Self-adaptive prediction and control system for anti-surge of steel strip uncoiler

Abstract

The application discloses a self-adaptive prediction and control system for anti-slip of a steel strip uncoiler, and relates to the technical field of steel strip uncoiler processing and automatic control. In the uncoiling preparation stage, the system quantifies three initial disturbance indexes of centering, stress and eccentricity, and performs closed-loop initial correction of automatic secondary centering or stress release. In the stable operation stage, an encoder and a decoder are used to real-time fuse multi-source time sequence data such as main shaft angular velocity, driving torque and steel strip tension, and generate a high-fidelity future slip trajectory prediction sequence. The system further triggers a hierarchical active compensation instruction according to the risk level of the predicted trajectory, drives the correction mechanism in a feedforward control mode, and ensures that the future trajectory converges as a whole within a preset safety threshold. The whole-process closed-loop intelligent control from initial state diagnosis to dynamic process prediction changes the traditional lag response into active intervention, and improves the accuracy of slip suppression.

Description

Technical Field

[0001] This invention relates to the field of steel strip uncoiler processing and automation control technology, specifically to an adaptive prediction and control system for preventing cross-movement in steel strip uncoilers. Background Technology

[0002] In cold and hot rolling production lines for steel strip, the uncoiler is a critical starting point, its function being to unwind the coiled steel strip and feed it into subsequent processing units. The stability of the uncoiling process directly determines the efficiency and product quality of the entire production line. However, during high-speed uncoiling, due to the dynamic changes in the physical properties of the steel coil itself (such as the rapid reduction in diameter from full roll to hollow core), the uneven distribution of interlayer tension, and the coupling effects of gaps and errors in the mechanical transmission system, the steel strip is highly susceptible to lateral displacement, i.e., "slippage." Severe slippage can lead to edge damage, interlayer slippage, steel piling, and even strip breakage, causing significant economic losses and safety hazards.

[0003] Existing technologies primarily address this problem in two ways. The first is to use mechanical limiting devices, such as guide rollers or centering rollers, to restrict the lateral movement of the steel strip through physical contact. This method is simple in structure, but it is a passive correction method and cannot eliminate the tension fluctuations that cause slippage at the source. Furthermore, it may cause secondary damage to the edges of the steel strip during high-speed operation.

[0004] Therefore, how to achieve real-time perception of various dynamic parameters during the unwinding process, and based on this, predict potential future slippage risks, and then proactively and adaptively adjust the control strategy, is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an adaptive prediction and control system for preventing slippage in steel strip uncoilers, thereby solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an adaptive prediction and control system for preventing slippage in a steel strip uncoiler, comprising:

[0007] The initial disturbance state modeling module is used in the uncoiling preparation stage to quantify the potential sources of movement introduced by the inherent defects of the steel coil and the installation process by collecting physical data during the equipment interaction process, so as to construct and evaluate the centering index, stress index and eccentricity index.

[0008] Each risk threshold is preset, and when the alignment quality is determined to be unqualified, a first warning instruction is generated;

[0009] When the stress concentration state is determined to be unqualified, a second early warning instruction is generated;

[0010] When the quality eccentricity is determined to be unqualified, a first compensation instruction is generated and executed.

[0011] Once the unwinding preparation stage is completed successfully, the unwinding machine enters the stable operation stage.

[0012] The dynamic operation trend prediction module is used to collect the spindle angular velocity, drive torque and real-time steel strip tension in real time during the stable operation phase of the uncoiler. The dynamic operation trend prediction module has a built-in hybrid prediction model constructed by the Long Short-Term Memory Network (LSTM) technology to generate a tumbling trajectory prediction sequence that characterizes the change of the edge position of the steel strip within a future time window.

[0013] The active compensation control module is used to generate a second compensation control command based on the predicted sequence of the slippage trajectory, and send the second compensation control command to the correction actuator of the uncoiler to perform feedforward adjustment before the slippage actually occurs.

[0014] Preferably, the initial disturbance state modeling module includes a working area division unit, a visual detection unit, and a sensor data acquisition unit;

[0015] The work area division unit is used to divide the unwinding operation process into an unwinding preparation area and a stable operation area, and to install a distributed multiphysics field sensor network in the unwinding preparation area and the stable operation area.

[0016] Distributed multiphysics sensor networks are deployed in a targeted manner based on the functions of different regions:

[0017] A vision inspection unit, including at least one high-resolution industrial camera and an auxiliary light source, is positioned facing the contact area between the inner wall of the steel coil and the main bearing plate, and performs the following steps:

[0018] Images of the contact area are acquired, and the contour lines of the inner wall of the steel coil and the outer edge contour lines of the main bearing plate are identified and extracted using an image edge detection algorithm. The maximum normal distance between the contour lines of the inner wall of the steel coil and the outer edge contour lines of the main bearing plate along the circumferential direction is calculated and defined as the maximum fitting gap.

[0019] Set the maximum allowable gap threshold. When the maximum fitting gap is less than the maximum allowable gap threshold, the first acquisition command is triggered and sent to the sensor data acquisition unit.

[0020] The sensor data acquisition unit is used to activate the distributed multiphysics sensor network after receiving the first acquisition command, and to acquire the load current time series of the feeding trolley during the entire pushing stroke, the pressure time series during the entire process of hydraulic tensioning of the main shaft, and the drive torque time series during the low-speed jogging rotation of the main shaft.

[0021] Preferably, the method for acquiring the time series of drive torque during low-speed jogging rotation of the spindle includes:

[0022] Continuously collect the time series of the spindle's drive torque and rotational speed;

[0023] When the spindle's drive torque time series is detected to be within the preset low-speed threshold range;

[0024] Record continuous data segments that meet the speed range conditions, and calculate the standard deviation or variance of the rotational speed;

[0025] A preset stability threshold is set. When the standard deviation or variance of the rotation speed is lower than the stability threshold, the continuous time window of the continuous data segment is recorded and identified as the low-speed jogging rotation stage. The data segment that completely corresponds to the time window of the low-speed jogging rotation stage is extracted from the drive torque time series and defined as the time series of the drive torque of the spindle during the low-speed jogging rotation.

[0026] Preferably, the initial disturbance state modeling module further includes a centering process quality identification unit, a stress concentration identification unit, and a mass eccentricity identification unit;

[0027] The centering process quality identification unit applies a moving average filtering algorithm to the load current time series of the loading trolley throughout its entire pushing stroke to generate a baseline trend current series representing the average load level of the pushing process. It then subtracts the load current time series of the loading trolley throughout its entire pushing stroke from the baseline trend current series point by point to generate a residual current series that only represents the instantaneous load fluctuations caused by collisions or misalignment during the pushing process. Based on the residual current series, it calculates the standard deviation and the maximum peak value of the absolute value of the residual current series. After normalizing the standard deviation and the maximum peak value of the absolute value of the residual current series, it calculates the standard deviation and the maximum peak value of the absolute value of the normalized residual current series using a weighted summation method to obtain the centering index, denoted as Sc.

[0028] Preferably, the stress concentration identification unit is used to perform discrete first-order difference operation on the pressure time series during the entire process of spindle hydraulic tensioning to generate a pressure growth rate series;

[0029] All data points in the pressure growth rate sequence are compared point by point with the preset benchmark growth rate threshold. The total number of all points exceeding the threshold is then filtered and counted, and recorded as the instantaneous pressure surge point count.

