An intelligent factory-based industrial process control method and system

Abstract

The application relates to the technical field of industrial process control, and particularly discloses an industrial process control method and system based on a smart factory, which comprises the following steps: collecting temperature values, speeds and thickness parameters of multiple discrete positions in the width direction of a strip steel to form a process observation set; projecting the process observation set onto a pre-stored two-dimensional heat transfer space mode to obtain a mode coefficient set and complete temperature field order reduction representation; dividing the width direction into multiple continuous segments according to temperature distribution curvature, giving each segment a heat transfer coefficient to be estimated, and splicing the mode coefficient into a joint state vector; recursively updating the joint state vector by using a set Kalman filter to obtain an estimated value of the heat transfer coefficient of each segment; and adjusting the output power of a heating device above each segment according to the estimated value of the heat transfer coefficient of each segment and the reconstructed temperature distribution; and the application can inhibit parameter physical drift, improve strip steel transverse temperature uniformity and control robustness.

Description

Technical Field

[0001] This invention relates to the field of industrial process control technology, and specifically to an industrial process control method and system based on a smart factory. Background Technology

[0002] In the field of industrial process control for smart factories, the strip heat treatment process in continuous annealing furnaces is a key process. The strip undergoes heating and homogenization processes within the annealing furnace, and the temperature uniformity along its width directly affects the mechanical properties and surface quality of the product. To achieve precise control of the transverse temperature distribution of the strip, traditional methods typically involve installing multiple temperature sensors along the width of the strip and adjusting the output power of each heating element based on the collected temperature values.

[0003] The existing technology has the following shortcomings:

[0004] In the heat treatment process of strip steel in continuous annealing furnace, the traditional one-dimensional lumped parameter digital twin model ignores the two-dimensional heat transfer effect in the width direction and does not model the dynamics, resulting in a physically unrealizable drift in the inverted heat transfer coefficient. Consequently, adjusting the heating power based on this drift parameter exacerbates the uneven transverse temperature of the strip steel. Summary of the Invention

[0005] The purpose of this invention is to provide an industrial process control method and system based on smart factories to solve the problems mentioned above.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] An industrial process control method based on a smart factory includes the following steps:

[0008] Step 1: Collect temperature measurements at multiple discrete locations along the width of the strip, and simultaneously acquire the strip running speed and thickness parameters to form a process observation set for the current moment;

[0009] Step 2: Project the process observation set onto the pre-stored two-dimensional heat transfer spatial modes, and obtain the modal coefficient set of each spatial mode by minimizing the projection error, thus completing the reduced-order characterization of the temperature field.

[0010] Step 3: Divide the strip width direction into multiple continuous segments, assign a segment constant heat transfer coefficient to be estimated to each segment, and concatenate all segment constant heat transfer coefficients with the modal coefficient set to form a joint state vector;

[0011] Step 4: Use ensemble Kalman filtering to recursively update the joint state vector, where each ensemble member independently performs prediction and observation correction, and extracts the average value of all members as the estimated value of the constant heat transfer coefficient of each segment at the current time.

[0012] Step 5: Based on the estimated constant heat transfer coefficients of each segment and the width-direction temperature distribution reconstructed from the modal coefficient set, adjust the output power of the heating device above each segment to make the reconstructed temperature distribution approximate the target uniform temperature field.

[0013] As a further aspect of the present invention: the reduced-order characterization of the temperature field specifically includes:

[0014] Based on the temperature measurements and their corresponding width-direction coordinates in the process observation set, a radial basis interpolation matrix is ​​constructed to fit the discrete temperature measurements into a continuously differentiable temperature distribution function along the width direction.

[0015] The temperature distribution function is integrated with each pre-stored spatial mode over a width interval to obtain the original coefficients of each mode.

[0016] A soft thresholding operation is applied to the original coefficients, setting coefficients whose absolute values ​​are less than a preset energy threshold to zero, and retaining the remaining coefficients as the final set of modal coefficients.

[0017] As a further aspect of the present invention: obtaining the original coefficients corresponding to each mode specifically includes:

[0018] The number of Gaussian integration nodes is determined based on the highest order of each spatial mode to be projected, and the corresponding integration node positions and integration weights are generated within the width interval.

[0019] At each integration node, the function value of the temperature distribution function and the modal values ​​of each spatial mode are calculated simultaneously.

[0020] Multiply the product of the function value and the modal value at the same node by the corresponding integral weight, and sum the weighted multiplications at all nodes to obtain the original coefficients of each spatial mode.

[0021] As a further aspect of the present invention: the process of constructing the joint state vector is as follows:

[0022] The temperature distribution curve along the width direction is reconstructed based on the obtained set of modal coefficients, and the second derivative of the temperature distribution curve is used to obtain the curvature variation sequence along the width.

[0023] All locations where the absolute value of curvature exceeds a preset threshold are selected as dividing points, and the width direction is divided into multiple continuous segments of unequal length based on these dividing points;

[0024] The constant heat transfer coefficient to be estimated is set independently in each segment, and all segment constant heat transfer coefficients are arranged sequentially with the original set of modal coefficients to form a joint state vector.

