Electric permanent magnet uncoiling composite magnetic field multi-mode adjusting system

Abstract

The invention provides an electro-permanent magnet uncoiling composite magnetic field multi-mode adjustment system which comprises a data acquisition and preprocessing module, a graph structure construction module, a dynamic prediction module, a historical matching prediction module, a magnetic force adjustment strategy module and an electro-permanent magnet execution mechanism. The data acquisition module constructs a multi-modal data cube through a temperature field, deformation, stress and magnesium powder concentration sensing network; the graph structure module discretizes the steel strip coil into a node network and calculates a mechanical association edge weight; the dynamic prediction module updates node features through the graph neural network and then outputs a sliding probability vector through the LSTM network; the historical matching module screens similar working conditions based on a dynamic time warping algorithm and generates magnetic force preset parameters; the strategy model fuses the probability vector and a preset parameter to generate a partition magnetic matrix; and the execution mechanism realizes millisecond-level magnetic field regulation and control through the reconfigurable magnetic pole array. According to the system, the magnetic field control precision can be improved, the interlayer sliding risk is reduced, and the self-adaptive adjustment effect is achieved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an electric permanent magnet unwinding composite magnetic field multi-mode adjustment system. Background Art

[0002] When the existing steel strip coil is extracted from the blast furnace and placed, a large amount of magnesium powder is added during the high-temperature firing process. During the cooling process, the magnesium powder will accumulate unevenly at one end of the steel strip coil, resulting in uneven distribution of interlayer bonding force. This unevenness will cause two major problems: first, the end face of the steel strip coil will form a significant concave deformation, affecting the subsequent processing accuracy; second, the risk of interlayer sliding is significantly increased, which can easily cause technical problems such as interlayer sliding, center movement, and end face damage during the unwinding process. The traditional unwinding system adopts a fixed magnetic field control mode and cannot perceive the temperature field distribution, geometric deformation, residual stress field, magnesium powder concentration and other multi-modal physical field changes of the steel strip coil in real time. It is even more difficult to adaptively adjust the magnetic field according to these dynamic changes. In addition, the existing technology lacks effective use of historical operating condition data and cannot achieve predictive adjustment based on similar operating conditions, resulting in obvious lag in magnetic field control.

[0003] In view of the above problems, the existing technology is in urgent need of improvement. Summary of the Invention

[0004] In order to solve the problems existing in the prior art, the present invention provides an electro-permanent magnetic unwinding composite magnetic field multi-modal adjustment system to improve the magnetic field control accuracy, reduce the risk of interlayer sliding, and achieve adaptive adjustment.

[0005] To achieve the above objectives, in a first aspect of the present invention, a multi-mode adjustment system for an electro-permanent magnetic unwinding composite magnetic field is provided, the system comprising: A data acquisition and preprocessing module is used to collect and preprocess the multimodal data of the steel strip coil. The multimodal data includes temperature field distribution data, geometric deformation data, residual stress field data, and magnesium powder concentration distribution data. The data acquisition and preprocessing module obtains the physical field information of the steel strip coil surface in real time through a distributed sensor array, and outputs a four-dimensional data cube after median filtering, Kriging interpolation, and normalization processing; A graph structure construction module, connected to the data acquisition and preprocessing module, is used to convert the preprocessed multimodal data into a spatiotemporal correlation graph. The nodes in the spatiotemporal correlation graph structure represent segmented regions of the steel strip coil, and each node has a feature vector. The feature vector includes the temperature, deformation, stress, and magnesium powder concentration of the corresponding region. The edges in the spatiotemporal correlation graph structure represent the physical interaction relationship between regions, and the weight of the edge is calculated based on the mechanical force and thermal stress between the regions; A dynamic prediction module, connected to the graph structure construction module, is used to analyze the spatiotemporal evolution characteristics of the graph structure and includes a graph neural network processing layer and a long short-term memory network prediction layer. The graph neural network processing layer is used to process the spatiotemporal association graph structure and update node features. The long short-term memory network prediction layer is used to predict the interlayer slip probability of each area of the steel strip coil based on the updated node features and output a slip probability vector. A historical matching prediction module, which is connected to the data acquisition and preprocessing module and is used to time-regularly match the multimodal data sequence in the current time window with the historical database, select the historical data set with the highest matching degree, and predict the magnetic parameters at the next moment based on the historical data set. It includes a time series database, a real-time matching unit, and a prediction output unit; The time series database stores Set historical working condition records, each record contains timestamp, multimodal data cube, magnetic parameter matrix and end face concavity measured value; The real-time matching unit is used to intercept the current time window data sequence, perform similarity matching on the current time window data sequence with the historical data, and calculate the similarity with the historical data using a dynamic time warping algorithm; The prediction output unit is used to generate a magnetic parameter prediction value at the next moment and output a weighted magnetic parameter based on the magnetic parameters recorded in the matched historical data and combined with environmental parameter compensation; A magnetic force adjustment strategy module, which is connected to the dynamic prediction module and the historical matching prediction module, is used to fuse the interlayer slip probability vector and the predicted magnetic parameters, generate a partitioned magnetic force parameter matrix based on a preset magnetic flux calculation rule, and realize adaptive allocation of the number of magnetic poles and magnetic flux density; The electro-permanent magnetic actuator is connected to the magnetic force adjustment strategy module, including a reconfigurable magnetic pole array for responding to magnetic parameter matrix instructions, and realizing dynamic adjustment of magnetic flux and N / S pole reorganization through excitation coil current regulation and polarity switching circuit.

