustment range is 20%-35%, the rolling force adjustment range is 30%-50KN, and the speed adjustment range is 1.5-2.5m / s.

[0070] Step 103: Establish an adaptive model for process parameters, including the correlation rules between process parameters. The reduction rate is strongly correlated with the billet temperature; excessively high temperatures lead to a decrease in the reduction rate, while excessively low temperatures lead to an increase. The rolling force is related to the billet hardness; the greater the hardness, the greater the rolling force. The rolling speed is related to the billet size and shape; the larger the size, the higher the speed.

[0071] Step 2: Collect and preprocess process parameters.

[0072] Step 201: Deploy a multi-sensor array at the billet inlet, between each rolling mill stand, and at the outlet, including a machine vision unit, a laser scanner, and a multi-band infrared thermal imager. The machine vision unit uses an industrial high-definition camera with a resolution of 20 million pixels for initial billet positioning and surface defect detection; the laser scanner uses a blue-green laser with a scanning range of ±15° and an accuracy of ±0.02mm; the infrared thermal imager uses a 640×480 resolution focal plane array with a temperature measurement range of 0-2000℃.

[0073] Step 202: Initial positioning of the steel billet and online detection of surface defects are achieved through a machine vision unit. Image recognition algorithms are used to identify the centerline of the steel billet, with positioning accuracy controlled within ±0.1mm; machine learning algorithms are used to identify surface defects, achieving an accuracy rate of over 95%.

[0074] Step 203: Measure the three-dimensional contour dimensions of the rolled piece in real time using a laser scanner. The scanning frequency is 100Hz, and the data update rate is 50Hz.

[0075] Step 204: Use a multi-band infrared thermal imager to collect the temperature field distribution along the entire length and width of the rolled piece. The temperature field resolution is 16×16 pixels, and the frame rate is 50Hz.

[0076] Step 205: Preprocess and standardize the collected data. Signal processing methods are used to filter out noise, and normalization methods are used to map the data to the [0,1] interval.

[0077] Step 3: Perform process parameter optimization calculations. Step 301: Input the preprocessed data into the data-driven decision-making core.

[0078] Step 302: Use a deep neural network to extract features from the input data and predict process parameters. The network structure is a 5-layer fully connected network with 64 hidden layer nodes, using the ReLU activation function, the mathematical expression of which is:

[0079]

[0080] in, These are the input values ​​of the neurons. The output layer uses a linear activation function.

[0081] Step 303: Optimize process parameters using reinforcement learning algorithms to make the model adapt to the latest production conditions. The Adam optimization algorithm is used with a learning rate of 0.001 and a batch size of 32.

[0082] Step 304: Optimize the number of decision trees and the maximum depth of trees in the random forest model using the Northern Eagle optimization algorithm. The number of decision trees ranges from 10 to 100, and the tree depth ranges from 3 to 10.

[0083] Step 4: Perform quality status difference analysis to optimize process parameters.

[0084] Step 401: Evaluate the quality status based on real-time acquired process parameters and simulation results. Principal components are extracted using PCA, and the principal components are calculated through the following optimization problem:

[0085]

[0086] in, It is a standardized process parameter data matrix. It is a unit eigenvector. It is the first principal component. Evaluation indicators include rolling force, temperature field uniformity, and dimensional accuracy.

[0087] Step 402: Filter outliers in the data using the Isolation Forest algorithm. The neighborhood size in the algorithm is 20, and the number of samples is 100.

[0088] Step 403: Use a tree model for feature selection to filter out the features most relevant to rolling force prediction. The ReliefF algorithm is used, with the number of features limited to 10.

[0089] Step 404: Feedback the optimized process parameters back to the slab rolling process for control. A fuzzy PID controller is used, with a control accuracy of ±5%.

[0090] Step 5: Determine whether the production end conditions have been met. If not, return to step 2; otherwise, proceed to step 6.

[0091] Step 6: Output the optimized process parameters.

[0092] Step 601: Export the optimized process parameters in Excel spreadsheet format.

[0093] Step 602: Visualize the process parameters. Use bar charts to display parameters such as reduction rate and rolling force for each pass, and use heat maps to display the temperature field distribution.

[0094] Step 603: Generate a production suggestion report, including parameters such as reduction rate, rolling force, and speed for each pass.

[0095] Example 2:

[0096] A data-driven intelligent billet rolling method for hot-rolled H-beams includes the following steps:

[0097] Step 1: Initialize system parameters and establish a process parameter database.