[0025] As a further aspect of the present invention: the reconstructing of the temperature distribution curve in the width direction based on the obtained set of modal coefficients specifically includes:

[0026] Several interpolation nodes are selected at equal intervals along the width direction. At each node, the preset value of each order spatial mode is multiplied by the coefficient of the corresponding order in the modal coefficient set to obtain the contribution value of each order.

[0027] The temperature reconstruction value of the interpolation node is obtained by summing all the order contribution values ​​at the same interpolation node.

[0028] The temperature reconstruction values ​​of all interpolation nodes are connected sequentially in the width direction to form a continuous temperature distribution curve.

[0029] As a further aspect of the present invention: the execution of prediction and observation correction specifically includes:

[0030] Based on the degree of dispersion of the estimated constant heat transfer coefficient of the same segment among all members at the previous moment, the adaptive disturbance amplitude of the corresponding segment is calculated.

[0031] For each set member, generate a random increment centered at zero with an adaptive perturbation amplitude as the variance, and superimpose the increment onto the joint state vector of the previous time step to obtain the prediction vector;

[0032] Each prediction vector is corrected using the process observation set at the current moment, and all corrected members are arithmetically averaged according to the same piecewise constant heat transfer coefficient to obtain the estimated value of the corresponding piecewise constant heat transfer coefficient.

[0033] As a further aspect of the present invention: the process of obtaining the prediction vector is as follows:

[0034] Extract the estimated value of the piecewise constant heat transfer coefficient from the joint state vector of all members at the previous moment, calculate the trace of its covariance matrix, and multiply the trace by a preset scaling factor to obtain the global scaling factor.

[0035] Multiply the adaptive perturbation amplitude of each segment by the global scaling factor to obtain the corrected perturbation amplitude of the corresponding segment.

[0036] For each member of the set, a random number is independently sampled according to the correction perturbation amplitude of each segment, and the random arrays of all segments are combined into a complete random increment vector.

[0037] As a further aspect of the present invention: the adjustment of the output power of the heating device above each segment specifically includes:

[0038] Calculate the difference between the average value of the reconstructed temperature distribution within each segment and the target value of the target uniform temperature field within the corresponding segment to obtain the segmented temperature difference;

[0039] Multiply the segmented temperature difference by a pre-calibrated proportional coefficient based on the estimated segmented constant heat transfer coefficient to obtain the preliminary power correction for the corresponding segment.

[0040] The initial power corrections for adjacent segments are weighted and averaged, with the weighting coefficient determined by the ratio of the estimated constant heat transfer coefficients of the two segments, and the final power adjustment value for each segment is output.

[0041] An industrial process control system based on a smart factory includes:

[0042] The process observation and acquisition module is used to acquire temperature measurements at multiple discrete locations along the width of the strip, and simultaneously obtain the strip running speed and thickness parameters to form a process observation set at the current moment.

[0043] The temperature field order reduction characterization module projects the process observation set onto the pre-stored two-dimensional heat transfer spatial modes, and obtains the modal coefficient set of each spatial mode by minimizing the projection error, thus completing the order reduction characterization of the temperature field.

[0044] The joint state vector construction module divides the strip width direction into multiple continuous segments, assigns a segment constant heat transfer coefficient to be estimated to each segment, and concatenates all segment constant heat transfer coefficients with the modal coefficient set to form a joint state vector.

[0045] The segmented heat transfer coefficient estimation module uses ensemble Kalman filtering to recursively update the joint state vector. Each ensemble member independently performs prediction and observation correction, and the average value of all members is extracted as the estimated value of the constant heat transfer coefficient of each segment at the current time.

[0046] The heating power adaptive adjustment module adjusts the output power of the heating device above each segment according to the estimated value of the constant heat transfer coefficient of each segment and the width direction temperature distribution reconstructed by the modal coefficient set, so that the reconstructed temperature distribution approaches the target uniform temperature field.

[0047] The beneficial effects of this invention are:

[0048] (1) This invention divides the strip width direction into multiple continuous segments and estimates the heat transfer coefficient for each segment independently. Combined with ensemble Kalman filtering, the joint state vector is recursively updated, which can effectively suppress the physical drift of parameters caused by neglecting the two-dimensional heat transfer effect in the one-dimensional simplified model. Based on the estimated constant heat transfer coefficient of each segment and the reconstructed temperature distribution in the width direction, the output power of the heating device above each segment is adjusted, so that the transverse temperature distribution can be finely controlled, significantly improving the temperature uniformity in the width direction of the strip, avoiding local overheating or uneven hardness, and improving product quality.