[0006] Furthermore, the historical matching prediction module further includes: Preload execution unit, when matching confidence When the historical matching result is output, the basic magnetic force parameters are sent to the electro-permanent magnetic actuator within 10ms; Exception handling unit, when the highest similarity score in the history matching process When the alarm is triggered, the system switches to pure model prediction mode; Database self-update unit, adding effective working condition records every day Group, record screening criteria is the measured value of end face concavity mm and interlayer sliding rate .

[0007] Furthermore, the real-time matching unit of the historical matching prediction module executes a dynamic time warping algorithm, which specifically includes performing dimensionality reduction processing on the current data window and historical data respectively, calculating the optimal time alignment path, and calculating the similarity score based on the cumulative distance of the path.

[0008] Furthermore, the prediction output unit executes the magnetic pole number allocation rule as follows: And the magnetic flux enhancement in the edge region is .

[0009] Furthermore, the electro-permanent magnetic actuator comprises independent magnetic pole units, each unit including a neodymium iron boron permanent magnet, an excitation coil, a polarity switching circuit and a magnetic flux closed-loop control system.

[0010] In a second aspect of the present invention, a method for adjusting a multi-mode of an electro-permanent magnetic unwinding composite magnetic field is provided, which is applied to the above-mentioned system, comprising: Step S100: collecting the temperature field distribution data, geometric deformation data, residual stress field data and magnesium powder concentration distribution data of the steel strip coil, and outputting the pre-processed data cube; Step S200: constructing a spatiotemporal association graph structure and calculating node feature vectors and edge weights; Step S300: Outputting an inter-layer sliding probability vector through a dynamic prediction module; Step S400: Execute historical matching prediction, perform dimensionality reduction compression, multi-level matching, dynamic weighted calculation, and preload execution in sequence; The dimension reduction compression calculation is: , retaining 95% variance; The multi-level matching is divided into coarse screening and fine screening, the coarse screening is to screen the Euclidean distance The fine screening performs DTW similarity calculation on the coarse screening result; The dynamic weighting combines similarity and temperature difference to calculate the weight, and the calculation formula is: , weight ; The specific operation of the preloading execution is to 50ms before the strategy model completes the calculation Medium magnetic flux Pre-energize the area; Step S500: fusing the sliding probability vector and the predicted magnetic parameters to generate a magnetic adjustment strategy; Step S600: driving the electro-permanent magnetic actuator to perform dynamic magnetic field adjustment.

[0011] Furthermore, the step S400 also includes similarity confidence verification and data aging weighting. The similarity confidence verification ,when When matching records are discarded, the data aging weighting method is to calculate the historical record weight .

[0012] Furthermore, the time series database update rule in step S400 is to add new records every day. Group, implement elimination rules , recalculate the principal component analysis matrix every 72 hours Optimize feature space.

[0013] Furthermore, the magnetic flux optimization in step S500 satisfies , where the rate of change of the magnetic flux density gradient in adjacent regions is .

[0014] Furthermore, the step S600 drives the electro-permanent magnetic actuator to perform dynamic magnetic field adjustment, including performing magnetic pole array response, so that the polarity switching time , through the Hall sensor to correct the magnetic flux deviation in real time to complete the closed-loop feedback control and control the total response delay .

[0015] The present invention provides an electric permanent magnet unwinding composite magnetic field multimodal adjustment system, which acquires multimodal data in real time through a data acquisition and preprocessing module, combines a graph structure construction module and a dynamic prediction module to perform spatiotemporal evolution analysis, and uses a historical matching prediction module to achieve predictive adjustment. It cooperates with a magnetic force adjustment strategy module to generate adaptive magnetic force parameters, effectively solving the problems of traditional systems being unable to perceive dynamic changes in real time and lacking historical data utilization. It has the advantages of improving magnetic field control accuracy, reducing the risk of interlayer sliding, avoiding end face damage, and achieving adaptive adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a structural diagram of an electric permanent magnet unwinding composite magnetic field multi-modal adjustment system provided by an embodiment of the present invention; Figure 2 A flow chart of a method for adjusting a multi-mode electromagnetic field of an electro-permanent magnetic unwinding composite magnetic field provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0018] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Therefore, features specified as "first" or "second" may explicitly or implicitly include one or more features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0019] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration or illustration". Any embodiment of the present invention described as "for example" is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any person skilled in the art to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed herein.

[0020] Reference Figure 1 , Figure 1 This is a structural diagram of an electric permanent magnet unwinding composite magnetic field multi-mode adjustment system provided by the present invention, an electric permanent magnet unwinding composite magnetic field multi-mode adjustment system comprising: The data acquisition and preprocessing module 10 is used to collect and preprocess the multimodal data of the steel strip coil. The multimodal data includes temperature field distribution data, geometric deformation data, residual stress field data, and magnesium powder concentration distribution data. The data acquisition and preprocessing module obtains the physical field information of the steel strip coil surface in real time through a distributed sensor array, and outputs a four-dimensional data cube after median filtering, Kriging interpolation and normalization processing.

[0021] Specifically, multimodal data refers to the temperature field distribution data, geometric deformation data, residual stress field data, and magnesium powder concentration distribution data generated during the production process of steel strip coils. Specifically, a distributed sensor array can be used to collect surface physical field information in real time. By using median filtering to eliminate pulse noise, Kriging interpolation to fill missing data, and normalization processing to eliminate dimensional differences, a four-dimensional data cube is formed to provide complete physical field characteristics for subsequent analysis.

