Method and system for closed-loop control of a corrugated board apparatus

Abstract

The invention relates to a method for closed-loop control of a corrugated board apparatus comprising: – recording a corrugated board curvature and at least one actual value of at least one manipulated variable of the corrugated board apparatus, – determining at least one correction value relating to the at least one manipulated variable for reducing the corrugated board curvature on the basis of the recorded corrugated board curvature and the at least one actual value, wherein the correction value is determined using a machine model created by means of a learning method, and – controlling the corrugated board apparatus according to the at least one correction value. The proposed solution also comprises a computer-implemented method, a system, a computer program product and a corrugated board apparatus.

Description

[0001] Method and system for controlling a corrugated board plant

[0002] Description

[0003] The proposed solution concerns a method for controlling a corrugated board plant and a corresponding system.

[0004] A control system generally refers to a technical process in which a specific process is continuously monitored and adjusted to achieve and maintain a desired target value. In corrugated board manufacturing, control systems are fundamentally known to operate based on various controlled variables. Most industrial applications utilize model-free controllers whose output signal includes a proportional component (P component), an integral component (I component), and / or a differential component (D component). The P component is proportional to the deviation of the controlled variable from a target value. The I component is proportional to the integral of the deviation over time. The D component is proportional to the differential of the deviation over time. Depending on the components considered, these are also referred to as P / Pl / PD / PlD controls.The coefficients of the different proportions are usually optimized on a physical or a data-driven system model formed by classical statistical methods (linear / polynomial regression).

[0005] However, despite regulation of the corrugated board machine, large-area warping (curvature) regularly occurs in the corrugated board during operation.

[0006] Therefore, there is a need to improve the processes and systems for controlling a corrugated board plant.

[0007] This problem is solved by a method according to claim 1 and a system according to claim 12.

[0008] Accordingly, the proposed procedure for controlling a corrugated board machine includes at least: detecting the curvature of the corrugated board and at least one actual value of at least one control variable of the corrugated board machine, determining at least one correction value relating to the at least one control variable to reduce the curvature of the corrugated board, based on the detected curvature of the corrugated board and the at least one actual value and using a machine model created with a learning method, and controlling the corrugated board machine according to the at least one correction value.

[0009] The machine model allows for a precise determination of how a change to at least one control variable of the corrugated board machine affects the corrugated board produced. Therefore, based on the machine model, it is possible to determine by what value the at least one control variable must be adjusted relative to its actual value to reduce the corrugated board curvature. This value is the correction value of the at least one control variable.

[0010] Creating a machine model by mapping physical relationships would be extremely complex for sophisticated dynamic systems and often only feasible with substantial investments in sensors. In most cases, the entirety of influencing factors is unknown. In contrast, the proposed method uses a machine learning algorithm to create a machine model, thus enabling the modeling of complex systems without requiring knowledge of all physical relationships. In particular, the proposed solution improves the modeling of nonlinear dynamic systems such as a corrugated board plant. This can enhance the control of the corrugated board plant with regard to reducing curvature, without necessitating the identification of all physical relationships influencing factors on the curvature.The proposed method can also be used to control a corrugated board plant during the production of various corrugated board products, thus improving scalability.

[0011] For example, the machine model can map at least one input value to at least one predicted output value. This can enable a time- and computationally efficient mapping of a manipulated variable to a property of the corrugated board produced.

[0012] For example, the at least one input value can represent the at least one actual value of the at least one manipulated variable, and the at least one output value can represent a property of the corrugated board. Thus, the machine model can be configured to determine a property that the corrugated board produced at the recorded actual value possesses. This allows a property of the corrugated board to be determined even before it is measured. This can be particularly advantageous if the property cannot be easily determined or can only be determined at a later stage in production.

[0013] In particular, the machine model can reduce the time it takes to determine a property of the corrugated board when the control variable is changed, compared to measuring the properties.

[0014] In principle, the machine model can be retrained using the metrologically recorded properties, such as the curvature of corrugated cardboard.

[0015] The machine model can be trained to map at least one input value to a property of the corrugated board that directly or indirectly determines the curvature of the board. For example, this property could be the distance of a corrugated board surface relative to a reference point or the radius of curvature of a corrugated board surface.

[0016] According to one embodiment of the proposed method, the machine model can map the at least one input value to a predicted corrugated board curvature. The output value of the machine model can therefore directly represent the corrugated board curvature. Deriving the corrugated board curvature from the output value is thus unnecessary. This can simplify the determination of the correction value.

[0017] According to a further embodiment of the proposed method, the at least one input value can include an actual value of the corrugated board curvature. Thus, within the framework of the proposed method, the measured corrugated board curvature can be used to determine the correction value in addition to the at least one actual value of the at least one manipulated variable. For example, a corrugated board curvature determined using the machine model can be compared with the actual value of the corrugated board curvature to verify the machine model. This can increase the reliability of the corrugated board curvature determined for the actual value of the at least one target value. According to a further embodiment of the proposed method, the at least one input value can include a target value of the corrugated board curvature. Therefore, determining the correction value can depend on a deviation of the measured corrugated board curvature from the target value.This can prevent unnecessary calculations to determine the correction value and unnecessary control processes for tolerable corrugated board curvatures.

[0018] For example, determining the correction value may depend on the fact that the target value of the corrugated board curvature deviates from the measured corrugated board curvature by more than a predetermined maximum value.

[0019] According to a further embodiment of the proposed method, the machine model can map the at least one actual value of the at least one manipulated variable, the actual value of the corrugated board curvature, and the target value of the corrugated board curvature to the at least one correction value of the at least one manipulated variable. Thus, the machine model can be trained to directly output the correction value for the at least one manipulated variable. Calculating the correction value by varying the value of the at least one manipulated variable can therefore be unnecessary. This can reduce the computational effort and time required to determine the correction value.