[0030] Among all the selected instantaneous pressure surge points, identify and record the largest pressure growth rate amplitude;

[0031] After normalizing the instantaneous pressure surge point count and the maximum pressure growth rate amplitude, the normalized instantaneous pressure surge point count and the maximum pressure growth rate amplitude are calculated by weighted summation to obtain the stress index, denoted as Sa.

[0032] Preferably, the mass eccentricity identification unit is used to apply the Fast Fourier Transform (FFT) algorithm to the time series of the driving torque during the low-speed jogging rotation of the spindle, transforming it from the time domain to the frequency domain to generate a torque spectrum.

[0033] The rotational speed is set according to the control system for low-speed jogging rotation, and the corresponding theoretical fundamental frequency is calculated. In the torque spectrum diagram, with the theoretical fundamental frequency as the center, a peak search is performed within a preset search window of ±5% or ±0.2Hz of the theoretical fundamental frequency to locate the spectral peak with the most concentrated energy. The spectral peak is the actual fundamental frequency component caused by the eccentricity of the steel coil mass. The spectral amplitude corresponding to the actual fundamental frequency component is extracted and recorded. The magnitude of the spectral amplitude is proportional to the amplitude of the periodic compensation torque that the drive system must apply to overcome the mass eccentricity. After dimensionless processing, it is defined as the eccentricity index, denoted as Me.

[0034] Preferably, the initial disturbance state modeling module further includes a first determination unit, a second determination unit, and a third determination unit;

[0035] The first determination unit is used to receive the alignment index Sc and compare the alignment index Sc with a preset alignment risk threshold. When the alignment index Sc is less than the alignment risk threshold, the alignment quality is determined to be qualified. When the alignment index Sc is greater than or equal to the alignment risk threshold, the alignment quality is determined to be unqualified, and a first warning instruction is generated, including: generating and executing a second automatic alignment instruction, which includes reducing the pushing speed of the loading trolley and reducing the pushing drive current.

[0036] After the second automatic alignment command is completed, the alignment index Sc is recalculated. If the newly calculated Sc value is lower than the unqualified threshold, the process continues; if it is still unqualified, it will switch to executing a mandatory manual intervention alarm.

[0037] The second determination unit receives the stress index Sa and compares it with a preset stress risk threshold. When the stress index Sa is less than the stress risk threshold, the stress concentration state is deemed acceptable. When the stress index Sa is greater than or equal to the stress risk threshold, the stress concentration state is deemed unacceptable, and a second warning instruction is generated. This instruction includes: generating and executing a stress release and reconstruction instruction, which controls the spindle hydraulic system to first perform a complete depressurization operation, and then re-execute hydraulic tensioning at a preset slow rate lower than the normal tensioning rate. After the stress release and reconstruction instruction is executed, the stress index Sa is recalculated. If the newly calculated stress index Sa is lower than the stress risk threshold, the process continues. If it is still greater than or equal to the stress risk threshold, a mandatory manual intervention alarm is triggered.

[0038] The third determination unit is used to receive the eccentricity index Me and compare the eccentricity index Me with a preset eccentricity risk threshold; when the eccentricity index Me is less than the eccentricity risk threshold, the quality eccentricity state is determined to be qualified; when the eccentricity index Me is greater than or equal to the eccentricity risk threshold, the quality eccentricity state is determined to be unqualified, triggering and executing a first compensation instruction, the first compensation instruction including:

[0039] Spectral analysis is performed on the time series of the drive torque during the low-speed jogging rotation of the spindle to identify the actual fundamental frequency component and its corresponding spectral amplitude of the periodic disturbance caused by mass eccentricity, and to generate a feedforward compensation signal for the downstream correction control system. The feedforward compensation signal is constructed as a periodic signal with a frequency synchronized with the actual fundamental frequency component and a phase that is 180 degrees out of phase with the periodic disturbance caused by mass eccentricity. The compensation amplitude is determined based on the eccentricity index Me.

[0040] Preferably, the dynamic operation trend prediction module includes a multi-source data synchronous acquisition unit and a real-time data preprocessing unit;

[0041] The multi-source data synchronous acquisition unit is used to collect the spindle angular velocity ω(t), driving torque T(t) and real-time tension F(t) of the steel strip in real time and synchronously at a preset sampling frequency during the stable operation phase of the uncoiler, so as to form a dynamic time-series data stream;

[0042] The real-time data preprocessing unit is used to normalize the dynamic time-series data stream and to segment the normalized data using a sliding window method to construct the input feature sequence.

[0043] Preferably, the dynamic running trend prediction module further includes an encoding feature fusion unit and a trajectory sequence decoding unit;

[0044] The coding feature fusion unit is used to perform the following operations:

[0045] The encoder maps the dynamic time-series data stream into a guiding vector containing time-series dynamic characteristics.

[0046] The trajectory sequence decoding unit is used to autoregressively generate a predicted sequence of undulating trajectories that characterizes the changes in the position of the steel strip edge within a future time window, using the guiding vector as the initial input or condition.

[0047] Preferably, before generating the second compensation control command, the predicted trajectory sequence is determined to determine whether there is any predicted point in the predicted trajectory sequence whose absolute value of position deviation is greater than a first preset position deviation threshold, or whose absolute value of predicted velocity is greater than a first preset velocity threshold; the first preset position deviation threshold is 5mm, and the first preset velocity threshold is 5mm / s;

[0048] If the existence is determined, a second active compensation command is triggered to generate a second compensation control command based on the predicted sequence of the axial movement trajectory. The second compensation control command is used to drive the correction actuator of the uncoiler to move towards the compensation target position at a preset maximum allowable operating speed or maximum allowable acceleration, adjust the lateral physical position of the uncoiler spindle, and control the flow and direction of hydraulic oil by adjusting the opening of the servo valve until the predicted position of any prediction point is within 2mm.

[0049] This invention provides an adaptive predictive and control system for preventing slippage in a steel strip uncoiler. It offers the following advantages:

[0050] (1) By performing quantitative diagnosis and closed-loop correction on the initial unwinding state, potential disturbances caused by misalignment, stress concentration, etc., are actively eliminated or suppressed before high-speed operation. This method improves the initial conditions from the root, lays a solid foundation for subsequent high-speed stable operation, and significantly reduces the risk of instability during operation.

[0051] (2) A deep learning model is used to analyze multi-source dynamic data streams in real time, enabling high-precision prediction of future trajectory changes. This predictive capability transforms traditional hysteresis feedback control into advanced feedforward compensation, gaining a valuable execution time window for correction actions and fundamentally overcoming the response delay of the physical system.

[0052] (3) A risk-prediction-based hierarchical control strategy was established, which only initiates high-intensity intervention when a severe trend of deviation is predicted. This intelligent decision-making mechanism achieves optimized allocation of control resources, avoids excessive actions by the actuators, and effectively protects the equipment and reduces energy consumption while ensuring the corrective effect.

[0053] (4) Through closed-loop intelligent control of the entire process, the control accuracy of the steel strip edge position and the system robustness under high-speed operation are significantly improved. This directly improves the winding quality and pass rate of the final product, enhances the production line's ability to cope with sudden working conditions, and provides strong technical support for improving overall production efficiency. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the system flow of the initial disturbance state modeling module of the present invention during the unwinding preparation stage;

[0055] Figure 2 This is a schematic diagram of the system flow of the dynamic operation trend prediction module and the active compensation control module of the present invention during the stable operation phase of the uncoiler. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0057] Example 1

[0058] Please see Figure 1 This invention provides an adaptive predictive and control system for preventing slippage in a steel strip uncoiler, comprising:

[0059] The initial disturbance state modeling module is used to quantify the potential sources of movement introduced by the inherent defects of the steel coil and the installation process by collecting physical data during the uncoiling preparation stage. This is done to construct the centering index Sc, stress index Sa, and eccentricity index Me. Once the uncoiling preparation stage is qualified, the uncoiling machine enters the stable operation stage.