[0022] In the specific implementation, the data acquisition and preprocessing module uses a distributed sensor array, including an infrared thermal imager, a laser displacement sensor, a stress strain gauge, and a capacitive magnesium powder concentration detector. The sensor array is evenly arranged along the axial and radial directions of the steel strip roll to collect physical field information in real time. The collected raw data is subjected to median filtering to remove noise, and then the sparse data points are spatially interpolated using the Kriging interpolation algorithm. Finally, normalization is performed to obtain a standardized four-dimensional data cube.

[0023] Through the above technical solutions, comprehensive collection of multimodal data is achieved, providing a rich data foundation for the subsequent construction of spatiotemporal correlation maps and prediction models.

[0024] The graph structure construction module 20 is connected to the data acquisition and preprocessing module and is used to convert the preprocessed multimodal data into a spatiotemporal correlation graph. The nodes in the spatiotemporal correlation graph structure represent the segmented areas of the steel strip coil. Each node has a characteristic vector, which includes the temperature, deformation, stress and magnesium powder concentration of the corresponding area. The edges in the spatiotemporal correlation graph structure represent the physical interaction relationship between the areas, and the weights of the edges are calculated based on the mechanical force and thermal stress between the areas.

[0025] Specifically, the spatiotemporal correlation graph refers to a topological structure that divides the steel strip coil into multiple segmented regions and establishes physical field correlation relationships. Specifically, it can be constructed by using nodes to represent regional units and edges to represent interactions between regions. The node feature vector contains temperature, deformation, stress and magnesium powder concentration parameters, and the edge weight is calculated based on the mechanical force and thermal stress to form a computable physical field dynamic propagation model.

[0026] In the specific implementation, the graph structure construction module divides the preprocessed data cube into several regions. Each region corresponds to a node in the spatiotemporal association graph. The node feature vector contains the average temperature, maximum deformation, principal stress value and magnesium powder concentration of the region. The edge weight is determined by calculating the mechanical force and thermal stress gradient between adjacent regions.

[0027] The dynamic prediction module 30 is a connection graph structure construction module, which is used to analyze the spatiotemporal evolution characteristics of the graph structure, including a graph neural network processing layer and a long short-term memory network prediction layer. The graph neural network processing layer is used to process the spatiotemporal correlation graph structure and update the node features. The long short-term memory network prediction layer is used to predict the inter-layer sliding probability of each area of the steel strip coil based on the updated node features and output the sliding probability vector.

[0028] Specifically, the dynamic prediction module refers to a hybrid prediction architecture that integrates graph neural networks and long-short-term memory networks. Specifically, the graph neural network processing layer can be used to update node feature vectors, and the long-short-term memory network prediction layer can be used to capture the time series evolution law, output the sliding probability vectors between layers in each region, and establish a mapping relationship between physical field changes and sliding risks.

[0029] In its implementation, the graph neural network processing layer in the dynamic prediction module uses a graph convolutional network architecture, extracting spatial correlation information from node features through multi-layer convolution operations. The long-short-term memory network prediction layer uses multiple LSTM units to construct a time series prediction model, taking as input the updated node feature sequence and outputting the inter-layer sliding probability for each region.

[0030] The historical matching prediction module 40 is connected to the data acquisition and preprocessing module and is used to time-regularly match the multimodal data sequence in the current time window with the historical database, select the historical data set with the highest matching degree, and predict the magnetic parameters at the next moment based on the historical data set. It includes a time series database, a real-time matching unit, and a prediction output unit. The time series database stores ≥10,000 sets of historical working condition records, each record containing a timestamp, a multimodal data cube, a magnetic parameter matrix, and the measured value of the end face concavity; The real-time matching unit is used to capture the current time window , the current time window data sequence is matched with the historical data for similarity, and the similarity with the historical data is calculated through the dynamic time warping algorithm .

[0031] Furthermore, the dynamic time warping algorithm also includes dimensionality reduction processing for the current data window and historical data respectively, calculating the optimal time alignment path and calculating the similarity score based on the cumulative distance of the path.

[0032] Dimensionality reduction can use principal component analysis or wavelet transform methods to compress the four-dimensional data cube into a two-dimensional feature space. For example, the temperature field and residual stress field are merged into thermodynamic feature vectors, and the geometric deformation and magnesium powder concentration are merged into the structural stability feature vector.

[0033] The optimal time-aligned path is implemented using a dynamic programming algorithm, where the path constraints are set such that the time offset of adjacent points does not exceed 3 sampling intervals, and the path search range is limited to a dynamic time warping window width of 20% of the current window length.

[0034] The prediction output unit is used to generate the predicted value output of the magnetic parameters at the next moment based on the magnetic parameters recorded in the matched historical data and combined with the environmental parameter compensation. Output weighted magnetic parameters: ,in , .

[0035] Furthermore, the prediction output unit executes the magnetic pole number allocation rule as follows: And the magnetic flux enhancement in the edge region is .

[0036] The similarity threshold is set to 0.95, and the historical magnetic flux threshold is set to 1.5T. These two conditions are combined through logical AND relationships to form screening conditions. When the conditions are met, the number of magnetic poles is allocated to 16, which matches the magnetic field fineness requirements under high-similarity working conditions. For cases where the conditions are not met, the number of magnetic poles is set to 8, and the magnetic flux in the edge area is enhanced to 1.2 times the average value of the central area. The enhancement ratio is determined based on the sudden increase in magnetic flux demand caused by stress concentration in the edge area. The differentiated allocation strategy of the number of magnetic poles and magnetic flux intensity is achieved through a dynamic adjustment mechanism. The increase or decrease in the number of magnetic poles directly affects the coverage density of the magnetic field distribution, and the edge enhancement operation is completed by adjusting the excitation coil current parameters.