[0098] Step 101: Establish a default value library for process parameters, including basic parameters such as reduction rate, rolling force, and speed for each pass. Specifically, the reduction rate for the first pass is set to 30%, the rolling force to 25 kN, and the speed to 2.0 m / s; the reduction rates for the second to fourth passes are 28%, 30%, and 32%, respectively, with rolling forces of 45 kN, 60 kN, and 55 kN, respectively, and the speed maintained at 2.2 m / s; the reduction rate for the fifth pass is 35%, the rolling force to 40 kN, and the speed to 2.5 m / s.

[0099] Step 102: Set the adjustment range of process parameters, including the maximum allowable deviation. The reduction rate adjustment range is 25%-35%, the rolling force adjustment range is 35%-55KN, and the speed adjustment range is 1.8-2.8m / s.

[0100] Step 103: Establish an adaptive model for process parameters, including the correlation rules between process parameters. The reduction rate is strongly correlated with the billet temperature; excessively high temperatures lead to a decrease in the reduction rate, while excessively low temperatures lead to an increase. The rolling force is related to the billet hardness; the greater the hardness, the greater the rolling force. The rolling speed is related to the billet size and shape; the larger the size, the higher the speed.

[0101] Step 2: Collect and preprocess process parameters.

[0102] Step 201: Deploy a multi-sensor array at the billet inlet, between each rolling mill stand, and at the outlet, including a machine vision unit, a laser scanner, and a multi-band infrared thermal imager. The machine vision unit uses an industrial high-definition camera with a resolution of 30 million pixels for initial billet positioning and surface defect detection; the laser scanner uses a green laser with a scanning range of ±20° and an accuracy of ±0.01mm; the infrared thermal imager uses a 1024×768 resolution focal plane array with a temperature measurement range of 0-2500℃.

[0103] Step 202: Initial positioning of the steel billet and online detection of surface defects are achieved through a machine vision unit. A deep learning algorithm is used to identify the centerline of the steel billet, with positioning accuracy controlled within ±0.05mm; a machine learning algorithm is used to identify surface defects, achieving an accuracy rate of over 98%.

[0104] Step 203: Measure the three-dimensional contour dimensions of the rolled piece in real time using a laser scanner. The scanning frequency is 150Hz, and the data update rate is 75Hz.

[0105] Step 204: Use a multi-band infrared thermal imager to collect the temperature field distribution along the entire length and width of the rolled piece. The temperature field resolution is 32×32 pixels, and the frame rate is 75Hz.

[0106] Step 205: Preprocess and standardize the collected data. Signal processing methods are used to filter out noise, and normalization methods are used to map the data to the [0,1] interval.

[0107] Step 3: Perform process parameter optimization calculations. Step 301: Input the preprocessed data into the data-driven decision-making core.

[0108] Step 302: Use a deep neural network to extract features from the input data and predict process parameters. The network structure is a 6-layer fully connected network with 128 hidden nodes, using the ReLU activation function, the mathematical expression of which is:

[0109]

[0110] in, These are the input values ​​of the neurons. The output layer uses a linear activation function.

[0111] Step 303: Optimize process parameters using reinforcement learning algorithms to adapt the model to the latest production conditions. The RMSprop optimization algorithm is used with a learning rate of 0.0001 and a batch size of 64.

[0112] Step 304: Optimize the number of decision trees and the maximum depth of trees in the random forest model using the Northern Eagle optimization algorithm. The number of decision trees ranges from 20 to 120, and the tree depth ranges from 4 to 12.

[0113] Step 4: Perform quality status difference analysis to optimize process parameters.

[0114] Step 401: Evaluate the quality status based on real-time acquired process parameters and simulation results. Principal components are extracted using PCA, and the principal components are calculated through the following optimization problem:

[0115]

[0116] in, It is a standardized process parameter data matrix. It is a unit eigenvector. It is the first principal component. Evaluation indicators include rolling force, temperature field uniformity, and dimensional accuracy.

[0117] Step 402: Filter outliers in the data using the Isolation Forest algorithm. The neighborhood size in the algorithm is 30, and the number of samples is 150.

[0118] Step 403: Use a tree model for feature selection to filter out the features most relevant to rolling force prediction. The ReliefF algorithm is used, with the number of features limited to 15.

[0119] Step 404: Feedback the optimized process parameters back to the slab rolling process for control. A fuzzy PID controller is used, with a control accuracy of ±3%.

[0120] Step 5: Determine whether the production end conditions have been met. If not, return to step 2; otherwise, proceed to step 6.

[0121] Step 6: Output the optimized process parameters. Step 601: Export the optimized process parameters as an Excel spreadsheet.

[0122] Step 602: Visualize the process parameters. Use bar charts to display parameters such as reduction rate and rolling force for each pass, and use heat maps to display the temperature field distribution.

[0123] Step 603: Generate a production suggestion report, including parameters such as reduction rate, rolling force, and speed for each pass.