[0020] According to a further embodiment of the proposed method, the machine model can be based on product metadata. The machine model can thus be trained to take into account different properties of the raw materials, the manufacturing process, and the finished corrugated board when determining the correction value for at least one control variable.

[0021] As an example, the machine model can be trained using training data that includes a large number of datasets differentiated according to product metadata.

[0022] According to one embodiment of the proposed method, the product metadata may include the basis weight of the paper and / or the total basis weight of the entire corrugated board and / or paper manufacturer data and / or information regarding a paper grade and / or the number of plies in the corrugated board and / or flute height and / or flute pitch and / or the number of sheets and / or the sheet dimensions. Likewise, the product metadata may include other parameters specific to a paper / board product. According to a further embodiment of the proposed method, the learning process may include supervised learning, self-supervised learning, reinforcement learning, or unsupervised learning.

[0023] In supervised learning, a learning algorithm is trained and validated using datasets that contain a corresponding output value for each input. Such datasets are called labeled datasets. An example would be a dataset with at least one actual value of at least one manipulated variable, to which a corrugated board curvature has been assigned. Another example would be a dataset consisting of an actual value of the corrugated board curvature, a target value of the corrugated board curvature, at least one actual value of at least one manipulated variable, and an assigned correction value.

[0024] The learning process can include various methods and algorithms that are suitable for learning from a data set and can be applied to new, unknown situations in order to, for example, recognize patterns or solve specific tasks within these situations.

[0025] In principle, a data set for training the machine model can be based on historical data.

[0026] According to an exemplary embodiment of the proposed method, the learning process can include ensemble learning or the training of an artificial neural network.

[0027] Ensemble methods are used in statistics and machine learning. They employ a finite set of different learning algorithms to achieve better results than a single algorithm could. While computing the results of this set of algorithms may take longer than evaluating a single algorithm, a nearly equivalent result can be achieved with significantly less computational depth.

[0028] A neural network comprises at least one input layer and one output layer, each containing at least one value. The values ​​of the different layers are interconnected via so-called artificial neurons. Based on the at least one value of the input layer, the at least one neuron calculates an output value, which represents an input value of another layer. The calculation of the output value can involve multiplying the at least one input value by at least one weight and summing over all weighted input values—if there is more than one value. This can also be described as applying a transfer function. The calculation of the output value can also involve applying a threshold and / or an activation function, which can introduce nonlinear dependencies between the sum of the weighted input values ​​and the output value.During the training of the neural network, the weighting function and optionally the activation function can be optimized in such a way that a deviation of the output values ​​from the corresponding output values ​​of the training data set is minimized.

[0029] According to a further embodiment of the proposed procedure, creating the machine model can include initial training with historical plant data.

[0030] According to a further embodiment of the proposed procedure, the historical plant data can include recorded values ​​of at least one control variable and corrugated board curvature measured in production at the recorded values.

[0031] According to a further embodiment of the proposed method, the machine model can be retrained using values ​​of at least one control variable and the measured corrugated board curvature recorded during operation. This can enable continuous adaptation of the control system to changing production conditions, e.g., due to wear.

[0032] According to a further embodiment of the proposed method, the machine model can be created using datasets from multiple corrugated board machines. This can increase the number of available datasets when creating the machine model and thus improve the reliability of the determined correction values.

[0033] According to a further embodiment of the proposed method, the data sets from the corrugated board plants can be uploaded to cloud storage during operation. This can enable the exchange of data sets from different corrugated board plants. According to another embodiment of the proposed method, the training and / or retraining can include minimizing a cost function. For example, the cost function could represent the squared error of the predicted output value relative to a reference value from the training data.

[0034] According to a further embodiment of the proposed procedure, a gradient-based method, an evolutionary method, or a probabilistic method can be used to minimize the deviation.

[0035] According to a further embodiment of the proposed method, training and / or retraining can be performed on a remote computer. This can enable centralized training and / or retraining of the machine model, thus conserving computing resources at the corrugated board production site.

[0036] According to a further embodiment of the proposed method, the remote computer can use the data records in the cloud storage.

[0037] The proposed method does not fundamentally restrict the way in which the machine model created by a learning process can be integrated. At least three different integration options are conceivable.

[0038] In a first variant, according to one embodiment of the proposed method, the control of at least one manipulated variable can be optimized using the machine model. In this variant, the machine model is therefore not part of the control loop that regulates the at least one manipulated variable as a function of the controlled variable, corrugated board curvature. However, the machine model can be used to optimize the control such that the at least one correction value determined by the control system reduces the corrugated board curvature as effectively as possible. The correction value is thus determined indirectly using the machine model and based on the at least one actual value of the at least one manipulated variable and the measured corrugated board curvature.This approach can combine the ability to model complex, nonlinear system behavior in corrugated board production, and thus reduce the predictive deviation of corrugated board curvature, with the simplicity and robustness of model-free control. According to a further embodiment of the proposed method, a time course of the predicted output value can be simulated for at least one test configuration of the control system using the machine model for optimization purposes. This allows for the determination of how the corrugated board production line is controlled with respect to corrugated board curvature in at least one test configuration.

[0039] According to a further embodiment of the proposed method, the transient response of the corrugated board curvature can be simulated in the test configuration. The transient response describes the temporal evolution of the controlled variable, i.e., the corrugated board curvature, after an adjustment of at least one manipulated variable by the correction value.

[0040] As is generally known, the time course of the controlled variable with respect to a setpoint targeted by the control system can exhibit the following three characteristic forms: a damped oscillatory case, a creeping case, and an aperiodic limiting case. Each of these cases can be assigned a characteristic time as a settling time.