[0049] (2) This invention uses a combination of radial basis interpolation and Gaussian integration to project discrete temperature measurements onto pre-stored two-dimensional heat transfer spatial modes. A sparse set of modal coefficients is obtained through soft thresholding, achieving a low-dimensional representation of the temperature field. Simultaneously, the random increment in the ensemble Kalman filter is dynamically adjusted using adaptive perturbation amplitude and global scaling coefficient, enabling the generation of the prediction vector to reflect changes in parameter uncertainty in real time. This method, without relying on an accurate mechanistic model, exhibits good adaptability to sensor noise, time delay, and unmodeled dynamics, ensuring the stability and physical rationality of the heat transfer coefficient estimation and enhancing the robustness of the control system. Attached Figure Description

[0050] The invention will now be further described with reference to the accompanying drawings.

[0051] Figure 1 This is a flowchart of the method of the present invention;

[0052] Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0053] 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.

[0054] Please see Figure 1 As shown, this invention is an industrial process control method based on a smart factory, comprising the following steps:

[0055] Step 1: Collect temperature measurements at multiple discrete locations along the width of the strip, and simultaneously acquire the strip running speed and thickness parameters to form a process observation set for the current moment;

[0056] Step 2: Project the process observation set onto the pre-stored two-dimensional heat transfer spatial modes, and obtain the modal coefficient set of each spatial mode by minimizing the projection error, thus completing the reduced-order characterization of the temperature field.

[0057] Step 3: Divide the strip width direction into multiple continuous segments, assign a segment constant heat transfer coefficient to be estimated to each segment, and concatenate all segment constant heat transfer coefficients with the modal coefficient set to form a joint state vector;

[0058] Step 4: Use ensemble Kalman filtering to recursively update the joint state vector, where each ensemble member independently performs prediction and observation correction, and extracts the average value of all members as the estimated value of the constant heat transfer coefficient of each segment at the current time.

[0059] Step 5: Based on the estimated constant heat transfer coefficients of each segment and the width-direction temperature distribution reconstructed from the modal coefficient set, adjust the output power of the heating device above each segment to make the reconstructed temperature distribution approximate the target uniform temperature field.

[0060] In step one, temperature measurements are collected at multiple discrete locations along the width of the strip, and the strip's running speed and thickness parameters are also acquired to form a process observation set for the current moment, specifically including:

[0061] Multiple infrared temperature probes are installed at equal intervals along the width direction perpendicular to the strip's running direction at the outlet side of the continuous annealing furnace or at the outlet of the soaking section. Each probe is aimed at a discrete position on the strip surface, and the sampling frequency is 2 Hz to acquire the real-time strip surface temperature value at the corresponding point. Simultaneously, a laser velocimeter and an X-ray thickness gauge are configured before the annealing furnace inlet to acquire the strip's running speed and thickness values, respectively, using the same time reference as the temperature sampling. All the above sensors are hardwired to the analog input channel of the programmable logic controller (PLC). The PLC arranges the temperature, speed, and thickness values ​​acquired at each discrete position at the same time into a multi-dimensional array in a predetermined order. This array represents the process observation set at the current moment.

[0062] In step two, the process observation set is projected onto the pre-stored two-dimensional heat transfer spatial modes. By minimizing the projection error, the modal coefficient set corresponding to each spatial mode is obtained, thus completing the reduced-order characterization of the temperature field. Specifically, this includes:

[0063] Construction of the Radial Basis Interpolation Matrix: In step two, a radial basis interpolation matrix is ​​constructed based on the temperature measurements and their corresponding width coordinates in the process observation set. Specifically, the width coordinates of all discrete locations in the process observation set are taken as interpolation base points, totaling 10 base points, each corresponding to a temperature measurement value. A thin-plate spline function is selected as the radial basis function, which is the square of the distance multiplied by the natural logarithm of the distance. For the width distance between any two base points, the square of the distance is calculated and then multiplied by the natural logarithm of the distance to obtain the radial basis function value. The radial basis function values ​​obtained by combining all base points pairwise are arranged into a 10x10 square matrix. A column vector of all 1s is added to the right side of this matrix, and a row vector of all 1s is added to the bottom. A zero is added to the lower right corner to form the augmented radial basis interpolation matrix.

[0064] To obtain the continuous temperature distribution function by solving for the interpolation coefficients: After obtaining the radial basis interpolation matrix, the 10 temperature measurements in the process observation set are arranged into a 10x1 vector, and a 0 is added to the bottom of this vector to form the augmented right-hand vector. By solving the linear equation system, i.e., multiplying the augmented right-hand vector by the inverse of the augmented radial basis interpolation matrix, a set of interpolation coefficients is obtained. This set of coefficients contains 10 radial basis coefficients, 1 linear coefficient, and 1 constant term. Based on these coefficients, for any position in the width direction, the radial basis function values ​​between that position and each base point are first calculated. Then, each radial basis function value is multiplied by its corresponding radial basis coefficient and summed. Finally, the linear coefficient is added, multiplied by the position coordinates, and the constant term is added. The result is the temperature distribution function value at that position. This temperature distribution function is continuous and first-order differentiable along the width direction.