[0060] The dynamic operation trend prediction module is used to collect the spindle angular velocity ω(t), driving torque T(t) and real-time strip tension F(t) in real time during the stable operation phase of the uncoiler. The dynamic operation trend prediction module has a built-in hybrid prediction model constructed by the Long Short-Term Memory Network (LSTM) technology to generate a tumbling trajectory prediction sequence that characterizes the change of the strip edge position within a future time window.

[0061] The active compensation control module is used to generate a second compensation control command based on the predicted sequence of the slippage trajectory, and send the second compensation control command to the correction actuator of the uncoiler to perform feedforward adjustment before the slippage actually occurs.

[0062] In this embodiment, in response to the problem that the correction control in the prior art is lagging and cannot suppress disturbances at the source, the present invention proposes a technical path from source diagnosis to process prediction and then to feedforward control.

[0063] First, by constructing centering, stress, and eccentricity indices during the uncoiling preparation stage, a quantitative assessment of inherent defects in the steel coil and potential sources of axial movement introduced during installation is achieved. This mechanism enables the system to grasp the initial boundary conditions leading to axial movement before high-speed operation begins, changing the passive situation of traditional control methods that can only respond after problems occur.

[0064] Secondly, the system combines multi-dimensional dynamic data during operation and uses a long short-term memory network model to predict future movement trajectories, generating a second compensation control command based on the prediction results. This method can proactively adjust before actual displacement occurs at the steel strip edge, avoiding the lag of traditional passive correction, thereby improving the timeliness and accuracy of control, helping to reduce the risk of steel strip edge damage and enhance production stability.

[0065] Example 2

[0066] This embodiment is an explanation based on Embodiment 1. Please refer to it. Figure 1 Specifically, the initial disturbance state modeling module includes a working area division unit, a visual detection unit, and a sensor data acquisition unit.

[0067] The work area division unit is used to divide the unwinding operation process into an unwinding preparation area and a stable operation area, and to install a distributed multiphysics field sensor network in the unwinding preparation area and the stable operation area.

[0068] Distributed multiphysics sensor networks are deployed in a targeted manner based on the functions of different regions:

[0069] In the uncoiling preparation area, the entire process of the feeding trolley pushing the steel coil to the main shaft and completing the tensioning is covered. At least the following are installed in this area: a current sensor coupled to the drive motor of the feeding trolley, a pressure sensor integrated into the pipeline of the main shaft hydraulic tensioning system, and a torque or speed sensor installed on the drive end of the main shaft.

[0070] In the stable operating zone, the steel strip is continuously unwound at the working speed. In this zone, at least the following are deployed: an edge position sensor deployed downstream of the uncoiler for non-contact measurement of the lateral displacement of the steel strip edge, and a torque or speed sensor is continuously used.

[0071] A vision inspection unit, including at least one high-resolution industrial camera and an auxiliary light source, is positioned facing the contact area between the inner wall of the steel coil and the main bearing plate, and performs the following steps:

[0072] Images of the contact area are acquired, and the contour lines of the inner wall of the steel coil and the outer edge contour lines of the main bearing plate are identified and extracted using an image edge detection algorithm. The maximum normal distance between the two contour lines along the circumferential direction is calculated and defined as the maximum fitting gap.

[0073] Set the maximum allowable gap threshold. When the maximum fitting gap is less than the maximum allowable gap threshold, the first acquisition command is triggered and sent to the sensor data acquisition unit.

[0074] The sensor data acquisition unit is used to activate the distributed multiphysics sensor network after receiving the first acquisition command, and to acquire the load current time series of the feeding trolley during the entire pushing stroke, the pressure time series during the entire process of hydraulic tensioning of the main shaft, and the drive torque time series during the low-speed jogging rotation of the main shaft.

[0075] In this embodiment, traditional systems typically proceed directly to the tensioning or jogging stage. If there is a significant initial misalignment of the steel coil, it may lead to equipment damage or invalid cut data. This invention introduces a visual inspection unit to pre-check the mechanical bonding gap, effectively preventing subsequent operations from being performed under substandard physical installation conditions. This reduces the risk of equipment damage and ensures the effectiveness of the entire diagnostic process from the outset.

[0076] By establishing a "visual inspection pass" pre-triggered condition, it is ensured that the subsequent current, pressure, and torque sequences are generated on a qualified mechanical reference. This eliminates the significant noise and interference introduced by severely flawed initial installation, allowing the subsequently calculated alignment index, stress index, and eccentricity index to more accurately reflect the inherent, subtle disturbance characteristics of the steel coil, providing a more precise initial state vector for subsequent dynamic trend prediction models. The integration of work area division, sensor deployment, visual inspection, condition triggering, and data acquisition into an automated closed-loop process replaces the traditional model that might rely on operator experience. This not only improves production efficiency but, more importantly, ensures that the initial state assessment before each uncoiling operation follows the same set of objective and quantitative standards, enhancing process consistency and stability.

[0077] Example 3

[0078] This embodiment is an explanation based on Embodiment 1. Please refer to it. Figure 1 Specifically, the method for acquiring the time series of drive torque during low-speed jogging rotation of the spindle includes:

[0079] Used for continuously acquiring the drive torque time series and speed time series of the spindle;

[0080] When the rotational speed value in the time series of the spindle's drive torque is identified to be within the low-speed threshold range, the preferred value for the low-speed threshold range is: greater than 1 RPM and less than 10 RPM.

[0081] Record continuous data segments that meet the speed range conditions, and calculate the standard deviation or variance of the rotational speed;

[0082] A preset stability threshold is set. When the standard deviation or variance of the rotation speed is lower than the stability threshold, the continuous time window of the continuous data segment is recorded and identified as the low-speed jogging rotation stage. The data segment that completely corresponds to the time window of the low-speed jogging rotation stage is extracted from the drive torque time series and defined as the time series of the drive torque of the spindle during the low-speed jogging rotation.

[0083] In this embodiment, a rough screening is first performed using a "low-speed threshold range" (preferably 1-10 RPM). The physical meaning of this range is as follows: speeds above 1 RPM ensure that the spindle has overcome static friction and entered dynamic rotation, avoiding data contamination from instantaneous current surges and aperiodic jitter; speeds below 10 RPM ensure the system is in a low dynamic response state, where mass eccentricity is the absolute dominant factor in periodic torque fluctuations, and the influence of other high-frequency disturbances (such as transmission chain vibration) is effectively suppressed. Within the aforementioned target range, this scheme introduces a crucial second verification—speed stability judgment. By calculating the standard deviation or variance of the speed within a continuous data segment and comparing it with the "stability threshold," a quantitative assessment of the rotational state is achieved. Standard deviation or variance is a statistical indicator that measures the degree of data dispersion; an extremely low speed standard deviation means that the spindle is rotating at an angular velocity very close to a constant value.

[0084] Different steel coil weights, sizes, and equipment conditions all affect the time and process required for the spindle to reach a stable low speed from rest. Traditional fixed-delay or fixed-duration data acquisition methods may, under certain operating conditions, capture data from the acceleration phase, leading to inaccurate results. This solution dynamically determines the optimal acquisition window by monitoring the statistical characteristics of the rotational speed in real time, enabling it to automatically adapt to various operating conditions without manual parameter adjustments. This ensures that high-quality, valid data segments are captured under any conditions, demonstrating strong robustness.