[0041] According to a further embodiment of the proposed method, an optimized control configuration that minimizes the settling time can be determined by varying the test configuration. The optimized configuration can thus be determined by simulating the settling behavior for different test configurations of the control system.

[0042] According to a further refinement of the proposed method, coefficients of a P, PI, PD, or PID control system can be adjusted during optimization. This enables the proposed method to be used for optimizing standardized and widely used control systems, particularly on corrugated board plants with an existing control system. This can reduce the costs of implementing the proposed method.

[0043] The configuration can be varied randomly or in a targeted manner, e.g. using a gradient method.

[0044] According to a further embodiment of the proposed method, the optimization, in particular the variation of the test configuration, can be implemented using evolutionary optimization or Bayesian model optimization. This can enable the calculation of a probability distribution for the optimized parameters. The reliability of the optimization can then be assessed using this probability distribution.

[0045] According to a further embodiment of the proposed procedure, the optimization can include transferring an optimized configuration to the control system. Accordingly, the optimization process can be carried out physically separate from the corrugated board plant.

[0046] According to a further embodiment of the proposed procedure, the control system can be designed with a plurality of control elements for each manipulated variable.

[0047] In a second variant, according to a further refinement of the proposed method, the curvature of corrugated boards can be predicted using the machine model within the framework of model predictive control. The machine model itself can thus be part of the control loop. The output value of the machine model can be the curvature of the corrugated boards. The curvature of the corrugated boards can therefore be indirectly determined by using the machine model. Based on the predicted curvature of the corrugated boards and a setpoint, at least one correction value can then be determined, for example, by an optimizer.

[0048] In general, model predictive control uses a discrete-time dynamic model of the process to be controlled to calculate the future behavior of the process as a function of the input signals. This allows the calculation of the optimal input values—in terms of a cost function—that lead to optimal output values. Input, output, and state constraints can be considered simultaneously. While the model behavior is predicted up to a certain time horizon, typically only the at least one input value is used for the next cycle, and the optimization is then repeated. Since the correction value can be determined by optimizing the at least one input value, i.e., the at least one manipulated variable, the proposed method according to the second variant can function without using the corrugated board curvature as a controlled variable.This eliminates the dead time between adjusting at least one control variable and detecting its effect on corrugated board curvature. This can reduce scrap and thus lower manufacturing costs.

[0049] Furthermore, the curvature of corrugated cardboard can be recorded for training or retraining of the machine model.

[0050] In principle, model predictive control can enable the mapping of multivariate manipulated variables to multivariate controlled variables while maintaining the possibility of mapping complex nonlinear system behavior and thus reducing the deviation in the prediction of corrugated board curvature.

[0051] According to a further embodiment of the proposed method, the predicted corrugated board curvature can include a temporal evolution of the curvature over a predetermined prediction horizon. This can reduce abrupt changes in at least one manipulated variable and thus improve the transient response of the controlled variable.

[0052] According to a further embodiment of the proposed method, at least one correction value of at least one control variable can be determined based on the predicted corrugated board curvature.

[0053] According to a further embodiment of the proposed procedure, a deviation between the predicted corrugated board curvature and the target value can be minimized to determine at least one correction value.

[0054] According to a further refinement of the proposed procedure, a gradient-based method, an evolutionary method, or a probabilistic method can be applied to minimize the deviation. This can improve convergence towards a theoretical optimum.

[0055] According to a further embodiment of the proposed procedure, the deviation can be determined for a plurality of time points within the prediction horizon.

[0056] According to a further refinement of the proposed method, the deviation can be integrated over the majority of time points, particularly the prediction horizon. Accordingly, the cost function can be the area under the curve of the predicted curvature and the curve of the target value.

[0057] In a third variant, according to a further refinement of the proposed method, the machine model can be used within the framework of model predictive control to determine at least one correction value for at least one manipulated variable. The machine model can thus map the at least one input value directly to the at least one correction value (reverse system dynamics). This can further simplify the determination of the correction value. An optimizer that determines the at least one correction value based on a predicted corrugated board curvature and a setpoint may become unnecessary.

[0058] According to a further embodiment of the proposed method, the at least one control variable can include a paper temperature and / or a steam temperature of a heated roller and / or a steam pressure and / or a wrap angle of a preheater roller and / or a web speed and / or a moisture content and / or a water flow of a spraying and / or a water distribution of a spraying and / or a quantity of a glue application and / or a distribution of a glue application and / or a temperature within a double facer.

[0059] There is no limit to the number of control variables and their associated correction values ​​that can be captured and controlled within the framework of the proposed procedure.

[0060] According to a further embodiment of the proposed procedure, at least one actual value can be recorded for a plurality of control variables.

[0061] According to a further embodiment of the proposed method, a plurality of actual values ​​can be recorded for at least one control variable. This can, for example, enable the recording of a spatial distribution.

[0062] According to a further embodiment of the proposed method, the paper temperature can encompass temperatures at a plurality of points distributed across the working width.

[0063] According to a further embodiment of the proposed method, the predicted and / or detected corrugated board curvature can be determined across the entire working width. In this further embodiment, the predicted and / or detected corrugated board curvature can be the curvature of precisely one corrugated board section, in particular one sheet or panel, from the plurality of corrugated board sections distributed across a working width. This can eliminate the need to detect a curvature across the entire working width and thus reduce the time required to determine a curvature.

[0064] According to a further embodiment of the proposed procedure, exactly one corrugated board section can be selected from the majority of corrugated board sections distributed over a working width based on a user choice and / or a predetermined characteristic.

[0065] For example, a user selection can specify that the curvature of a particular corrugated board section, especially a sheet or panel, should be used as the corrugated board curvature. The curvature of adjacent corrugated board sections then does not need to be recorded.

[0066] According to a further embodiment of the proposed method, the predetermined characteristic can be a position along the working width or a relation of the curvature of the corrugated board section to the curvature of all corrugated board sections distributed over the working width.