[0065] Determining Gaussian Integral Nodes and Weights: After obtaining the continuously differentiable temperature distribution function, it is necessary to integrate this function with pre-stored spatial modes of various orders. There are five pre-stored spatial modes, each stored discretely at 100 equally spaced points along the width direction. The number of Gaussian integral nodes is determined based on the highest order of 5, and six Gaussian integral nodes are selected. Within the width interval, i.e., from coordinate 0 on one edge of the strip to coordinate 1 on the other edge, the positions of the six integral nodes in the interval from -1 to 1 are obtained by looking up the standard Legendre polynomial root table. Then, a linear transformation is used to map them to the width interval from 0 to 1, and the integral weight corresponding to each node is calculated.

[0066] Calculate the temperature distribution function and nodal values ​​for each mode: At each Gaussian integral node, perform two calculations simultaneously: First, substitute the node position coordinates into the temperature distribution function expression determined by the interpolation coefficients to obtain the temperature function value for that node; Second, for pre-stored spatial modes from order 1 to order 5, read the preset values ​​for that mode at the node position (since the modes are stored as discrete points, cubic spline interpolation is used to obtain the mode values ​​at the node positions). Thus, at each node, one temperature function value and five mode values ​​are obtained.

[0067] The weighted summation yields the original coefficients for each order: For the first-order spatial mode, traverse all six Gaussian integration nodes. At each node, multiply the temperature function value by the mode value of that order, and then multiply by the integration weight corresponding to that node to obtain the weighted product at that node. Sum the weighted products at the six nodes; the result is the original coefficient of the first-order spatial mode. The same process is used to calculate the original coefficients for the second, third, fourth, and fifth-order spatial modes. Each original coefficient is a real value representing the projection amplitude of the temperature distribution function onto that mode.

[0068] The soft-thresholding operation yields the final modal coefficient set: After obtaining all five original coefficients, a soft-thresholding operation is performed. First, a preset energy threshold is set, which is the sum of the squares of all original coefficients multiplied by 0.01. Then, each original coefficient is evaluated: if the absolute value of the original coefficient is less than the preset energy threshold, the coefficient is set to zero; if it is equal to or greater than the threshold, its original value is retained. The five coefficients after the above processing are arranged in order of order to form the final modal coefficient set. The modes corresponding to the non-zero coefficients in this set constitute the main reduced-order representation of the temperature field, completing the reduced-order representation of the temperature field.

[0069] It should be noted that the two-dimensional heat transfer spatial modes are obtained in advance in the following way: a two-dimensional heat transfer finite element model in the width direction of the strip is established, typical heating boundary conditions are applied, and snapshots of the temperature distribution in the width direction under different working conditions under steady state are collected; intrinsic orthogonal decomposition is performed on all snapshots, and the eigenvectors with the first 5 energy proportions exceeding 99.9% are taken as spatial modes and stored in the controller in the form of 200 equidistant scattered points in the width direction.

[0070] In step three, the strip width direction is divided into multiple continuous segments, each segment is assigned a segmental constant heat transfer coefficient to be estimated, and all segmental constant heat transfer coefficients are concatenated with the modal coefficient set to form a joint state vector, specifically including:

[0071] Determining the interpolation nodes and the acquisition method of preset values ​​for each spatial mode: In step three, the temperature distribution curve in the width direction is first reconstructed based on the obtained set of modal coefficients. In the width direction, i.e., within the interval from coordinate 0 on one side of the strip edge to coordinate 1 on the other side edge, 200 interpolation nodes are selected at equal intervals, with a spacing of 0.005 between adjacent nodes. There are five pre-stored spatial modes. Each mode has corresponding preset values ​​stored at the same 200 interpolation node positions. These preset values ​​are obtained through offline two-dimensional finite element simulation and intrinsic orthogonal decomposition. Five values ​​are stored at each node, corresponding to the amplitudes of modes 1 through 5 at that node.

[0072] Calculate the temperature reconstruction value for each interpolation node: For each interpolation node, perform the following calculations: Multiply the preset value of the first-order spatial mode at that node by the first coefficient in the modal coefficient set to obtain the first-order contribution value; multiply the preset value of the second-order spatial mode at that node by the second coefficient in the modal coefficient set to obtain the second-order contribution value; and so on, until the preset value of the fifth-order spatial mode at that node is multiplied by the fifth coefficient in the modal coefficient set to obtain the fifth-order contribution value. Then, add these five contribution values ​​together; the sum is the temperature reconstruction value at that interpolation node. Repeat the above process to calculate the temperature reconstruction values ​​for all 200 interpolation nodes.

[0073] To form a continuous temperature distribution curve: Arrange the 200 calculated temperature reconstruction values ​​in ascending order of their width coordinates. Using cubic spline interpolation, construct a cubic spline curve passing through all the model points, with the width coordinate of each interpolation node and the corresponding temperature reconstruction value as model points. This curve is a cubic polynomial between adjacent nodes, possessing continuous first and second derivatives at the nodes, thus forming a continuously differentiable temperature distribution curve along the width direction.