[0085] Example 4

[0086] This embodiment is an explanation based on Embodiment 1. Please refer to it. Figure 1 Specifically, the initial disturbance state modeling module also includes a centering process quality identification unit, a stress concentration identification unit, and a mass eccentricity identification unit;

[0087] The centering process quality identification unit applies a moving average filtering algorithm to the load current time series of the loading trolley throughout its entire pushing stroke to generate a baseline trend current series representing the average load level of the pushing process. It then subtracts the load current time series of the loading trolley throughout its entire pushing stroke from the baseline trend current series point by point to generate a residual current series that only represents the instantaneous load fluctuations caused by collisions or misalignment during the pushing process. Based on the residual current series, it calculates the standard deviation and the maximum peak value of the absolute value of the residual current series. After normalizing the standard deviation and the maximum peak value of the absolute value of the residual current series, it calculates the standard deviation and the maximum peak value of the absolute value of the normalized residual current series using a weighted summation method to obtain the centering index, denoted as Sc.

[0088] The stress concentration identification unit is used to perform discrete first-order difference operation on the pressure time series of the entire process of spindle hydraulic tensioning to generate a pressure growth rate sequence. Each data point in the pressure growth rate sequence represents the instantaneous rate of change of pressure at the corresponding sampling time.

[0089] All data points in the pressure growth rate sequence are compared point by point with the preset benchmark growth rate threshold. The total number of all points exceeding the threshold is then filtered and counted, and recorded as the instantaneous pressure surge point count.

[0090] Among all the selected instantaneous pressure surge points, identify and record the largest pressure growth rate amplitude;

[0091] After normalizing the instantaneous pressure surge point count and the maximum pressure growth rate amplitude, the normalized instantaneous pressure surge point count and the maximum pressure growth rate amplitude are calculated by weighted summation to obtain the stress index, denoted as Sa.

[0092] The mass eccentricity identification unit applies a Fast Fourier Transform (FFT) algorithm to the time series of the drive torque during low-speed jogging rotation of the spindle, transforming it from the time domain to the frequency domain to generate a torque spectrum. The torque spectrum shows the energy distribution of the torque signal at different frequencies. Based on the speed set by the control system for low-speed jogging rotation, the corresponding theoretical fundamental frequency is calculated. In the torque spectrum, peak search is performed within a preset, extremely small frequency search window centered on the theoretical fundamental frequency to locate the spectral peak with the most concentrated energy. This spectral peak is the actual fundamental frequency component caused by the mass eccentricity of the steel coil. The spectral amplitude corresponding to the actual fundamental frequency component is extracted and recorded. The magnitude of the spectral amplitude is proportional to the amplitude of the periodic compensation torque that the drive system must apply to overcome the mass eccentricity. After dimensionless processing, it is defined as the eccentricity index, denoted as Me.

[0093] Technical principle: The centering index (Sc) captures disturbances during the "installation process".

[0094] The physical process, Sc, is calculated by analyzing the load current during the process of the feeding trolley pushing the steel coil. This physical action, "pushing and feeding," is a separate and brief event in the uncoiling preparation stage.

[0095] The uniqueness of the time window: Once the coil is pushed onto the spindle and tensioned, the loading trolley has completed its task and returns, and its load current no longer contains any information related to the centering quality. Therefore, the data window for capturing the quality of the centering process exists only in this specific stage of the uncoiling preparation phase.

[0096] A chart example of the calculation process for the median index (Sc):

[0097] Prerequisites:

[0098] Calculation formula: ;

[0099] Weighting: Set weights w1=0.4 (standard deviation weight) and w2=0.6 (maximum peak weight). This weighting slightly emphasizes the severity of a single, severe impact event.

[0100] Normalization method: All normalizations (Norm) in the table use the maximum value normalization method, i.e., normalized value = original value / maximum value of the feature in all samples. A chart example of the calculation process for the median index (Sc) is shown in Table 1 below;

[0101] Table 1: Example of the calculation process for the Centralization Index (Sc):

[0102]

[0103] Diagnostic objective: Sc, in essence, is a quantitative assessment of the "quality of the initial 'handshake' between the steel coil and the spindle." The quality of this "handshake" (whether it is smooth, whether there is any collision or jamming) directly determines an important initial boundary condition for subsequent operation.

[0104] The physical principle of the centering index (Sc): This index is based on the application of Newton's second law in a motor drive system. The load current of the loading trolley motor is proportional to its output torque, which is directly used to overcome all resistance during the pushing process. The baseline current extracted through moving average filtering represents the basic power consumption required for ideal, smooth pushing. The residual current sequence accurately reflects the instantaneous mechanical shock and friction caused by collisions, scrapes, or jamming between the steel coil and the guide device. Therefore, the standard deviation of the residual current quantifies the smoothness of the process (continuous small disturbances), while the maximum peak value quantifies the impact intensity of the process (occasional large disturbances). The combination of these two constitutes a complete physical profile of the dynamic quality of the "installation process".

[0105] The stress index (Sa) captures the disturbances of the "tightening process";

[0106] The physical process, Sa, is calculated by analyzing the rate of pressure change during the tensioning of the steel coil in the spindle hydraulic system. This physical action, "hydraulic tensioning," is also a core step in the uncoiling preparation stage.

[0107] The uniqueness of the time window: The tensioning process is designed to open the main bearing plates and tightly grip the inner wall of the steel coil. Pressure surges and unevenness during this process directly reflect geometric defects in the inner wall of the steel coil or instability in the tensioning process. Once tensioning is complete, in the stable operation phase, the hydraulic system's main task is "pressure maintenance," and the pressure curve becomes relatively flat, making it impossible to capture the characteristic information of the tensioning instant. Therefore, the characteristic data for diagnosing the distribution of tensioning stress only exist in this specific stage of uncoiling preparation.

[0108] A graphical example of the stress index (Sa) calculation process:

[0109] Prerequisites:

[0110] Calculation formula: ;

[0111] Weighting: Set weights α=0.5 (surge frequency weight) and β=0.5 (surge intensity weight). This allocation assumes that the frequency of a problem and the severity of an individual incident are equally important.

[0112] Key parameters: Set the baseline growth rate threshold Pratethreshold = 10 bar / s. Any pressure increase exceeding this rate is considered a "surge" event. Normalization method: Use maximum value normalization.

[0113] Table 2 below shows a graphical example of the stress index Sa calculation process;

[0114] Table 2: Example of the stress index Sa calculation process:

[0115]

[0116] Diagnostic objective: Sa is the quantification of the "initial state of the steel coil being fixed". A fixed state with uneven stress distribution is like a tire that is installed in a twisted manner; it will inevitably produce vibration and instability when rotating at high speed.

[0117] The eccentricity index (Me) captures perturbations in the "inherent properties of steel coils";

[0118] The physical process, Me, is calculated by analyzing the drive torque of the spindle during low-speed jogging rotation. This "low-speed jogging" is a specially designed diagnostic action.

[0119] Safety: Directly starting at high speed when the degree of eccentricity of the steel coil is unknown is extremely dangerous. Low-speed jogging (e.g., 510 RPM) is a safe and controllable "check-up" method.