[0067] According to a further embodiment of the proposed method, the one corrugated board section, in particular the corrugated board sheet or the panel, can be the corrugated board section with the greatest curvature of the majority of corrugated board sections distributed over a working width.

[0068] According to a further embodiment of the proposed method, the curvature of the corrugated board can be detected with at least one sensor, in particular with at least one camera and / or at least one laser sensor.

[0069] According to a further embodiment of the proposed method, the at least one sensor can be arranged above the conveyor belt. According to a further embodiment of the proposed method, the at least one sensor can be arranged between a longitudinal cutting unit and a stacking tray on the corrugated board machine.

[0070] According to a further embodiment of the proposed method, the sensor-acquired measured values ​​can be fitted to a polynomial to calculate a curvature.

[0071] According to a further embodiment of the proposed method, the prediction of the at least one output value using the machine model can take into account a time difference between a correction of the at least one manipulated variable and its effect on the predicted output value. This can improve training / retraining, since in the training datasets the actual values ​​of the at least one manipulated variable and the measured corrugated board curvature are typically separated by a dead time.

[0072] According to a further embodiment of the proposed method, the time difference can include a dead time that the corrugated board needs to be transported from a location where a machine element acts on the corrugated board according to the control variable to a location of a detection device that records the output value.

[0073] According to a further embodiment of the proposed method, the dead time can be a multidimensional quantity for different machine elements and detection devices arranged along the conveyor belt.

[0074] According to a further embodiment of the proposed method, the target value of the corrugated board curvature can be time-dependent. This allows the target value to be varied, for example, during the transition from one product to another.

[0075] According to a further embodiment of the proposed method, the target value can be specified based on product metadata. For example, the product metadata can also include data regarding post-processing, such as a planned storage time. According to another embodiment of the proposed method, a vanishing curvature can be specified as the target value. The goal can therefore be to produce corrugated board without any curvature.

[0076] According to a further embodiment of the proposed procedure, an up-warp, a down-warp, or an S-warp can be specified as the target value.

[0077] Furthermore, the aforementioned task is also solved by the proposed system. The proposed system therefore comprises at least: a sensor for detecting corrugated board curvature, an electronically controllable actuator for reading and setting at least one control variable of the corrugated board machine, and an electronic control unit connected to the at least one sensor and the at least one actuator, which is configured to

[0078] • to detect corrugated cardboard curvature via at least one sensor,

[0079] • to read at least one actual value of at least one manipulated variable, and

[0080] • to determine a correction value relating to the at least one control variable for reducing the corrugated board curvature based on the recorded corrugated board curvature and the at least one actual value, wherein the correction value is determined using a machine model created by a learning process, and

[0081] • to control the corrugated board machine via at least one positioning device according to at least one correction value.

[0082] The proposed system is designed to use the machine model to precisely determine how a change to at least one control variable of the corrugated board machine affects the corrugated board produced. Based on the machine model, it is thus possible to determine by what value the at least one control variable must be adjusted relative to its actual value to reduce the corrugated board curvature. This value is the correction value of the at least one control variable. The proposed system uses a machine model created with a machine learning process and therefore enables the modeling of complex systems without requiring knowledge of all the underlying physical relationships. In particular, the proposed system improves the representation of nonlinear dynamic systems, such as a corrugated board machine.This can improve the control of the corrugated board plant with regard to a reduction in corrugated board curvature, without requiring the recording of all physical relationships of the influencing factors on the corrugated board curvature.

[0083] According to an exemplary embodiment of the proposed system, the control unit can be connected to a storage unit and configured to store and execute the created machine model.

[0084] According to an exemplary embodiment of the proposed system, the control unit can be configured to map at least one input value to at least one predicted output value using the machine model. This can enable a time- and computationally efficient mapping of a manipulated variable to a property of the corrugated board produced.

[0085] According to an exemplary embodiment of the proposed system, the at least one input value can include the at least one actual value of the at least one manipulated variable.

[0086] According to an exemplary embodiment of the proposed system, the control unit can be configured to map the machine model to a predicted corrugated board curvature from at least one input value. The output value of the machine model can therefore directly represent the corrugated board curvature. Deriving the corrugated board curvature from the output value is thus unnecessary. This can simplify the determination of the correction value.

[0087] According to an exemplary embodiment of the proposed system, the at least one input value can include an actual value of the corrugated board curvature. Thus, the proposed system can use the measured corrugated board curvature, along with the at least one actual value of the at least one manipulated variable, to determine the correction value. For example, a corrugated board curvature determined by the machine model can be compared with the actual value of the corrugated board curvature to verify the machine model. This can increase the reliability of the corrugated board curvature determined for the actual value of the at least one setpoint variable.

[0088] According to an exemplary embodiment of the proposed system, at least one input value can include a target value for the corrugated board curvature. Thus, determining the correction value can depend on any deviation of the measured corrugated board curvature from the target value. This can reduce the deviation over time and improve control quality. Furthermore, unnecessary calculations for determining the correction value and unnecessary control operations for tolerable corrugated board curvatures can be avoided.

[0089] For example, determining the correction value may depend on the fact that the target value of the corrugated board curvature deviates from the measured corrugated board curvature by more than a predetermined maximum value.

[0090] According to an exemplary embodiment of the proposed system, the control unit can be configured to map the at least one actual value of the at least one manipulated variable, the actual value of the corrugated board curvature, and the target value of the corrugated board curvature to the at least one correction value of the at least one manipulated variable using the machine model. Accordingly, the machine model can be trained to directly output the correction value for the at least one manipulated variable. Calculating the correction value by varying the value of the at least one manipulated variable can thus be unnecessary. This can reduce computational effort and the time required to determine the correction value.