[0037] The historical matching prediction module also includes a preloading execution unit, an exception handling unit and a database self-update unit. The preloading execution unit is configured so that when the matching confidence exceeds 0.9, the basic magnetic parameters are sent to the electro-permanent magnetic actuator within 10 milliseconds. The preloading execution unit triggers the actuator action in advance during the data matching stage by coupling the confidence threshold with the millisecond-level response mechanism.

[0038] The exception handling unit is configured so that when the highest similarity score in the historical matching process is lower than 0.6, the third-level alarm is triggered and the prediction mode is switched to the pure model prediction mode. The exception handling unit maintains the continuity of system operation when data matching fails by coordinating the similarity threshold with the mode switching mechanism.

[0039] The database self-update unit is configured to add no less than 50 sets of valid working condition records every day. The new records must meet the screening criteria that the measured value of the end face concavity is less than 2.5 mm and the interlayer sliding rate is less than 5%. The database self-update unit ensures the timeliness of the historical database by combining a dynamic update mechanism with quality screening conditions.

[0040] Specifically, historical matching prediction refers to a working condition matching method based on a dynamic time warping algorithm. Specifically, a time series database can be used to store historical working condition records, and the current data window and historical data set can be matched through dimensionality reduction compression and similarity calculation. Combined with environmental parameter compensation, weighted magnetic parameter prediction values are generated to achieve parameter optimization based on historical experience.

[0041] Furthermore, the current data window is processed by dimensionality reduction to generate a feature vector sequence, which is matched with the records in the historical database that have undergone the same dimensionality reduction process. The dynamic time warping algorithm traverses all possible alignment paths under constraints and determines the optimal path by comparing local feature differences. The cumulative distance value on the optimal path is normalized and converted into a similarity score. The lower the score value, the higher the matching degree.

[0042] For example, when the time series length of the historical data record is 120 seconds and the current window is 100 seconds, the algorithm automatically adjusts the time axis scaling ratio and searches for the optimal phase alignment method within the allowable time offset range. In this process, the data dimension is compressed to less than 30% of the original dimension, and the computing resource consumption is reduced by about 65%. At the same time, through precise time axis alignment, the similarity evaluation error rate is reduced from 12% of the traditional method to 4.5%. This technical solution forms a synergistic relationship with the previous spatiotemporal correlation graph structure. The feature vector after dimensionality reduction can be directly mapped to the feature space of the graph node, so that the historical matching results and the output of the dynamic prediction module have a consistent physical meaning representation basis.

[0043] During the similarity calculation process, the similarity score output by the dynamic time warping algorithm and the magnetic flux data in the historical database are synchronously retrieved. When the similarity exceeds 0.95 and the associated historical magnetic flux record is greater than 1.5T, the system determines that the current operating conditions require a high-density magnetic field distribution. At this time, the configuration of 16 magnetic poles reduces the magnetic pole spacing to 50% of the original density, thereby improving the magnetic field gradient control accuracy. For cases where the threshold is not reached, the configuration of 8 magnetic poles reduces the system energy consumption by reducing the number of magnetic poles. At the same time, the magnetic flux enhancement in the edge area is achieved by independently controlling the excitation coil current in the area, for example, increasing the current intensity to 120% of the baseline value.

[0044] This allocation rule achieves dynamic adaptation of the number of magnetic poles and magnetic flux intensity through dual threshold conditions. The magnetic pole increment strategy under high similarity conditions can suppress the probability of interlayer sliding by 18%, while the edge enhancement operation increases the local magnetic field constraint force to 1.3 times that of the unenhanced state, effectively responding to sudden changes in magnetic flux demand caused by residual stress. Through the coordinated adjustment of the number of magnetic poles and edge magnetic flux, the system reduces the risk of interlayer sliding in the edge area to below 5% while ensuring the stability of the magnetic field in the core area.

[0045] When the matching confidence output by the real-time matching unit exceeds 0.9, the preload execution unit immediately calls the optimal magnetic parameters in the historical database. Before the dynamic prediction module completes the calculation, the basic magnetic parameters have been transmitted to the electro-permanent magnetic actuator, thereby eliminating the impact of system response delay on the timeliness of magnetic adjustment.

[0046] During the historical data matching process, if the highest similarity score of all historical records is lower than 0.6, the exception handling unit automatically terminates the current matching process and activates the pure model prediction mode to ensure the continuous and stable operation of the system under abnormal working conditions.

[0047] The database self-update unit automatically performs data cleaning operations every day. New valid working condition records are stored in the time series database. At the same time, historical records with a storage time of more than 72 hours and a usage frequency below the threshold are eliminated, thereby maintaining the long-term accuracy of the prediction model.

[0048] For example, when the measured value of the end face concavity of a batch of steel strip coils reaches 3.0 mm, the working condition record is automatically excluded from the valid data. The three functional units form a closed-loop control through timing control and data interaction. The preloading execution unit shortens the response time, the exception handling unit provides a fault-tolerant mechanism, and the database self-update unit optimizes data quality. The three work together to solve the problems of real-time control delay, abnormal working condition response and data aging.

[0049] In the specific implementation, the time series database of the historical matching prediction module adopts a distributed storage architecture. The real-time matching unit uses a sliding time window to intercept the current data sequence, and calculates the similarity with the historical data through a dynamic time warping algorithm. The prediction output unit obtains preliminary magnetic parameters based on the weighted average of the similarity, and then compensates for factors such as ambient temperature.