[0120] Signal purity: At low speeds, torque disturbances caused by factors such as tension fluctuations and air resistance are very small. At this point, the periodic fluctuations in torque are almost entirely dominated by mass eccentricity. This allows for very clean and accurate extraction of the fundamental frequency component caused by eccentricity through FFT analysis. However, if torque is analyzed at high speeds, the signal will be mixed with a large amount of other noise, making it difficult to accurately separate the single factor of eccentricity.

[0121] The physical principle of the stress index (Sa): This index is based on fluid pressure and solid mechanics. Ideally, during hydraulic tensioning, the main bearing plate expands uniformly, the inner wall of the steel coil experiences uniform stress, and the system pressure should rise smoothly. However, if there are irregularities on the inner wall of the steel coil (such as ellipticity or protrusions), or if there is pre-stress release between layers, the contact state between the bearing plate and the inner wall will abruptly change. This abrupt change in mechanical contact state will instantly alter the load impedance of the hydraulic cylinder, resulting in a spike or "surge" in the instantaneous rate of increase of pressure (i.e., the first derivative) on the pressure curve. Therefore, the count of surge points reflects the frequency of uneven contact events, while the amplitude of the maximum rate of increase reflects the intensity of the most severe stress redistribution; both together quantify the initial stress distribution state when the steel coil is fixed.

[0122] A graphical example of the eccentricity index (Me) calculation process:

[0123] Prerequisites:

[0124] Calculation model: Me=Norm(Apeak), where Apeak is the peak amplitude of the actual torque spectrum found near the theoretical fundamental frequency.

[0125] Normalization method: All normalizations (Norm) in the table use the maximum value normalization method, that is, normalized value = original value / maximum value of the feature in all samples.

[0126] Physical background: The greater the mass eccentricity, the greater the periodic compensation torque required by the motor to "lift" the heavier side during uniform rotation. The amplitude of this compensation torque is directly reflected in the fundamental frequency amplitude Apeak of the torque spectrum. A graphical example of the calculation process for the intermediate value of the eccentricity index (Me) is shown in Table 3 below.

[0127] Table 3: Example of the calculation process for intermediate values ​​of the eccentricity index (Me):

[0128]

[0129] Diagnostic Purpose: Me is the quantification of the inherent physical property of "uniformity of mass distribution of the steel coil itself". It is an examination conducted before the "run" to check whether the "athlete" itself has any inherent imbalance problems.

[0130] The physical principle of the eccentricity index (Me): This index is based on rotational dynamics. Any rigid body with eccentric mass rotating about a fixed axis generates a centrifugal force at its center of mass, which in turn produces a periodic torque associated with the axis of rotation. To maintain uniform spindle rotation, the drive motor must provide a periodic compensating torque of equal magnitude and opposite direction. The frequency of this compensating torque is exactly the same as the spindle rotation frequency (fundamental frequency), and its amplitude is proportional to the mass eccentricity moment (eccentric mass × eccentric radius). By performing a Fast Fourier Transform (FFT) on the drive torque under low-speed steady-state conditions, the amplitude of this fundamental frequency component can be accurately separated from the complex time-domain signal, thus directly and quantitatively measuring the inherent mass imbalance of the steel coil.

[0131] In this embodiment, the present invention clearly decomposes the ambiguous problem of "poor initial state" into three independent, quantifiable physical indicators: Sc (installation process quality), Sa (fixed stress state), and Me (inherent quality attribute). When a steel coil is determined to be high-risk, the system can clearly indicate whether it is caused by collisions during installation (high Sc), uneven inner wall during tensioning (high Sa), or eccentricity of the steel coil itself (high Me). This precise attribution capability provides a direct and objective basis for decision-making in subsequent process improvements, equipment maintenance, or incoming material quality control. Before the steel coil enters high-speed operation, the present invention conducts a comprehensive check on the three key initial disturbance dimensions through a standardized diagnostic process. This constitutes an effective "quality threshold," allowing the system to determine the potential risk level of the current operation before uncoiling based on the comprehensive evaluation results of (Sc, Sa, Me). For high-risk steel coils, intervention measures such as early warning, speed reduction, or reinstallation can be taken to control potential cross-movement risks before they occur, transforming passive correction into proactive risk management, which helps improve the operational stability and yield of the entire production line.

[0132] Example 5

[0133] This embodiment is an explanation based on Embodiment 4. Please refer to it. Figure 1 Specifically, the initial disturbance state modeling module also includes a first decision unit, a second decision unit, and a third decision unit;

[0134] The first determination unit receives the alignment index Sc and compares it with a preset alignment risk threshold. When the alignment index Sc is less than the alignment risk threshold, the alignment quality is determined to be qualified; when the alignment index Sc is greater than or equal to the alignment risk threshold, the alignment quality is determined to be unqualified, and a first warning instruction is generated. This includes generating and executing a second automatic alignment instruction, which contains modified alignment control parameters designed to address the causes of the first alignment failure (usually impact or jamming).

[0135] The centering control parameters include at least: reducing the feeding trolley's pushing speed and reducing the pushing drive current;

[0136] After the second automatic alignment command is completed, the alignment index Sc is recalculated. If the newly calculated Sc value is lower than the unqualified threshold, the process continues; if it is still unqualified, it will switch to executing a mandatory manual intervention alarm.

[0137] Reduce the feeding trolley's push speed: Set the push speed to a preset percentage of the normal speed (e.g., 50%) to achieve a smoother, more flexible contact.

[0138] Reduce the upper limit of the push drive current: limit the maximum thrust of the loading trolley to prevent the steel coil from jamming with the main shaft or causing a violent impact due to excessive force;

[0139] The second determination unit receives the stress index Sa and compares it with a preset stress risk threshold. When the stress index Sa is less than the stress risk threshold, the stress concentration state is deemed acceptable. When the stress index Sa is greater than or equal to the stress risk threshold, the stress concentration state is deemed unacceptable, and a second warning instruction is generated. This instruction includes: generating and executing a stress release and reconstruction instruction, which controls the spindle hydraulic system to first perform a complete depressurization operation, and then re-execute hydraulic tensioning at a preset slow rate lower than the normal tensioning rate. After the stress release and reconstruction instruction is executed, the stress index Sa is recalculated. If the newly calculated stress index Sa is lower than the stress risk threshold, the process continues. If it is still greater than or equal to the stress risk threshold, a mandatory manual intervention alarm is triggered.

[0140] The third determination unit is used to receive the eccentricity index Me and compare the eccentricity index Me with a preset eccentricity risk threshold; when the eccentricity index Me is less than the eccentricity risk threshold, the quality eccentricity state is determined to be qualified; when the eccentricity index Me is greater than or equal to the eccentricity risk threshold, the quality eccentricity state is determined to be unqualified, triggering and executing a first compensation instruction, the first compensation instruction including:

[0141] Spectral analysis is performed on the time series of the drive torque during the low-speed jogging rotation of the spindle to identify the actual fundamental frequency component and its corresponding spectral amplitude of the periodic disturbance caused by mass eccentricity, and to generate a feedforward compensation signal for the downstream correction control system. The feedforward compensation signal is constructed as a periodic signal with a frequency synchronized with the actual fundamental frequency component and a phase that is 180 degrees out of phase with the periodic disturbance caused by mass eccentricity. The compensation amplitude is determined based on the eccentricity index Me.

[0142] The first, second, and third judgment units construct a hierarchical risk management and control framework, mitigating or suppressing potential operational risks during the open-book preparation stage, fundamentally ensuring stability and security during high-speed operation. This invention establishes a clear risk management process through preset risk thresholds and step-by-step execution instructions.