[0091] According to an exemplary embodiment of the proposed system, the control unit can be configured to determine a mapping of the machine model depending on product metadata.

[0092] According to an exemplary embodiment of the proposed system, the control unit can be connected to and configured with a metadata acquisition device to acquire product metadata via the metadata acquisition device, in particular via user input or reading of stored information, which may be stored, for example, on an RFID chip, in a barcode, in a data matrix code (e.g. QR code), in a printed image or with a UV marking.

[0093] According to an exemplary embodiment of the proposed system, the product metadata can include the basis weight of the paper and / or the number of layers of the corrugated board and / or the flute height and / or the flute pitch.

[0094] According to an exemplary embodiment of the proposed system, the control unit can be configured to implement the learning process as a supervised learning process, a self-supervised learning process, a reinforcement learning process, or an unsupervised learning process.

[0095] The learning process can include various methods and algorithms suitable for learning from a data set and applying them to new, unknown situations in order to, for example, recognize patterns or solve specific tasks within these situations.

[0096] According to an exemplary embodiment of the proposed system, the control unit can be configured to perform the learning process as ensemble learning or the training of an artificial neural network.

[0097] According to an exemplary design of the proposed system, the control unit can be configured to perform initial training with historical plant data for the creation of the machine model.

[0098] According to an exemplary embodiment of the proposed system, the historical plant data can include recorded values ​​of at least one control variable and corrugated board curvature measured in production at the recorded values.

[0099] According to an exemplary embodiment of the proposed system, the control unit can be configured to retrain the machine model with values ​​of at least one manipulated variable and the corrugated board curvature recorded during operation.

[0100] According to an exemplary embodiment of the proposed system, the control unit can be configured to train and / or retrain the machine model with data sets from a plurality of corrugated board plants.

[0101] According to an exemplary embodiment of the proposed system, the control unit can be configured to receive the data sets for training and / or retraining from a cloud storage.

[0102] According to an exemplary embodiment of the proposed system, the control unit is configured to minimize a cost function during training and / or post-training. According to an exemplary embodiment of the proposed system, the cost function can relate to a deviation of the predicted output value, in particular a predicted corrugated board curvature, from a reference value, in particular a measured corrugated board curvature, derived from the training data.

[0103] According to an exemplary embodiment of the proposed system, the control unit can be configured to apply a gradient-based procedure, an evolutionary procedure, or a probabilistic method to minimize the deviation.

[0104] According to an exemplary embodiment of the proposed system, the control unit can be configured to store and / or read data for initial training and / or post-training on the storage unit.

[0105] According to an exemplary embodiment of the proposed system, the system can include a remote computer equipped with the control unit.

[0106] According to an exemplary configuration of the proposed system, the remote computer can be configured to perform training and / or retraining with the datasets in the cloud storage.

[0107] The proposed system does not fundamentally restrict the way in which the machine model created by a learning process can be integrated. At least three different integration options are conceivable.

[0108] In a first variant, according to an exemplary embodiment of the proposed system, the control unit can be configured to optimize the control of at least one manipulated variable using the machine model. In this variant, the machine model is therefore not part of the control loop that regulates the at least one manipulated variable as a function of the controlled variable, corrugated board curvature. However, the machine model can be used to optimize the control such that the at least one correction value determined by the control system reduces the corrugated board curvature. The correction value is thus determined indirectly using the machine model and based on the at least one actual value of the at least one manipulated variable and the measured corrugated board curvature.This can combine the ability to model complex nonlinear system behavior in corrugated board production and thus reduce the deviation in the prediction of corrugated board curvature with the simplicity and robustness of model-free control.

[0109] According to an exemplary embodiment of the proposed system, the control unit can be configured to simulate the time course of the predicted output value, in particular the predicted corrugated board curvature, for at least one test configuration of the control system in order to optimize the system using the machine model. Accordingly, the system can be configured to determine how the corrugated board plant is controlled with respect to the corrugated board curvature in at least one test configuration of the control system.

[0110] According to an exemplary embodiment of the proposed system, the control unit can be configured to simulate the transient response of the corrugated board curvature during the test configuration. The transient response can be described as the temporal evolution of the controlled variable, i.e., the corrugated board curvature, after an adjustment of at least one manipulated variable by the correction value.

[0111] According to an exemplary embodiment of the proposed system, the control unit can be configured to determine an optimized control configuration that minimizes settling time by varying the test configuration. The optimized configuration can thus be determined by simulating the settling behavior for different test configurations of the control system.

[0112] According to an exemplary configuration of the proposed system, the control unit can be configured to adjust coefficients of a P, PI, PD, or PID controller during optimization. This enables the proposed method to be used for optimizing standardized and widely used controllers, particularly on corrugated board plants with existing control systems. This can reduce the costs of implementing the proposed method.

[0113] According to an exemplary embodiment of the proposed system, the control unit can be configured to perform the optimization using evolutionary optimization. According to an exemplary embodiment of the proposed system, the control unit can be configured to transfer an optimized configuration to the control system.

[0114] According to an exemplary design of the proposed system, the control can be configured with a plurality of control elements for each manipulated variable.

[0115] In a second variant, according to an exemplary embodiment of the proposed system, the control unit can be configured to predict the curvature of the corrugated board using the machine model within the framework of model predictive control. The machine model is thus itself part of the control loop. The output value of the machine model is the curvature of the corrugated board. The curvature of the corrugated board is therefore not directly, but indirectly, through the use of the machine model. Based on the predicted curvature of the corrugated board and a setpoint, at least one correction value can then be determined, for example, by an optimizer.

[0116] According to an exemplary embodiment of the proposed system, the predicted corrugated board curvature can include a temporal development of the corrugated board curvature over a predetermined prediction horizon.