[0074] Calculate the curvature variation sequence along the width and set a threshold: Take the second derivative of the continuous temperature distribution curve to obtain the curvature variation sequence along the width direction. Specifically, at the midpoint of every two adjacent interpolation nodes, calculate the second derivative value of the temperature distribution curve; this value is the approximate curvature value at that midpoint, resulting in a total of 199 curvature values. Then, calculate the absolute values ​​of these 199 curvature values ​​and multiply the median of all absolute values ​​by 3 as the preset threshold.

[0075] Divide the strip into continuous segments along its width: Starting from one edge of the strip, scan the absolute value sequence of curvature point by point along the width direction. Whenever the absolute value of curvature exceeds a preset threshold, record that position as a boundary point. After scanning, several boundary points are obtained. Based on these boundary points, the width interval is divided into multiple continuous segments. Each segment does not contain any boundary points, and the segment length is determined by the distance between adjacent boundary points; therefore, the lengths of the segments may not be equal.

[0076] Constructing the joint state vector: Within each consecutive segment, an independent constant heat transfer coefficient to be estimated is defined. The initial value of this coefficient is the average value of the equivalent heat transfer coefficients obtained from offline simulation at each interpolation node within that segment. The constant heat transfer coefficients corresponding to all consecutive segments are arranged into a sub-vector in the width direction from left to right. This sub-vector is then sequentially concatenated with the modal coefficient set (containing 5 coefficients) obtained in step two. That is, all segmental constant heat transfer coefficients are arranged first, followed by the 5 coefficients from the modal coefficient set, forming a complete joint state vector. This joint state vector is the basic object for subsequent recursive updates.

[0077] In step four, the joint state vector is recursively updated using ensemble Kalman filtering, where each ensemble member independently performs prediction and observation correction. The average value of all members is extracted as the estimated value of the constant heat transfer coefficient for each segment at the current time. Specifically, this includes:

[0078] The overall process of recursive update is as follows: In step four, the joint state vector is recursively updated using ensemble Kalman filtering. The ensemble has 100 members, each holding a copy of the joint state vector. The update process begins with the posterior estimate from the previous time step. First, a prediction step is performed to generate a prediction vector for each member. Then, the process observation set acquired at the current time step is used to correct each prediction vector. Finally, all corrected members are arithmetically averaged using the same piecewise constant heat transfer coefficient to obtain the estimated value of each piecewise constant heat transfer coefficient at the current time step.

[0079] Calculate the adaptive disturbance amplitude: For the first For each piecewise constant heat transfer coefficient, the estimated value corresponding to that piecewise constant is extracted from the joint state vector of all 100 members at the previous time step. This set of estimated values ​​is denoted as a numerical sequence. First, the arithmetic mean of the sequence is calculated. Then, the square of the difference between each estimated value and the mean is calculated sequentially. The sum of all squares is divided by the number of members minus one (i.e., 99) to obtain the variance of the sequence. Taking the square root of the variance yields the standard deviation, denoted as . Preset a constant coefficient. The value is 2.0. Then the... Adaptive perturbation amplitude of each segment Calculated using the following formula: ;

[0080] in, For the first The standard deviation of the piecewise constant heat transfer coefficients at the previous time step among all member estimates The default constant is 2.0.

[0081] Calculate the global scaling factor: Extract the complete set of joint state vectors from the joint state vectors of all 100 members at the previous time step. The dimension of the joint state vector is denoted as D, where D equals the number of segments plus the number of modal coefficients (in this embodiment, the number of segments is 8 and the number of modal coefficients is 5, so D=13). First, calculate the arithmetic mean vector of the joint state vectors of all members. Then, for each member, calculate the difference vector between its state vector and the mean vector. Squaring each component of this difference vector and summing it over all members, then dividing by the number of members minus one, yields the covariance matrix. The covariance matrix is ​​a D x D square matrix. The trace of the covariance matrix is ​​obtained by summing the D elements on its diagonal (i.e., the variances of each component), denoted as [equation missing]. Preset a scaling factor. The value is 0.05. Therefore, the global scaling factor is... Calculated using the following formula: ;

[0082] in, Let the trace be the covariance matrix of the joint state vector of all members at the previous time step. The preset scaling factor is 0.05.

[0083] Obtain the corrected disturbance amplitude: calculate the adaptive disturbance amplitude for each segment. Multiplying each value by the global scaling factor γ yields the corrected perturbation amplitude for that segment, denoted as... .Right now The correction perturbation amplitude is used to control the magnitude of the random increment, so that it can reflect the variability of the parameters themselves and adaptively scale according to the uncertainty of the overall state.

[0084] Generate a random increment vector for each set member: For each set member (100 in total), generate a random number independently for each segment. The random number follows a normal distribution with a mean of zero and a variance equal to the magnitude of the correction perturbation for that segment. The square of . Specifically, it is generated by calling the standard normal distribution random number generation function to obtain a standard random number, and then multiplying it by . The random increments for each segment are obtained. The random increments of segments 1 through 8 are arranged in segment order into an 8-dimensional random sub-vector. Meanwhile, for the five coefficients in the modal coefficient set, since their variation is indirectly reflected by the update of the heat transfer coefficient, no random perturbation is applied (i.e., the random increments of the modal coefficients are all zero). The segmented random sub-vectors are concatenated with the zero vector of the modal coefficients to form a complete D-dimensional random increment vector, where D=13.