[0050] Furthermore, the preloading execution unit sets the matching confidence threshold to 0.9. When the matching confidence exceeds the threshold, the basic magnetic parameters are sent to the electro-permanent magnetic actuator within 10ms after the historical matching result is output. The exception handling unit sets the similarity score threshold to 0.6. When the highest similarity score in the historical matching process is lower than the threshold, the third-level alarm is triggered and the mode is switched to pure model prediction. The database self-update unit adds no less than 50 groups of valid working condition records every day. The record screening criteria are that the measured value of the end face concavity is less than 2.5 mm and the interlayer sliding rate is less than 5%.

[0051] It solves the problem of real-time response delay in the historical data matching process, improves the timeliness of the actuator's action, establishes an emergency response mechanism under abnormal working conditions, reduces the risk of system loss of control, realizes the dynamic update of the historical database, and ensures the long-term accuracy of the prediction model.

[0052] When the real-time matching unit of the historical matching prediction module executes the dynamic time warping algorithm, it first performs dimensionality reduction on the current data window and historical data. Specifically, the principal component analysis method is used to compress the features of the multidimensional time series data, retaining the principal components with an explained variance ratio of 95%. Secondly, the optimal time alignment path is calculated, and the dynamic programming algorithm is further used to construct the cumulative distance matrix. The optimal alignment path is obtained by backtracking, and finally the similarity score is calculated based on the cumulative distance of the path. The local distance on the optimal path is accumulated as the global similarity measure, and normalized to obtain the final similarity score.

[0053] Efficient matching and precise alignment of high-dimensional multimodal data are achieved. Dimensionality reduction processing reduces computing resource consumption while retaining key feature information. The calculation of the optimal time alignment path overcomes the time misalignment problem of the traditional fixed window alignment method and accurately captures the temporal correlation between the data window and the historical records. The similarity calculation method based on the cumulative distance of the path can better reflect the overall matching degree of the multidimensional time series than a single distance indicator, thereby improving the reliability of historical working condition matching.

[0054] The prediction output unit executes the magnetic pole number allocation rule. When there are records with a similarity greater than 0.95 and a historical magnetic flux greater than 1.5T, 16 magnetic poles are allocated. Specifically, the prediction output unit first checks whether there are records with a similarity greater than 0.95 in the historical matching results. If so, it further checks whether the historical magnetic flux corresponding to the record is greater than 1.5T. When these two conditions are met at the same time, the prediction output unit will send an instruction to the electro-permanent magnetic actuator to configure 16 magnetic poles. This configuration is suitable for working conditions that require high matching and high magnetic field strength. In other cases, the prediction output unit will allocate 8 magnetic poles and perform special treatment on the edge area. The magnetic flux in the edge area will be enhanced to 1.2 times the average value of the central area.

[0055] Dynamic adaptation of the number of magnetic poles and magnetic flux intensity is achieved. By combining the dual thresholds of similarity and historical magnetic flux, the system can accurately identify working conditions that require high magnetic field intensity and improve the fineness of magnetic field distribution by increasing the number of magnetic poles. This method can more accurately suppress interlayer sliding, especially under highly matched working conditions. For other cases, the strategy of adopting an 8-pole configuration and strengthening the magnetic flux in the edge area can effectively cope with the sudden increase in magnetic flux demand caused by residual stress or deformation in the edge area. This differentiated allocation strategy enhances the magnetic field constraint force in the edge area, thereby avoiding interlayer sliding caused by local stress concentration, taking into account both system efficiency and adjustment accuracy, improving the accuracy of magnetic flux distribution and working condition adaptation, and effectively solving the problem of insufficient adaptation of magnetic flux distribution and working conditions in historical matching prediction.

[0056] The preloading execution unit preloads basic magnetic parameters with high matching degree through the high-speed cache mechanism, and triggers execution immediately when the matching confidence reaches the preset threshold. The exception handling unit monitors the matching process in real time, and automatically switches the prediction mode and issues a multi-level alarm when an anomaly is detected. The database self-update unit regularly screens and adds high-quality new working condition records, while eliminating low-value historical data, solving the real-time response delay problem in the historical data matching process, improving the timeliness of the actuator's action, establishing an emergency handling mechanism under abnormal conditions, reducing the risk of system loss of control, realizing the dynamic update of the historical database, and ensuring the long-term accuracy of the prediction model.

[0057] The magnetic adjustment strategy module 50 is connected to the dynamic prediction module and the historical matching prediction module, and is used to fuse the interlayer sliding probability vector and the predicted magnetic parameters, generate a partitioned magnetic parameter matrix based on the preset magnetic flux calculation rules, and realize the adaptive allocation of the number of magnetic poles and magnetic flux density.

[0058] Specifically, the magnetic adjustment strategy module connects the dynamic prediction module and the historical matching prediction module to fuse the interlayer sliding probability vector and the predicted magnetic parameters, and generates a partitioned magnetic parameter matrix based on the preset magnetic flux calculation rules to achieve adaptive allocation of the number of magnetic poles and magnetic flux density.

[0059] In the specific implementation, the magnetic adjustment strategy module integrates the interlayer slip probability vector and the predicted magnetic parameters, and uses a fuzzy logic controller to generate a partitioned magnetic parameter matrix. The control rules are based on a preset magnetic flux calculation formula to achieve adaptive allocation of the number of magnetic poles and magnetic flux density.