[0117] According to an exemplary embodiment of the proposed system, the control unit can be equipped with an optimizer that is set up to determine at least one correction value of at least one manipulated variable based on the predicted corrugated board curvature.

[0118] According to an exemplary embodiment of the proposed system, the optimizer can be configured to minimize the deviation between the predicted corrugated board curvature and the target value in order to determine at least one correction value.

[0119] According to an exemplary embodiment of the proposed system, the optimizer can be configured to apply a gradient-based, evolutionary, or probabilistic method to minimize the deviation. According to an exemplary embodiment of the proposed system, the optimizer can be configured to determine the deviation for a plurality of time points within the prediction horizon.

[0120] According to an exemplary embodiment of the proposed system, the optimizer can be configured to integrate the deviation over the majority of time points, in particular the prediction horizon.

[0121] In a third variant, according to an exemplary embodiment of the proposed system, the control unit can be configured to determine at least one correction value for at least one manipulated variable using the machine model within the framework of model predictive control. The machine model can thus directly map the at least one input value to the at least one correction value. This can further simplify the determination of the correction value. The optimizer may then be unnecessary.

[0122] According to an exemplary embodiment of the proposed system, the control unit can be configured to record at least one actual value for a plurality of control variables.

[0123] According to an exemplary embodiment of the proposed system, the at least one control variable can include a paper temperature and / or a steam temperature of a heated roller and / or a wrap angle of a preheater roller and / or a web speed and / or a moisture content and / or a water flow of a spray and / or a water distribution of a spray and / or a quantity of a glue application and / or a distribution of a glue application and / or a temperature within a double facer.

[0124] According to an exemplary embodiment of the proposed system, the control unit can be set up to capture at least one manipulated variable and a plurality of actual values.

[0125] According to an exemplary embodiment of the proposed system, the paper temperature can encompass temperatures at multiple points distributed across the working width. According to an exemplary embodiment of the proposed system, the control unit can be configured to determine the predicted and / or detected corrugated board curvature across the entire working width.

[0126] According to an exemplary embodiment of the proposed system, the predicted and / or detected corrugated board curvature can be a curvature of exactly one corrugated board section, in particular one corrugated board sheet or panel, from the plurality of corrugated board sections distributed over a working width.

[0127] According to an exemplary embodiment of the proposed system, the control unit can be configured to select exactly one corrugated board section from the majority of corrugated board sections distributed over a working width based on a user selection and / or a predetermined characteristic.

[0128] According to an exemplary embodiment of the proposed system, the control unit can be connected to an input device and configured to receive a user selection for choosing exactly one corrugated board section and / or the characteristic.

[0129] According to an exemplary embodiment of the proposed system, the predetermined characteristic can be a position along the working width or a relation of the curvature of the corrugated board section to the curvature of all corrugated board sections distributed across the working width.

[0130] According to an exemplary embodiment of the proposed system, the one corrugated board section can be the corrugated board section with the greatest curvature of all the corrugated board sections distributed across the working width.

[0131] According to an exemplary embodiment of the proposed system, the control unit can be configured to detect the corrugated board curvature with at least one sensor, in particular with at least one camera and / or at least one laser sensor.

[0132] According to one exemplary embodiment of the proposed system, the at least one sensor can be arranged above the conveyor belt. According to another exemplary embodiment of the proposed system, the at least one sensor can be arranged between a longitudinal cutting unit and a stacking tray on the corrugated board machine.

[0133] According to an exemplary embodiment of the proposed system, the control unit can be configured to fit a polynomial to the sensor-acquired measurements for calculating the curvature of one of the corrugated cardboard sections.

[0134] According to an exemplary embodiment of the proposed system, the control unit can be configured to take into account, when predicting the at least one output value with the machine model, a time difference between a correction of the at least one manipulated variable and an effect on the predicted output value.

[0135] According to an exemplary embodiment of the proposed system, the time difference can include a dead time that the corrugated board needs to be transported from a location where an actuating device acts on the corrugated board according to the actuating variable to a location of a detection device that captures the output value.

[0136] According to an exemplary embodiment of the proposed system, the dead time can be a multidimensional quantity for different positioning devices and detection devices arranged along the conveyor belt.

[0137] According to an exemplary embodiment of the proposed system, the target value of the corrugated board curvature can be time-dependent.

[0138] According to an exemplary design of the proposed system, the control unit can be configured to specify the target value depending on product metadata.

[0139] According to an exemplary embodiment of the proposed system, the target value can relate to a vanishing curvature.

[0140] According to an exemplary embodiment of the proposed system, the setpoint can relate to an up-warp, a down-warp, or an S-warp. In particular, the proposed system can be configured to carry out the proposed procedure.

[0141] The aforementioned task is also solved by a computer program product that includes machine-readable instructions which, upon execution, cause a control unit of the proposed system to carry out the proposed procedure.

[0142] Likewise, the aforementioned task is solved by a corrugated board plant using the proposed system.

[0143] The proposed statements regarding the advantages and possible design of the proposed procedure apply analogously to the proposed system, the proposed computer program product and the proposed corrugated board plant.

[0144] The attached figures illustrate possible implementation variants of the proposed solution.

[0145] This shows:

[0146] Figure 1 shows a schematic side view of an exemplary

[0147] Corrugated board plant with a first embodiment of the proposed system for controlling the corrugated board plant;

[0148] Figure 2 shows a perspective view of a schematic structure of a single-wall corrugated cardboard;

[0149] Figure 3 shows a perspective view of a schematic structure of a double-wall corrugated cardboard;

[0150] Figure 4A shows a cross-section along the working width of a curved

[0151] Corrugated cardboard sheet;

[0152] Figure 4B shows a cross-section of a longitudinally cut corrugated board sheet;

[0153] Figures 5, 6 and 7 are schematic side views of a corrugated board plant with further embodiment variants of the proposed system; and Figures 8 to 13 are flowcharts of different embodiments of the proposed method.