[0085] The prediction vector is obtained by superposition: For each member of the set, its joint state vector from the previous time step is added element-wise to the complete random increment vector to obtain the prediction vector for that member. Specifically, the value of each component in the prediction vector is equal to the value of the corresponding component from the previous time step plus the corresponding random increment value. Since the random increment corresponding to the modal coefficient is zero, the predicted value of the modal coefficient remains the same as the previous time step. The same operation is performed on all 100 members to obtain 100 prediction vectors.

[0086] Correction and Estimation Extraction: Each prediction vector is corrected using the process observation set acquired at the current moment. The correction process follows the standard update formula of ensemble Kalman filtering: First, the predicted temperature value corresponding to each prediction vector is calculated (obtained by combining the modal coefficients and spatial modes in the prediction vector, and then combining them with the piecewise heat transfer coefficient for heat transfer calculation). Then, the difference between the observed value and the predicted value is calculated, and multiplied by the Kalman gain (calculated from the prediction error covariance and the observation noise covariance). The product is added to the prediction vector to obtain the corrected analysis vector. After completing the correction for all 100 members, for each piecewise constant heat transfer coefficient, the values ​​of the corresponding components in the corrected analysis vectors of these 100 members are added together and then divided by 100 to obtain the estimated value of that piecewise constant heat transfer coefficient. This estimated value is the output at the current moment and is used for power adjustment in step five.

[0087] In step five, based on the estimated constant heat transfer coefficients of each segment and the width-direction temperature distribution reconstructed from the modal coefficient set, the output power of the heating device above each segment is adjusted to make the reconstructed temperature distribution approximate the target uniform temperature field. Specifically, this includes:

[0088] Calculating the segmented temperature difference: In step five, based on the estimated constant heat transfer coefficients of each segment obtained in step four and the reconstructed width-direction temperature distribution in step three, power adjustment calculations are performed for each segment. For the i-th segment, the width interval corresponding to the segment is taken, and the temperature reconstruction values ​​at all interpolation nodes within the interval are extracted from the reconstructed continuous temperature distribution curve. The arithmetic mean of these temperature reconstruction values ​​is calculated as the average reconstructed temperature of the segment. Simultaneously, a target uniform temperature field is pre-set, which is a constant target temperature value along the entire strip width direction, for example, 850 degrees Celsius. The difference between the average reconstructed temperature of the segment and this constant target temperature value is the segmented temperature difference. If the difference is positive, it indicates that the current temperature of the segment is higher than the target value, and the heating power needs to be reduced; if it is negative, the heating power needs to be increased.

[0089] Calculate the initial power correction: For each segment, a proportionality coefficient is pre-calibrated through offline experiments. This proportionality coefficient is related to the heat transfer characteristics of that segment. The calibration method is as follows: Under steady-state conditions, the output power of the heating device above the segment is changed by a known amount, the change in the strip temperature of that segment is measured, and the ratio of the power change to the temperature change is taken as the proportionality coefficient for that segment. In actual operation, based on the estimated constant heat transfer coefficient of that segment obtained in step four, the offline-calibrated proportionality coefficient is corrected in real time: the estimated constant heat transfer coefficient of that segment is divided by the offline-calibrated standard heat transfer coefficient to obtain a correction factor. Then, the offline-calibrated proportionality coefficient is multiplied by this correction factor to obtain the proportionality coefficient applicable at the current moment. The segment temperature difference is multiplied by this proportionality coefficient to obtain the initial power correction for that segment. A positive initial power correction indicates that power needs to be increased, and a negative one indicates that power needs to be reduced.

[0090] The final power adjustment value is obtained by weighted averaging: To eliminate temperature oscillations caused by mutual interference between power adjustments in adjacent segments, the initial power correction amounts of adjacent segments are weighted and averaged. Specifically, for the i-th segment, the initial power correction amounts of three segments—the (i-1)-th, i-th, and (i+1)-th segments—are taken (for segments located at the edge of the strip, only the two adjacent segments are taken). Weighting coefficients are set, where the weighting coefficient of the i-th segment itself is 0.6, and the weighting coefficients of the two adjacent segments are each 0.2. If the ratio of the estimated heat transfer coefficient between the i-th segment and its adjacent segments deviates significantly from 1, the weighting coefficients are fine-tuned: the estimated constant heat transfer coefficient of the i-th segment is divided by the estimated constant heat transfer coefficient of the adjacent segments, resulting in two ratios. Each ratio is multiplied by 0.2 and used as the weighting coefficient for the corresponding adjacent segment. At the same time, the weighting coefficient of the i-th segment is adjusted to 1 minus the sum of the two adjacent weighting coefficients. Multiply the initial power correction for each segment by its corresponding weighting coefficient and sum them to obtain the final power adjustment value for that segment. Add this adjustment value to the current output power of the heating device above that segment to make the reconstructed temperature distribution approximate the target uniform temperature field.