[0143] The first layer (automatic correction): For alignment and stress problems that can be solved by adjusting operating parameters, the system performs autonomous repairs, which is the first line of defense for risk mitigation.

[0144] The second layer (feedforward compensation): For unavoidable eccentricity problems, the system generates a compensation signal, transforming an unknown and destructive disturbance into a known and controllable factor. This is a key means of risk mitigation.

[0145] The third layer (human intervention): The system only escalates to a mandatory alarm after automatic correction fails, ensuring that the operator's attention is directed to difficult problems that the system cannot resolve on its own. This hierarchical framework ensures that the vast majority of potential risks are effectively addressed before high-speed operation, greatly reducing the probability of steel belt slippage, equipment vibration, or even damage during high-speed operation, and providing a solid guarantee for the safe, stable, and efficient operation of the entire production line.

[0146] In this embodiment, through the first, second, and third decision units, an intelligent upgrade from passive diagnosis to proactive closed-loop correction is achieved, significantly improving the system's autonomy and robustness. Traditional systems, upon detecting an anomaly, typically only issue an alarm and wait for manual intervention. This invention, through three decision units, constructs an automated "diagnosis-decision-execution-re-verification" closed-loop control logic:

[0147] If the alignment quality is not up to standard (Sc exceeds the limit), the system does not shut down directly. Instead, it autonomously tries a more "gentle" secondary alignment strategy (reducing speed and thrust). This adaptive retry mechanism can automatically resolve most problems caused by improper initial contact.

[0148] For stress concentration defects (Sa exceeding the standard), the system can perform a standardized "stress release and reconstruction" operation, actively eliminating uneven stress and restoring the initial fixed state through pressure relief and slow re-tensioning. This closed-loop correction capability enables the system to autonomously handle common anomalies in the uncoiling preparation stage, reducing unnecessary downtime and manual intervention, and greatly enhancing the automation level of the entire uncoiling process and its adaptability (robustness) to different working conditions.

[0149] Example 6

[0150] This embodiment is an explanation based on Embodiment 1. Please refer to it. Figure 1 Specifically, the dynamic operation trend prediction module includes a multi-source data synchronous acquisition unit and a real-time data preprocessing unit;

[0151] The multi-source data synchronous acquisition unit is used to collect the spindle angular velocity ω(t), driving torque T(t) and real-time tension F(t) of the steel strip in real time and synchronously at a preset sampling frequency during the stable operation phase of the uncoiler, so as to form a dynamic time-series data stream;

[0152] The real-time data preprocessing unit is used to normalize the dynamic time-series data stream and to segment the normalized data using a sliding window method to construct the input feature sequence.

[0153] The dynamic operation trend prediction module also includes an encoding feature fusion unit and a trajectory sequence decoding unit;

[0154] The coding feature fusion unit is used to perform the following operations:

[0155] The encoder maps the dynamic time-series data stream into a guiding vector containing time-series dynamic characteristics.

[0156] The trajectory sequence decoding unit is used to guide the vector as the initial input or condition to autoregressively generate a predicted sequence of undulating trajectories that characterizes the changes in the position of the steel strip edge within a future time window.

[0157] Data example of encoder-decoder model:

[0158] The most classic and effective way to implement this process is to use an encoder-decoder architecture;

[0159] The encoder is responsible for "reading" and "understanding" the input historical time-series data. The encoder's task is to compress a variable-length historical information into a fixed-length context vector containing core dynamic features. This vector is the guiding vector in this invention.

[0160] The decoder is responsible for both "writing" and "predicting." It receives the guiding vector generated by the encoder as initial instructions and then generates the predicted sequence step by step.

[0161] Step 1: Input Data Preparation and Matrixing

[0162] The system first needs to convert continuous physical signals into digital matrices that the model can process.

[0163] Sampling frequency: 100Hz (sampled once every 0.01 seconds);

[0164] Historical time window: 0.5 seconds (i.e., using data from the past 50 time points);

[0165] Forecast time window: 0.2 seconds (i.e., forecasting 20 points in the future);

[0166] At any time t, the system will construct an input feature sequence (Input-Sequence-Matri) with dimensions [50,3].

[0167] Example data scenario: Suppose that in the past 0.5 seconds, the tension F(t) of the steel strip experienced a sudden and sharp drop, which is a typical precursor to the instability and slippage of the steel strip.

[0168] Input feature sequence (dimension: 50×3):

[0169] Each row represents data at a single time point [ω, T, F]. The following is an example of the normalized feature sequences, as shown in Table 4:

[0170] Table 4: Examples of Normalized Feature Sequences

[0171]

[0172] The data is normalized (e.g., scaled to between 01 and 11) to facilitate model training.

[0173] This 50×3 matrix fully captures the dynamic process of the "tension instability" event.

[0174] Step 2: The encoder generates the guiding vector:

[0175] The encoder (set to be a recurrent neural network such as LSTM or GRU) will "read" the above feature sequence line by line.

[0176] At t0.50s, the encoder reads the first line [0.51,0.65,0.88] and updates its internal hidden state.

[0177] At t0.49s, it reads the second line [0.50,0.64,0.88] and, combined with the hidden state from the previous step, updates its hidden state again.

[0178] ...This process continues until the encoder has read the last line of data (at the current time t).

[0179] After processing all 50 time points, the encoder's final hidden state is a fixed-length vector. This vector has compressed all the important dynamic information from the past 0.5 seconds (especially the tension drop and its chain reaction) into it.

[0180] This final hidden state is the guiding vector.

[0181] Example data: Set the dimension of the guiding vector to 128.

[0182] Input-Sequence-Matri(50×3)>[Encoder]>Guidance-Vector(1×128)

[0183] Guidance-Vector=[0.98,0.45,0.12,...,0.77]

[0184] Each value in this vector has no direct physical meaning, but their combination pattern is "understood" by the model as a signal that "a rapid unidirectional deviation caused by a sudden drop in tension is about to occur."

[0185] Step 3: Decoder autoregression generates predicted sequences:

[0186] Now, the decoder (also LSTM or GRU) begins its work. Its task is to generate the trajectories [position, velocity] for the next 20 time points, guided by the guidance vector.

[0187] The meaning of "autoregressive" is that the next prediction depends on the prediction result of the previous step.

[0188] The generation process is as follows:

[0189] Predicting step 1 (t+0.01s):

[0190] Input: The decoder takes the Guidance Vector as its initial hidden state. It also takes a special "start" marker (e.g., a zero vector [0,0]) as input.

[0191] Output: The decoder calculates and outputs the first predicted point P1=[0.15,1.8].

[0192] Predicting step 2 (t+0.02s):

[0193] Input: The decoder takes the output P1 from the previous step as input for this step. Its hidden state is also inherited from the previous step.

[0194] Output: Based on its internal state and input P1, the decoder calculates and outputs the second prediction point P2=[0.34,2.1].

[0195] Predicting step 3 (t+0.03s):

[0196] Input: Use the output P2 from the previous step as the input for this step.

[0197] Output: The decoder outputs the third prediction point P3=[0.56,2.5].

[0198] ...repeatedly: this process is repeated 20 times.

[0199] Ultimately, the decoder will generate a complete sequence of predicted trajectories.