[0060] The electro-permanent magnetic actuator 60 is connected to the magnetic force adjustment strategy module, and includes a reconfigurable magnetic pole array for responding to magnetic parameter matrix instructions, and realizing dynamic adjustment of magnetic flux and N / S pole reorganization through excitation coil current regulation and polarity switching circuit.

[0061] The electro-permanent magnetic actuator consists of independent magnetic pole units, each of which includes a neodymium iron boron permanent magnet, an excitation coil, a polarity switching circuit, and a magnetic flux closed-loop control system.

[0062] Furthermore, the NdFeB permanent magnet is configured to provide a basic magnetic field, and its remanence characteristics are set to 1.2T, as a static magnetic field source support; the excitation coil is wound on both sides of the permanent magnet, and a positive or reverse superposition magnetic field is generated through bidirectional current input, and the current regulation accuracy is controlled within The polarity switching circuit is connected to the excitation coil loop, and the H-bridge topology is used to achieve current direction switching. The switching time is constrained to The magnetic flux closed-loop control system is integrated in each magnetic pole unit, including a Hall sensor array and a proportional integral differential controller. The sampling frequency is set to 1kHz, and the magnetic flux deviation is controlled within .

[0063] Specifically, the reconfigurable magnetic pole array refers to an actuator with polarity switching and magnetic flux closed-loop control functions. It can adopt a combined structure of neodymium iron boron permanent magnets and excitation coils. By adjusting the coil current to change the magnetic flux intensity and switching the circuit to change the magnetic pole polarity, the magnetic field distribution is reorganized within 10ms to form a composite magnetic field that matches the physical field of the steel strip coil.

[0064] Furthermore, when receiving the magnetic parameter matrix instruction, the magnetic flux closed-loop control system first obtains the current magnetic flux measured value through the Hall sensor, and calculates the difference with the instruction target value. The proportional integral differential controller generates a pulse width modulation signal according to the difference to drive the excitation coil to generate a compensation current. Under the working condition where polarity reversal is required, the polarity switching circuit changes the current direction by switching the conduction state of the H bridge, so that the basic magnetic field of the permanent magnet and the excitation magnetic field are vector-superimposed or offset. The magnetic flux output of each independent magnetic pole unit is individually regulated, and the magnetic field coupling degree between adjacent units is limited to .

[0065] For example, when the concavity of the end surface of the steel strip coil reaches 2.0mm, the magnetic pole units in the edge area are configured to output 1.8T magnetic flux, while the center area maintains 1.2T, and the magnetic flux density gradient change rate is controlled at 12% / cm. Through this split control architecture, the magnetic pole reorganization response time is shortened from 200ms in the traditional solution to less than 150ms, and the interlayer slip rate is reduced from 7% to 3.5%.

[0066] In the specific implementation, the electropermanent magnetic actuator consists of a reconfigurable magnetic pole array composed of multiple independently controlled electropermanent magnetic units. Each unit contains a permanent magnet, an excitation coil and a polarity switching circuit. By adjusting the excitation current and switching the polarity, dynamic adjustment of the magnetic flux and N / S pole reorganization are achieved, thereby forming a composite magnetic field distribution that matches the physical field of the steel strip coil.

[0067] Furthermore, the electro-permanent magnetic actuator includes independent magnetic pole units, each of which is composed of a neodymium iron boron permanent magnet, an excitation coil, a polarity switching circuit and a magnetic flux closed-loop control system. The neodymium iron boron permanent magnet serves as the basic magnetic source to provide a stable magnetic field. The excitation coil is wound around the outside of the permanent magnet, and the superposition or offset of the magnetic flux is achieved by adjusting the coil current. The polarity switching circuit is composed of an H-bridge circuit, which can quickly change the current direction of the excitation coil, thereby achieving reversal of the magnetic pole direction. The magnetic flux closed-loop control system includes a Hall sensor, a signal conditioning circuit and a PID controller, which monitors the magnetic flux output of the magnetic pole unit in real time and performs closed-loop adjustment according to the set value.

[0068] A reconfigurable magnetic pole array is composed of multiple independent magnetic pole units. Each magnetic pole unit can be independently controlled to achieve precise adjustment of the local magnetic field. By collaboratively controlling multiple magnetic pole units, a complex magnetic field distribution can be formed to meet the magnetic force requirements of different areas. The layout of the magnetic pole units can be flexibly adjusted according to the size and shape of the steel strip coil to adapt to steel strip coils of different specifications.

[0069] In practical applications, the magnetic adjustment strategy module generates a partitioned magnetic parameter matrix based on the interlayer slip probability vector and predicted magnetic parameters. This matrix contains the target magnetic flux and polarity information for each magnetic pole unit. The electropermanent magnetic actuator receives these instructions, and each independent magnetic pole unit then adjusts the excitation coil current and polarity to achieve dynamic adjustment of the magnetic flux and reorganization of the N / S poles. The entire process is coordinated by a central controller to ensure the coordinated work of each magnetic pole unit to form the desired composite magnetic field distribution.

[0070] High-precision, fast-response dynamic adjustment of magnetic flux and magnetic pole reorganization are achieved. The design of independent magnetic pole units overcomes the limitations of traditional integrated magnetic pole structures, significantly improving the magnetic flux adjustment accuracy and polarity switching speed. The split design enables each magnetic pole unit to independently respond to control instructions, and at the same time achieve adaptive adjustment of the overall magnetic field distribution through collaborative work. This flexible magnetic field control capability effectively suppresses the sliding phenomenon between steel strip coil layers and meets the demand for rapid dynamic adjustment of the bonding force between steel strip coil layers. In addition, the magnetic flux closed-loop control system ensures the stability and accuracy of the magnetic field output, further improving the reliability and control accuracy of the system.