[0091] Please see Figure 2 As shown, an industrial process control system based on a smart factory includes:

[0092] The process observation and acquisition module is used to acquire temperature measurements at multiple discrete locations along the width of the strip, and simultaneously obtain the strip running speed and thickness parameters to form a process observation set at the current moment.

[0093] The temperature field order reduction characterization module projects the process observation set onto the pre-stored two-dimensional heat transfer spatial modes, and obtains the modal coefficient set of each spatial mode by minimizing the projection error, thus completing the order reduction characterization of the temperature field.

[0094] The joint state vector construction module divides the strip width direction into multiple continuous segments, assigns a segment constant heat transfer coefficient to be estimated to each segment, and concatenates all segment constant heat transfer coefficients with the modal coefficient set to form a joint state vector.

[0095] The segmented heat transfer coefficient estimation module uses ensemble Kalman filtering to recursively update the joint state vector. Each ensemble member independently performs prediction and observation correction, and the average value of all members is extracted as the estimated value of the constant heat transfer coefficient of each segment at the current time.

[0096] The heating power adaptive adjustment module adjusts the output power of the heating device above each segment according to the estimated value of the constant heat transfer coefficient of each segment and the width direction temperature distribution reconstructed by the modal coefficient set, so that the reconstructed temperature distribution approaches the target uniform temperature field.

[0097] The working principle of this invention is as follows: First, temperature measurements are collected at multiple discrete locations along the width of the strip, while simultaneously acquiring the strip's running speed and thickness parameters, forming a process observation set for the current moment. Then, this process observation set is projected onto a pre-stored two-dimensional heat transfer spatial mode. A set of modal coefficients is obtained through radial basis interpolation, Gaussian integration, and soft thresholding, completing the reduced-order characterization of the temperature field. Next, the temperature distribution curve along the width is reconstructed based on the modal coefficient set. The strip width is divided into multiple continuous segments according to the curvature change, and each segment is assigned a constant heat transfer coefficient to be estimated. These segments are then concatenated with the modal coefficient set to form a joint state vector. Finally, an ensemble Kalman filter is used... The filter recursively updates the joint state vector. By calculating the adaptive perturbation amplitude and the global scaling factor, a random increment is generated for each set member and superimposed to obtain the prediction vector. After correction, the average value of all members is extracted as the estimated value of the constant heat transfer coefficient of each segment. Finally, based on the estimated value of the constant heat transfer coefficient of each segment and the reconstructed width-direction temperature distribution, the temperature difference between the reconstructed average temperature of each segment and the target temperature value is calculated. Multiplying this difference by the proportional coefficient corrected by the heat transfer coefficient, the initial power correction amount is obtained. Then, the correction amounts of adjacent segments are weighted and averaged to obtain the final power adjustment value of the heating device above each segment, so that the reconstructed temperature distribution approximates the target uniform temperature field.

[0098] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. An industrial process control method based on smart factories, characterized in that, Includes the following steps: Step 1: Collect temperature measurements at multiple discrete locations along the width of the strip, and simultaneously acquire the strip running speed and thickness parameters to form a process observation set for the current moment; Step 2: Project the process observation set onto the pre-stored two-dimensional heat transfer spatial modes, and obtain the modal coefficient set of each spatial mode by minimizing the projection error, thus completing the reduced-order characterization of the temperature field. Step 3: Divide the strip width direction into multiple continuous segments, assign a segment constant heat transfer coefficient to be estimated to each segment, and concatenate all segment constant heat transfer coefficients with the modal coefficient set to form a joint state vector; Step 4: Use ensemble Kalman filtering to recursively update the joint state vector, where each ensemble member independently performs prediction and observation correction, and extracts the average value of all members as the estimated value of the constant heat transfer coefficient of each segment at the current time. Step 5: Based on the estimated constant heat transfer coefficients of each segment and the width-direction temperature distribution reconstructed from the modal coefficient set, adjust the output power of the heating device above each segment to make the reconstructed temperature distribution approximate the target uniform temperature field.

2. The industrial process control method based on a smart factory according to claim 1, characterized in that, The reduced-order characterization of the temperature field specifically includes: Based on the temperature measurements and their corresponding width-direction coordinates in the process observation set, a radial basis interpolation matrix is ​​constructed to fit the discrete temperature measurements into a continuously differentiable temperature distribution function along the width direction. The temperature distribution function is integrated with each pre-stored spatial mode over a width interval to obtain the original coefficients of each mode. A soft thresholding operation is applied to the original coefficients, setting coefficients whose absolute values ​​are less than a preset energy threshold to zero, and retaining the remaining coefficients as the final set of modal coefficients.