[0200] Data example:

[0201] Guiding vector: [0.98, 0.45, 0.12, ..., 0.77] (Signal: Sudden drop in tension, indicating a large deflection)

[0202] The generated predicted sequence of roving trajectories (dimension: 20×2) is shown in Table 5:

[0203] Table 5: Predicted Sequence of Movement Trajectory

[0204]

[0205] This 20×2 prediction sequence is the direct input to the active compensation control module, which will calculate a precise compensation command based on this sequence to offset this severe deviation. At t+0.19s and t+0.20s, 5.2 exceeds the first preset speed threshold of 5mm / s; 5.4 exceeds the first preset position threshold of 5mm, thus triggering the second compensation control command.

[0206] Before generating the second compensation control command, the predicted trajectory sequence is judged to determine whether there is any predicted point in the predicted trajectory sequence whose absolute value of position deviation is greater than a preset first preset position deviation threshold, or whose absolute value of predicted velocity is greater than a preset first preset velocity threshold; the first preset position threshold is 5mm, and the first preset velocity threshold is 5mm / s;

[0207] If the existence is determined, a second active compensation command is triggered. This command is used to generate a second compensation control command based on the predicted sequence of the axial movement trajectory. The second compensation control command is used to drive the correction actuator of the uncoiler to move towards the compensation target position at the preset maximum allowable operating speed or maximum allowable acceleration. This adjusts the lateral physical position of the uncoiler spindle. By adjusting the opening of the servo valve (by controlling the voltage / current signal), the flow and direction of the hydraulic oil are controlled until the predicted position of any prediction point is within 2mm.

[0208] In this embodiment, traditional belt slip correction systems rely on edge position sensors detecting actual deviations before responding, resulting in inherently delayed control. This invention, through an encoder-decoder model, can analyze the dynamic correlations in multi-source data streams such as angular velocity, torque, and tension in real time. It predicts the slip trajectory over a future period (e.g., 0.2 seconds) before the slippage physically occurs and is captured by sensors. As shown in the example, in the initial stage of a sudden tension drop, the model can generate a predictive sequence pointing to a future severe slippage of 5.41 mm. This 0.2-second "foresight" window is crucial, enabling the compensation system to shift from passively "correcting deviations" to actively "counteracting disturbances," fundamentally overcoming the delay between physical measurement and mechanical execution. Belt slippage is the result of nonlinear and dynamic coupling of factors such as angular velocity, driving torque, and tension. Simple linear models or threshold judgments cannot capture this complex intrinsic correlation. The encoder-decoder architecture employed in this invention compresses multi-dimensional time-series data over a past period (e.g., 0.5 seconds) into a "guiding vector" containing core dynamic characteristics. This vector is a highly abstract and generalized representation of the system's current operating state (such as the "precursor to tension instability" in the example). The decoder, based on this precise "guidance," can autoregressively generate a high-fidelity future trajectory sequence containing point-by-point position and velocity. This approach not only predicts "whether there will be deviation," but also more accurately answers "how there will be deviation" (trajectory shape) and "how fast there will be deviation" (velocity change), providing unprecedentedly refined input to the control system. This invention does not compensate for all prediction deviations with maximum capacity. Instead, it divides the control strategy into two levels: "regular compensation" and "emergency intervention," by setting clear thresholds (such as position > 5mm or velocity > 5mm / s). Regular, small prediction deviations may be handled by a milder control law, while a second active compensation command is triggered only when the model predicts a severe surge that will exceed safety limits, driving the actuator at maximum speed or acceleration. This hierarchical strategy ensures that high-intensity control actions that impact the equipment are only invoked when absolutely necessary, avoiding frequent and violent actuator movements when dealing with minor deviations, thereby reducing energy consumption, minimizing mechanical wear, and extending equipment lifespan. The stopping condition for the second active compensation command is not "the current position returns to within 2mm", but rather "the predicted position of any predicted point is within 2mm". This means that the goal of the control system is to actively shape an acceptable "future trajectory". It will continuously apply a strong control action until the maximum risk point within the entire prediction time window is successfully suppressed within the safety threshold (2mm). This control logic, which targets the future state, can more effectively cope with the inertia and delay of the system, preventing the deviation from exceeding the limit due to insufficient correction, or causing new oscillations due to overcorrection, thereby achieving a higher level of control accuracy and stability.

[0209] The threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value, it is acceptable.

[0210] The above formulas are all derived from software simulation using a large amount of data and are selected to be close to the actual values. The coefficients in the formulas are set by those skilled in the art according to the actual situation. The above description is only a preferred embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any equivalent substitutions or changes made by those skilled in the art within the technical scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the protection scope of the present invention.

Claims

1. An adaptive predictive and control system for preventing slippage in a steel strip uncoiler, characterized in that, include: The initial disturbance state modeling module is used in the uncoiling preparation stage to quantify the potential sources of movement introduced by the inherent defects of the steel coil and the installation process by collecting physical data during the equipment interaction process, so as to construct and evaluate the centering index, stress index and eccentricity index. The initial disturbance state modeling module includes a working area division unit, a visual detection unit, and a sensor data acquisition unit; The work area division unit is used to divide the unwinding operation process into an unwinding preparation area and a stable operation area, and to install a distributed multiphysics field sensor network in the unwinding preparation area and the stable operation area. Distributed multiphysics sensor networks are deployed in a targeted manner based on the functions of different regions: A vision inspection unit, including at least one high-resolution industrial camera and an auxiliary light source, is positioned facing the contact area between the inner wall of the steel coil and the main bearing plate, and performs the following steps: Images of the contact area are acquired, and the contour lines of the inner wall of the steel coil and the outer edge contour lines of the main bearing plate are identified and extracted using an image edge detection algorithm. The maximum normal distance between the contour lines of the inner wall of the steel coil and the outer edge contour lines of the main bearing plate along the circumferential direction is calculated and defined as the maximum fitting gap. Set the maximum allowable gap threshold. When the maximum fitting gap is less than the maximum allowable gap threshold, the first acquisition command is triggered and sent to the sensor data acquisition unit. The sensor data acquisition unit is used to activate the distributed multiphysics sensor network after receiving the first acquisition command, and to acquire the load current time series of the feeding trolley during the entire pushing stroke, the pressure time series during the entire process of the main shaft hydraulic tensioning, and the drive torque time series during the low-speed jogging rotation of the main shaft. The initial disturbance state modeling module also includes a centering process quality identification unit, a stress concentration identification unit, and a mass eccentricity identification unit; The centering process quality identification unit applies a moving average filtering algorithm to the load current time series of the loading trolley throughout its entire pushing stroke to generate a baseline trend current series representing the average load level of the pushing process. It then subtracts the load current time series of the loading trolley throughout its entire pushing stroke from the baseline trend current series point by point to generate a residual current series that only represents the instantaneous load fluctuations caused by collisions or misalignment during the pushing process. Based on the residual current series, it calculates the standard deviation and the maximum peak value of the absolute value of the residual current series. After normalizing the standard deviation and the maximum peak value of the absolute value of the residual current series, it calculates the standard deviation and the maximum peak value of the absolute value of the normalized residual current series using a weighted summation method to obtain the centering index, denoted as Sc. The stress concentration identification unit is used to perform discrete first-order difference operations on the pressure time series during the entire process of spindle hydraulic tensioning to generate a pressure growth rate series. All data points in the pressure growth rate sequence are compared point by point with the preset benchmark growth rate threshold. The total number of all points exceeding the threshold is then filtered and counted, and recorded as the instantaneous pressure surge point count. Among all the selected instantaneous pressure surge points, identify and record the largest pressure growth rate amplitude; After normalizing the instantaneous pressure surge point count and the maximum pressure growth rate amplitude, the normalized instantaneous pressure surge point count and the maximum pressure growth rate amplitude are calculated by weighted summation to obtain the stress index, denoted as Sa. The mass eccentricity identification unit is used to apply the Fast Fourier Transform (FFT) algorithm to the time series of the driving torque during the low-speed jogging rotation of the spindle, transforming it from the time domain to the frequency domain to generate a torque spectrum. The theoretical fundamental frequency is calculated based on the speed set by the control system for low-speed jogging rotation. In the torque spectrum diagram, with the theoretical fundamental frequency as the center, a peak search is performed within a preset search window of ±5% or ±0.2Hz of the theoretical fundamental frequency to locate the spectral peak with the most concentrated energy. The spectral peak is the actual fundamental frequency component caused by the mass eccentricity of the steel coil. The spectral amplitude corresponding to the actual fundamental frequency component is extracted and recorded. The magnitude of the spectral amplitude is proportional to the amplitude of the periodic compensation torque that the drive system must apply to overcome the mass eccentricity. After dimensionless processing, it is defined as the eccentricity index, denoted as Me. Each risk threshold is preset, and when the alignment quality is determined to be unqualified, a first warning instruction is generated; When the stress concentration state is determined to be unqualified, a second early warning instruction is generated; When the quality eccentricity condition is determined to be unqualified, a first compensation instruction is generated and executed. Once the unwinding preparation stage is completed successfully, the unwinding machine enters the stable operation stage. The dynamic operation trend prediction module is used to collect the spindle angular velocity, drive torque and real-time steel strip tension in real time during the stable operation phase of the uncoiler. The dynamic operation trend prediction module has a built-in hybrid prediction model constructed by the Long Short-Term Memory Network (LSTM) technology to generate a tumbling trajectory prediction sequence that characterizes the change of the edge position of the steel strip within a future time window. The active compensation control module is used to generate a second compensation control command based on the predicted swerving trajectory sequence, and send the second compensation control command to the correction actuator of the unwinding machine to perform feedforward adjustment before the swerving actually occurs.