[0071] In one or more embodiments, Figure 2 As shown, the present invention discloses a multi-modal adjustment method for an electro-permanent magnetic unwinding composite magnetic field, comprising: Step S100: collecting the temperature field distribution data, geometric deformation data, residual stress field data and magnesium powder concentration distribution data of the steel strip coil, and outputting the pre-processed data cube;

[0072] Step S200: constructing a spatiotemporal association graph structure and calculating node feature vectors and edge weights;

[0073] Step S300: Outputting an inter-layer sliding probability vector through a dynamic prediction module;

[0074] Step S400: Execute historical matching prediction, perform dimensionality reduction compression, multi-level matching, dynamic weighted calculation, and preload execution in sequence; Dimensionality reduction compression is calculated as , retaining 95% variance; Multi-level matching is divided into coarse screening and fine screening. Coarse screening is to screen the Euclidean distance The fine screening performs DTW similarity calculation on the coarse screening results. Dynamic weighting combines similarity and temperature difference to calculate weights. The calculation formula is: , weight ; The specific operation of preloading is to execute the Medium magnetic flux Pre-energize the area.

[0075] Furthermore, step S400 also includes similarity confidence verification and data aging weighting. Similarity confidence verification , when Confidence<0.8, matching records are discarded, and the data aging weighting method is to calculate the historical record weight .

[0076] Similarity confidence verification filters matching records by setting a confidence threshold. The threshold range can be set between 0.7 and 0.9. For example, 0.8 is determined to be the effective dividing point through experimental verification. Data aging weighting dynamically adjusts the weight coefficient of historical records through the time decay function. The decay function can adopt an exponential decay model, and the time window can be set to the last 30 days to 90 days. The synergistic effect of the two is reflected in the fact that after the confidence verification filters the low-quality matching results, the data aging weighting evaluates the timeliness of the remaining valid records. The weight of records whose timestamps and the current time interval exceed the set threshold is proportionally decayed. For example, the weight of records with an interval of more than 72 hours is reduced to 50% of the original value. This collaborative processing can simultaneously eliminate noise interference and data timeliness deviation, significantly reduce the prediction deviation caused by historical data mismatch, and improve the system's adaptability to changes in working conditions, thereby achieving more accurate magnetic parameter control.

[0077] The time series database update rule in step S40 is to add new records every day Group, implement elimination rules , recalculate the principal component analysis matrix every 72 hours Optimize feature space.

[0078] New records added daily The group's rules ensure that the current working condition change trend is effectively captured by setting the minimum data supplement threshold. The newly added data covers multi-dimensional parameters such as temperature field, deformation, stress and magnesium powder concentration. The data screening standard is the measured value of end face concavity. And the interlayer sliding rate , thereby ensuring the validity of the stored data. The elimination rule calculation adopts a dual evaluation mechanism based on time decay and feature contribution. The continuous addition of new records ensures the database's coverage of the current working conditions. The elimination rule suppresses storage expansion by screening low-value data. The periodic optimization of the principal component analysis matrix enables the feature space to dynamically adapt to changes in working conditions and eliminates the decline in matching accuracy caused by environmental parameter drift. The dynamic balance between database scale and quality improves the computational efficiency of similarity matching. At the same time, the implicit correlation extraction of the feature space enhances the matching accuracy between historical data and current working conditions.

[0079] Step S500: Fusing the sliding probability vector and the predicted magnetic parameters to generate a magnetic adjustment strategy.

[0080] Furthermore, the magnetic flux optimization in step S500 satisfies , where the rate of change of the magnetic flux density gradient in adjacent regions is , which effectively suppresses the local stress mutation phenomenon during the regulation of the bonding force between the steel strip layers, making the magnetic flux distribution curve continuous and smooth.

[0081] Step S600: driving the electro-permanent magnetic actuator to perform dynamic magnetic field adjustment.

[0082] Furthermore, step S600 drives the electro-permanent magnetic actuator to perform dynamic magnetic field adjustment, including performing magnetic pole array response, so that the polarity switching time , through the Hall sensor to correct the magnetic flux deviation in real time to complete the closed-loop feedback control and control the total response delay .

[0083] The inertia delay during the magnetic pole switching process is eliminated, allowing the magnetic field polarity adjustment to instantly respond to the dynamic changes in the interlayer bonding force of the steel strip coil. The magnetic flux closed-loop control mechanism effectively suppresses the magnetic field output deviation caused by the interference of the permanent magnet remanence and the ambient temperature, avoiding the deterioration of the end face concavity caused by the accumulation error of the magnetic flux. The optimized compression of the response time of the entire system enables the magnetic field control rhythm to accurately match the real-time deformation rate of the steel strip coil, fundamentally solving the problem of increased interlayer slip rate caused by control lag.

[0084] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0085] The present invention optimizes the uncoiling process of steel strip coils through an innovative multi-modal control architecture. Its multi-source sensing network deeply integrates temperature field, geometric deformation, residual stress and magnesium powder concentration data, completely eliminating traditional monitoring blind spots and realizing holographic and accurate perception of interlayer status; the original dynamic prediction engine and historical database high-speed matching mechanism predict the sliding risk within milliseconds and output the magnetic pre-adjustment strategy, significantly improving the system response speed and accident prevention capabilities; the magnetic control model based on physical field characteristics uses edge-enhanced magnetic field generation technology and risk-graded magnetic flux control to simultaneously overcome the industry problems of end face concave damage control and ultra-high energy consumption; the self-evolving database system continuously absorbs effective working conditions and optimizes feature space to ensure that the equipment maintains excellent adaptability in the processing of full-specification steel strip coils, and ultimately achieves the triple progress of essential improvement in end face quality, breakthrough enhancement in interlayer stability and systematic optimization of energy efficiency.