3. The industrial process control method based on a smart factory according to claim 2, characterized in that, Obtaining the original coefficients corresponding to each mode specifically includes: The number of Gaussian integration nodes is determined based on the highest order of each spatial mode to be projected, and the corresponding integration node positions and integration weights are generated within the width interval. At each integration node, the function value of the temperature distribution function and the modal values ​​of each spatial mode are calculated simultaneously. Multiply the product of the function value and the modal value at the same node by the corresponding integral weight, and sum the weighted multiplications at all nodes to obtain the original coefficients of each spatial mode.

4. The industrial process control method based on a smart factory according to claim 1, characterized in that, The process of constructing the joint state vector is as follows: The temperature distribution curve along the width direction is reconstructed based on the obtained set of modal coefficients, and the second derivative of the temperature distribution curve is used to obtain the curvature variation sequence along the width. All locations where the absolute value of curvature exceeds a preset threshold are selected as dividing points, and the width direction is divided into multiple continuous segments of unequal length based on these dividing points; The constant heat transfer coefficient to be estimated is set independently in each segment, and all segment constant heat transfer coefficients are arranged sequentially with the original set of modal coefficients to form a joint state vector.

5. The industrial process control method based on a smart factory according to claim 4, characterized in that, The reconstructing of the temperature distribution curve along the width direction based on the obtained set of modal coefficients specifically includes: Several interpolation nodes are selected at equal intervals along the width direction. At each node, the preset value of each order spatial mode is multiplied by the coefficient of the corresponding order in the modal coefficient set to obtain the contribution value of each order. The temperature reconstruction value of the interpolation node is obtained by summing all the order contribution values ​​at the same interpolation node. The temperature reconstruction values ​​of all interpolation nodes are connected sequentially in the width direction to form a continuous temperature distribution curve.

6. The industrial process control method based on a smart factory according to claim 1, characterized in that, The execution of prediction and observation correction specifically includes: Based on the degree of dispersion of the estimated constant heat transfer coefficient of the same segment among all members at the previous moment, the adaptive disturbance amplitude of the corresponding segment is calculated. For each set member, generate a random increment centered at zero with an adaptive perturbation amplitude as the variance, and superimpose the increment onto the joint state vector of the previous time step to obtain the prediction vector; Each prediction vector is corrected using the process observation set at the current moment, and all corrected members are arithmetically averaged according to the same piecewise constant heat transfer coefficient to obtain the estimated value of the corresponding piecewise constant heat transfer coefficient.

7. The industrial process control method based on a smart factory according to claim 6, characterized in that, The process of obtaining the prediction vector is as follows: Extract the estimated value of the piecewise constant heat transfer coefficient from the joint state vector of all members at the previous moment, calculate the trace of its covariance matrix, and multiply the trace by a preset scaling factor to obtain the global scaling factor. Multiply the adaptive perturbation amplitude of each segment by the global scaling factor to obtain the corrected perturbation amplitude of the corresponding segment. For each member of the set, a random number is independently sampled according to the correction perturbation amplitude of each segment, and the random arrays of all segments are combined into a complete random increment vector.

8. The industrial process control method based on a smart factory according to claim 1, characterized in that, The adjustment of the output power of the heating device above each segment specifically includes: Calculate the difference between the average value of the reconstructed temperature distribution within each segment and the target value of the target uniform temperature field within the corresponding segment to obtain the segmented temperature difference; Multiply the segmented temperature difference by a pre-calibrated proportional coefficient based on the estimated segmented constant heat transfer coefficient to obtain the preliminary power correction for the corresponding segment. The initial power corrections for adjacent segments are weighted and averaged, with the weighting coefficient determined by the ratio of the estimated constant heat transfer coefficients of the two segments, and the final power adjustment value for each segment is output.

9. An industrial process control system based on a smart factory, characterized in that, An industrial process control method based on a smart factory as described in any one of claims 1-8, comprising: The process observation and acquisition module is used to acquire temperature measurements at multiple discrete locations along the width of the strip, and simultaneously obtain the strip running speed and thickness parameters to form a process observation set at the current moment. The temperature field order reduction characterization module projects the process observation set onto the pre-stored two-dimensional heat transfer spatial modes, and obtains the modal coefficient set of each spatial mode by minimizing the projection error, thus completing the order reduction characterization of the temperature field. The joint state vector construction module divides the strip width direction into multiple continuous segments, assigns a segment constant heat transfer coefficient to be estimated to each segment, and concatenates all segment constant heat transfer coefficients with the modal coefficient set to form a joint state vector. The segmented heat transfer coefficient estimation module uses ensemble Kalman filtering to recursively update the joint state vector. Each ensemble member independently performs prediction and observation correction, and the average value of all members is extracted as the estimated value of the constant heat transfer coefficient of each segment at the current time. The heating power adaptive adjustment module adjusts the output power of the heating device above each segment according to the estimated value of the constant heat transfer coefficient of each segment and the width direction temperature distribution reconstructed by the modal coefficient set, so that the reconstructed temperature distribution approaches the target uniform temperature field.

Images