2. The adaptive prediction and control system for preventing slippage in a steel strip uncoiler according to claim 1, characterized in that, The method for acquiring the time series of the driving torque of the spindle during low-speed jogging rotation includes: Continuously collect the time series of the spindle's drive torque and rotational speed; When the spindle's drive torque time series is detected to be within the preset low-speed threshold range; Record continuous data segments that meet the speed range conditions, and calculate the standard deviation or variance of the rotational speed; A preset stability threshold is set. When the standard deviation or variance of the rotation speed is lower than the stability threshold, the continuous time window of the continuous data segment is recorded and identified as the low-speed jogging rotation stage. The data segment that completely corresponds to the time window of the low-speed jogging rotation stage is extracted from the drive torque time series and defined as the time series of the drive torque of the spindle during the low-speed jogging rotation.

3. The adaptive prediction and control system for preventing slippage in a steel strip uncoiler according to claim 1, characterized in that, The initial disturbance state modeling module further includes a first determination unit, a second determination unit, and a third determination unit; The first determination unit is used to receive the alignment index Sc and compare the alignment index Sc with a preset alignment risk threshold. When the alignment index Sc is less than the alignment risk threshold, the alignment quality is determined to be qualified. When the alignment index Sc is greater than or equal to the alignment risk threshold, the alignment quality is determined to be unqualified, and a first warning instruction is generated, including: generating and executing a second automatic alignment instruction, the second automatic alignment instruction including reducing the pushing speed of the loading trolley and reducing the pushing drive current. After the second automatic alignment command is completed, the alignment index Sc is recalculated. If the newly calculated Sc value is lower than the unqualified threshold, the process continues; if it is still unqualified, it will switch to executing a mandatory manual intervention alarm. The second determination unit is used to receive the stress index Sa and compare the stress index Sa with a preset stress risk threshold. When the stress index Sa is less than the stress risk threshold, the stress concentration state is determined to be qualified. When the stress index Sa is greater than or equal to the stress risk threshold, the stress concentration state is determined to be unqualified, and a second early warning command is generated, including: generating and executing a stress release and reconstruction command. The stress release and reconstruction command controls the spindle hydraulic system to first perform a complete depressurization operation, and then re-execute hydraulic tensioning at a preset slow rate lower than the normal tensioning rate. After the stress release and reconstruction command is executed, the stress index Sa is recalculated. If the newly calculated stress index Sa value is lower than the stress risk threshold, the process continues. If it is still greater than or equal to the stress risk threshold, it switches to executing a mandatory manual intervention alarm. The third determination unit is used to receive the eccentricity index Me and compare the eccentricity index Me with a preset eccentricity risk threshold; when the eccentricity index Me is less than the eccentricity risk threshold, the quality eccentricity state is determined to be qualified; when the eccentricity index Me is greater than or equal to the eccentricity risk threshold, the quality eccentricity state is determined to be unqualified, triggering and executing a first compensation instruction, the first compensation instruction including: Spectral analysis is performed on the time series of the drive torque during the low-speed jogging rotation of the spindle to identify the actual fundamental frequency component and its corresponding spectral amplitude of the periodic disturbance caused by mass eccentricity, and to generate a feedforward compensation signal for the downstream correction control system. The feedforward compensation signal is constructed as a periodic signal with a frequency synchronized with the actual fundamental frequency component and a phase that is 180 degrees out of phase with the periodic disturbance caused by mass eccentricity. The compensation amplitude is determined based on the eccentricity index Me.

4. The adaptive prediction and control system for preventing slippage in a steel strip uncoiler according to claim 1, characterized in that, The dynamic operation trend prediction module includes a multi-source data synchronous acquisition unit and a real-time data preprocessing unit. The multi-source data synchronous acquisition unit is used to collect the spindle angular velocity ω(t), driving torque T(t), and real-time tension F(t) of the steel strip in real time and synchronously at a preset sampling frequency during the stable operation phase of the uncoiler, so as to form a dynamic time-series data stream. The real-time data preprocessing unit is used to normalize the dynamic time-series data stream and to segment the normalized data using a sliding window method to construct an input feature sequence.

5. The adaptive prediction and control system for preventing slippage in a steel strip uncoiler according to claim 4, characterized in that, The dynamic operation trend prediction module also includes an encoding feature fusion unit and a trajectory sequence decoding unit; The coding feature fusion unit is used to perform the following operations: The encoder maps the dynamic time-series data stream into a guiding vector containing time-series dynamic characteristics. The trajectory sequence decoding unit is used to autoregressively generate the undulating trajectory prediction sequence that characterizes the change in the position of the steel strip edge within the future time window, using the guiding vector as the initial input or condition.

6. The adaptive prediction and control system for preventing slippage in a steel strip uncoiler according to claim 5, characterized in that, Before generating the second compensation control command, the predicted trajectory sequence is determined to determine whether there is any predicted point in the predicted trajectory sequence whose absolute value of position deviation is greater than a first preset position deviation threshold, or whose absolute value of predicted velocity is greater than a first preset velocity threshold; the first preset position deviation threshold is 5mm, and the first preset velocity threshold is 5mm / s; If the existence is determined, a second active compensation command is triggered to generate a second compensation control command based on the predicted sequence of the erratic trajectory. The second compensation control command is used to drive the correction actuator of the uncoiler to move towards the compensation target position at a preset maximum allowable operating speed or maximum allowable acceleration, adjust the lateral physical position of the uncoiler spindle, and control the flow and direction of hydraulic oil by adjusting the opening of the servo valve until the predicted position of any prediction point is within 2mm.

Images