[0154] Figure 1 shows an exemplary corrugated board production line with a system for controlling the corrugated board production line according to a first embodiment of the proposed solution. In the corrugated board production line shown in Figure 1, two unwinding devices 101 and 102 are provided for two material webs M1 and M2 to feed the material webs M1 and M2, typically made of paper, to a first single-sided machine E1, for example, in the form of a so-called "single facer". A second material web M2 is guided between two corrugating rollers W1 and W2 of the first single-sided machine E1. The pair of corrugating rollers W1 and W2 creates a corrugation in the second material web M2, thus producing a corrugated layer from the second material web M2, which – after passing a gluing device L of the first single-sided machine E1 – is bonded to the first material web M1 by a pressure device AP.At the output of the first single-sided machine E1, a single-sided corrugated board web WPB is provided. This corrugated board web WPB is transported via a (high-level) transport device to a bridge B of the corrugated board plant for further processing.

[0155] In the further manufacturing process, the corrugated board web WPB is fed along a conveying direction R to a preheating unit E2. This preheating unit E2 is also passed through by a third material web M3. This third material web M3 is supplied by a separate unwinding device 103 and, if necessary, printed by a printing unit DR before entering the preheating unit E2.

[0156] After passing through the preheating unit E2, one of the webs WPB, M3 is coated with glue in a gluing unit E3 equipped with a gluing system. In the subsequent manufacturing process, the glued corrugated board web WPB is then bonded to the third material web M3 in a pressure unit E4. For this purpose, the pressure unit E4 – particularly when using conventional starch glue for bonding the corrugated board web WPB and the material web M3 – can include one or more heating plates. The corrugated board web WPB and the third material web M3 are pressed together, for example, by a motor-driven endless pressure belt guided by guide rollers. The pressure unit E4 can be configured as a combined tension and heating section. The second material web M2 is then used, for example, to produce a flute layer 21 for a corrugated board WP as shown in Figure 7. The first material web M1 serves to form one of the top layers 11 or 12.The third material web M3 is used to produce the further cover layer 12 or 12 for the single-wall corrugated board WP.

[0157] In the embodiment of the corrugated board machine shown in Figure 9, the pressure unit E4 produces a single-wall corrugated board, which is then fed to a longitudinal cutting or creasing unit E5 of the corrugated board machine in the further manufacturing process. The longitudinal cutting or creasing unit E5 cuts the continuous single-wall corrugated board web lengthwise.

[0158] Subsequently, a further cutting process takes place in a cross-cutting unit E6 to produce corrugated board WP, ​​ready for further processing, in the form of corrugated board sheets from the continuous corrugated board web. These corrugated board sheets are conveyed via a conveyor belt F of the corrugated board plant shown in Figure 9 to a stacking tray E7. From the stacking tray E7, the corrugated board sheets can be transported away in stacks.

[0159] The proposed system for controlling the corrugated board production line comprises a sensor S located above the conveyor belt FB between the longitudinal cutting unit E5 and the transverse cutting unit E6 for detecting corrugated board curvature, an electronically controlled actuator SV1 for reading and setting at least one control variable of the corrugated board production line, and an electronic control unit ECU connected to the at least one sensor S and the at least one actuator SV1. The control unit ECU is configured to detect corrugated board curvature via the at least one sensor S, to read at least one actual value of the at least one control variable, and, based on the detected corrugated board curvature and the at least one actual value, to determine a correction value for the at least one control variable to reduce the corrugated board curvature.Here, the correction value is determined using a machine model F created by a learning process. Furthermore, the control unit ECU is configured to control the corrugated board machine via at least one positioning device SV1 according to the at least one correction value.

[0160] The proposed system is therefore designed to use machine model F to specifically determine how a change to at least one control variable of the corrugated board machine affects the corrugated board produced. Thus, based on machine model F, it is possible to determine by what value the at least one control variable must be adjusted relative to its actual value in order to reduce the corrugated board curvature. This value is the correction value of the at least one control variable.

[0161] Figures 2 and 3 each show a section of the structure of a corrugated cardboard in perspective view, whereby the corrugated cardboard WP shown does not have any curvature.

[0162] Figure 2 shows a single-wall corrugated board WP. This board has a single flute 21 arranged between two linerboards 11 and 12. The flute 21 has a sinusoidal cross-section, while the linerboards 11 and 12 are flat. One linerboard is referred to, for example, as the inner layer 11, while the other is referred to as the outer layer 12.

[0163] In a double-wall corrugated board WP according to Figure 3, two corrugation layers 21, 22 are provided between the two liner layers 11, 12. A flat intermediate layer 13 is arranged between these corrugation layers 21, 22. The corrugation types of the corrugation layers 21, 22 can be fundamentally different or identical. In other words, in the first case, the sinusoidal profiles of the corrugation layers 21, 22 can differ from one another with respect to their wave pitch, i.e., the distance between two immediately consecutive wave crests or between two immediately consecutive wave troughs, and / or with respect to their wave height.

[0164] Figure 4A shows a cross-section of a curved corrugated board sheet along its working width AB. The corrugated board sheet WPB has no cut along its entire working width AB. Two opposite ends of the corrugated board sheet WPB, located along the working width AB, are curved in a vertical direction relative to a central area of ​​the corrugated board sheet WPB. The corrugated board sheet WPB thus exhibits a curvature that is particularly pronounced at the edges. In contrast, the curvature of the corrugated board sheet WPB is less pronounced in the central area than at the edges.

[0165] It is generally possible for a corrugated board (WPB) to be under tension. Such tension can relax during cutting, particularly longitudinal cutting, leading to pronounced curvature of the longitudinally cut WPB. Figure 4B shows a cross-section of a corrugated board (WPB) longitudinally cut into three sections WP, WP', and WP' along the working width AB.