[0086] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An electric permanent magnet unwinding composite magnetic field multi-mode adjustment system, characterized in that: include: A data acquisition and preprocessing module is used to acquire and preprocess the multimodal data of the steel strip coil, wherein the multimodal data includes temperature field distribution data, geometric deformation data, residual stress field data, and magnesium powder concentration distribution data; A graph structure construction module, connected to the data acquisition and preprocessing module, for converting the preprocessed multimodal data into a spatiotemporal correlation graph; A dynamic prediction module, including a graph neural network processing layer and a long short-term memory network prediction layer. The dynamic prediction module is connected to the graph structure construction module to analyze the spatiotemporal evolution characteristics of the graph structure; A historical matching prediction module, which is connected to the data acquisition and preprocessing module and is used to time-regularly match the multimodal data sequence in the current time window with the historical database, select the historical data set with the highest matching degree, and predict the magnetic parameters at the next moment based on the historical data set. It includes a time series database, a real-time matching unit, and a prediction output unit; A magnetic force adjustment strategy module, which is connected to the dynamic prediction module and the historical matching prediction module, is used to fuse the interlayer slip probability vector and the predicted magnetic parameters, generate a partitioned magnetic force parameter matrix based on a preset magnetic flux calculation rule, and realize adaptive allocation of the number of magnetic poles and magnetic flux density; The electro-permanent magnetic actuator is connected to the magnetic force adjustment strategy module, including a reconfigurable magnetic pole array for responding to magnetic parameter matrix instructions, and realizing dynamic adjustment of magnetic flux and N / S pole reorganization through excitation coil current regulation and polarity switching circuit.

2. The system according to claim 1, wherein: The historical matching prediction module also includes: Preload execution unit, when matching confidence When the historical matching result is output, the basic magnetic force parameters are sent to the electro-permanent magnetic actuator within 10ms; Exception handling unit, when the highest similarity score in the history matching process When the alarm is triggered, the system switches to pure model prediction mode; Database self-update unit, adding effective working condition records every day Group, record screening criteria is the measured value of end face concavity mm and interlayer sliding rate .

3. The system according to claim 1, wherein: The real-time matching unit of the historical matching prediction module is used to intercept the current time window data sequence, perform similarity matching on the current time window data sequence with the historical data, and calculate the similarity with the historical data through the dynamic time warping algorithm, which specifically includes performing dimensionality reduction processing on the current data window and the historical data respectively, calculating the optimal time alignment path, and calculating the similarity score based on the cumulative distance of the path.

4. The system according to claim 1, wherein: The prediction output unit of the history matching prediction module is used to generate the magnetic parameter prediction value of the next moment and output the weighted magnetic parameter according to the magnetic parameters recorded in the matched historical data and the environmental parameter compensation. The prediction output unit executes the magnetic pole number allocation rule as follows: And the magnetic flux enhancement in the edge region is .

5. The system according to claim 1, wherein: The electro-permanent magnetic actuator comprises Independent magnetic pole units, each unit includes NdFeB permanent magnets, excitation coils, polarity switching circuits and magnetic flux closed-loop control systems.

6. A method for adjusting the multi-mode of an electric permanent magnet unwinding composite magnetic field, applied to the system according to any one of claims 1 to 5, characterized in that include: Step S100: collecting the temperature field distribution data, geometric deformation data, residual stress field data and magnesium powder concentration distribution data of the steel strip coil, and outputting the pre-processed data cube; Step S200: constructing a spatiotemporal association graph structure and calculating node feature vectors and edge weights; Step S300: Outputting an inter-layer sliding probability vector through a dynamic prediction module; Step S400: Execute historical matching prediction, perform dimensionality reduction compression, multi-level matching and dynamic weighted calculation and pre-load execution in sequence, The dimension reduction compression calculation is: , retaining 95% variance; The multi-level matching is divided into coarse screening and fine screening, the coarse screening is to screen the Euclidean distance The fine screening performs DTW similarity calculation on the coarse screening result; The dynamic weighting combines similarity and temperature difference to calculate the weight, and the calculation formula is: , weight ; The specific operation of the preloading execution is to 50ms before the strategy model completes the calculation Medium magnetic flux Pre-energize the area; Step S500: fusing the sliding probability vector and the predicted magnetic parameters to generate a magnetic adjustment strategy; Step S600: driving the electro-permanent magnetic actuator to perform dynamic magnetic field adjustment.

7. The method according to claim 6, characterized in that The step S400 also includes similarity confidence verification and data aging weighting. The similarity confidence verification , when Confidence<0.8, the matching record is discarded. The data aging weighting method is to calculate the historical record weight .

8. The method according to claim 6, characterized in that The time series database update rule in step S400 is to add new records every day Group, implement elimination rules , recalculate the principal component analysis matrix every 72 hours Optimize feature space.

9. The method according to claim 6, characterized in that The magnetic flux optimization in step S500 satisfies , where the rate of change of the magnetic flux density gradient in adjacent regions is .

10. The method according to claim 6, characterized in that The step S600 drives the electro-permanent magnetic actuator to perform dynamic magnetic field adjustment, including performing magnetic pole array response, so that the polarity switching time , through the Hall sensor to correct the magnetic flux deviation in real time to complete the closed-loop feedback control and control the total response delay .

Images