[0124] Finally, it should be pointed out that the above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A data-driven intelligent billet rolling method for hot-rolled H-beams, characterized in that, The method includes the following steps: Step 1: Initialize the parameters of the automated control system for the rolling process and establish a process parameter database; Step 2: Collect process parameters and preprocess the process parameters; Step 3: Perform process parameter optimization calculations; The preprocessed operating data is input into the data-driven decision-making core, and a deep neural network is used to extract high-dimensional nonlinear features and output the initial predicted values ​​of process parameters. Through reinforcement learning algorithms, adaptive adjustments based on real-time feedback are achieved; at the same time, optimization algorithms are used to globally optimize hyperparameters, thus implementing process parameter control strategies. Step 4: Conduct a quality status assessment on the real-time collected process parameters and the initial predicted values ​​of the process parameters, perform quality status difference analysis, and optimize the process parameters. Step 5: Determine whether the production end conditions have been met. If not, return to step 2; otherwise, proceed to step 6. Step 6: Output the optimized process parameters.

2. The data-driven intelligent billet rolling method for hot-rolled HH-section steel according to claim 1, characterized in that, Step 1 includes: Step 101: Establish a default value library for process parameters, wherein the process parameters are basic parameters including the reduction rate, rolling force, and speed for each pass; Step 102: Set the adjustment range of process parameters, including the maximum allowable deviation value; Step 103: Establish an adaptive model for process parameters, including the correlation rules between process parameters.

3. The data-driven intelligent billet rolling method for hot-rolled HH-section steel according to claim 1, characterized in that, Step 2 includes: Step 201: Deploy a multi-sensor array at the inlet of the billet mill, between each stand of the rolling mill, and at the outlet; Step 202: Initially position the billet and detect surface defects online using a machine vision unit to obtain initial position data and surface defect data of the billet; Step 203: Use a laser scanner to measure the three-dimensional contour dimensions of the rolled piece in real time after each pass, and obtain the three-dimensional contour data of the rolled piece in each pass. Step 204: Use a multi-band infrared thermal imager to collect the temperature field distribution of the entire length and width of the rolled piece, and obtain the temperature data of the rolled piece; Step 205: Preprocess and standardize the collected data.

4. The data-driven intelligent billet rolling method for hot-rolled HH-section steel according to claim 1, characterized in that, Step 3 includes: Step 301: Input the preprocessed data into the data-driven decision-making core; Step 302: Use a deep neural network to extract features from the input data and predict process parameters; Step 303: Optimize process parameters using reinforcement learning algorithms to enable the model to adapt to new production conditions; Step 304: Use the Northern Eagle optimization algorithm to optimize the number of decision trees and the maximum depth of trees in the random forest model.

5. The data-driven intelligent billet rolling method for hot-rolled HH-section steel according to claim 1, characterized in that, Step 4 includes: Step 401: Evaluate the quality status based on the real-time acquired process parameters and simulation calculation results; Step 402: Filter outliers in the data using the Isolation Forest algorithm; Step 403: Use a tree model to select features and filter out the features most relevant to rolling force prediction; Step 404: Feed back the optimized process parameters to the slab rolling process for control.

6. The data-driven intelligent billet rolling method for hot-rolled H-beams according to claim 1, characterized in that, Step 6 includes: Step 601: Export the optimized process parameters in Excel spreadsheet format; Step 602: Visualize the process parameters; Step 603: Generate a production suggestion report.

7. A data-driven intelligent billet rolling system for hot-rolled H-beams as described in claim 1, characterized in that, The system includes: The execution layer receives optimized process parameter setting instructions and converts them into adaptive control signals for the rolling process through a fuzzy PID controller; it drives the adaptive actuator to adjust the execution parameters; and it monitors and dynamically corrects the position and attitude of the rolled piece in real time based on machine vision or laser detection feedback. The output module is used to visualize and generate reports on the key raw data of real-time control signals, actuator status, and process stability indicators generated by the output execution layer during the control process. The sensing layer is used to collect process status data of the rolling process in real time through a multi-sensor array; The data preprocessing module is used to convert the received sensing layer process status data into structured data by removing equipment noise and electromagnetic interference through signal filtering, eliminating sensor dimension differences and numerical range deviations through standardization and normalization. The decision-making level uses a deep neural network to extract high-dimensional nonlinear features from the structured data processed by the data preprocessing module and makes initial predictions of process parameters and quality indicators. By using reinforcement learning algorithms, process parameters are dynamically optimized and adaptively adjusted based on real-time production feedback. The optimization algorithm is used to globally optimize the hyperparameters of the key model, and the key variables are selected by combining the feature selection method, and the optimized process parameter control strategy is output.

Images