[0166] In principle, it is conceivable and possible that the corrugated board WPB, through the longitudinal section along the working width AB, has a plurality of curved sections WP, WP',WP" which are not necessarily curved to the same degree or in the same orientation.

[0167] In alternative embodiments of the proposed system that differ from the one shown in Figure 1, a different arrangement of the sensor S would also be conceivable and possible, provided that the location and orientation of the sensor S are suitable for determining the corrugated board curvature. In particular, an arrangement along the conveyor belt FB behind the longitudinal cutting unit E5 is advantageous, since after longitudinal cutting of the corrugated board web WPB, stresses in the corrugated board web WBP can relax and thereby form a curvature.

[0168] In the embodiment shown in Figure 1, the positioning device SV1 is arranged on a preheating unit E2 of the corrugated board machine. For example, the positioning device SV1 could be a device that controls a paper temperature. In particular, it would be conceivable and possible to control the wrap angle of the heated rollers of the preheating unit E2. In this case, the positioning device SV1 is configured to send an actual value of the wrap angle as the actual value of the manipulated variable to the control unit ECU. Based on the actual value of the wrap angle and the corrugated board curvature detected by the sensor S, the control unit ECU determines the correction value for the wrap angle using the machine model F. The positioning device SV1 is then controlled by the control unit ECU according to the correction value, i.e., the wrap angle is adjusted by the correction value.

[0169] In further embodiments, the proposed system can also comprise a plurality of different electronically controllable actuators SV1, SV2 for a plurality of manipulated variables, each connected to the control unit ECU and configured to send an actual value of the respective manipulated variable to the control unit ECU. The actuators SV1, SV2 can also each be configured to control the respective manipulated variable depending on a correction value determined by the control unit ECU. Likewise, the control unit ECU can be connected to more than one sensor S and configured to detect the corrugated board curvature via the plurality of sensors S.

[0170] In further embodiments of the proposed system that differ from the embodiment shown in Figure 1, the system can also be designed with a storage unit MEM on which the machine model F is stored.

[0171] Figure 5 shows a corrugated board production line with a further embodiment of the proposed system, in which the control unit ECU is configured to download data sets for training and / or retraining from a cloud storage system CLOUD and also to store and / or read data for initial training and / or retraining on the storage unit MEM. Furthermore, the control unit ECU is connected to a metadata acquisition device META for capturing product metadata.

[0172] The proposed system can thus be configured to obtain training and / or retraining data not only directly from a corrugated board machine, but also from the MEM cloud storage system via the COULD cloud storage. This allows machine model F to access a larger number of data sets than if only the data sets from a single corrugated board machine were available. Furthermore, the system is equipped with the META metadata acquisition device to capture product metadata, whereby machine model F is trained to consider various product metadata, such as properties of the raw materials, the manufacturing process, and the finished corrugated board, when determining the correction value for at least one control variable.

[0173] The metadata acquisition device META can be configured, in particular, to capture product metadata via user input or by reading stored information. The stored information can be, for example, on an RFID chip, in a barcode, in a data matrix code (e.g., QR code), in a printed image, or in a UV marking.

[0174] In another alternative embodiment of the proposed system, the control unit ECU can also be spatially separated from the corrugated board machine and / or configured to train / retrain the machine model exclusively with data from a cloud storage system (CLOUD) or a local storage unit (MEM). Figure 6 shows a corrugated board machine with a variant of the proposed system, in which the control unit ECU is part of a remote computer (RC) and the system further comprises a transceiver T coupled to the at least one actuator (SV1) and the at least one sensor (S). The actuator (SV1) has a control element (C1) for controlling the at least one control variable, which is configured to determine a correction value for the at least one control variable and send it to the control unit ECU or the actuator (SV1).Such a control element C1 can be, for example, a P, PI, PD, or PID controller. Accordingly, the system is configured with control element C1 to regulate at least one manipulated variable based on the corrugated board curvature as the controlled variable, i.e., to determine at least one correction value for the at least one manipulated variable, by which the manipulated variable is adjusted by the actuator SV1.

[0175] The transceiver T is configured to send the at least one actual value of the manipulated variables and the measured corrugated board web curvature to the remote computer RC, which uses the machine model F to optimize the control element C1 for the at least one manipulated variable, i.e., to determine an optimized configuration of the control element C1. The optimized configuration is sent via the transceiver T to the control element C1, so that the control element C1 indirectly determines the correction value for the at least one actuator SV1 using the machine model F.

[0176] The machine model F is regularly retrained, or upon explicit request, using the actual values ​​of the manipulated variable and the recorded corrugated board web curvature.

[0177] In principle, it is also conceivable and possible that the proposed system has further control elements C2, C3.

[0178] Figure 7 shows a corrugated board production line with a further embodiment of the proposed system featuring cascaded paper temperature control. The system includes two actuators, SV1 and SV2, each equipped with control elements C1 and C2 (i.e., two temperature actuators), located in the preheater E2. Each of the two control elements C1 and C2 in the preheater E2 is configured to use one of the temperatures 0_1, 0_2, or 0_3 as the manipulated variable and the paper temperature as the controlled variable. The higher-level control element C3 uses the target paper temperature from the two control elements C1 and C2 as the manipulated variable and the corrugated board curvature as the controlled variable. The higher-level control element C3 is optimized by the ECU of the proposed system to reduce the corrugated board curvature using machine model F.

[0179] Figure 8 shows a flowchart of a first embodiment of the proposed method. Here, the curvature of the corrugated board and at least one actual value of at least one control variable of the corrugated board machine are detected. Subsequently, at least one correction value relating to the at least one control variable is determined to reduce the curvature of the corrugated board based on the detected curvature and the at least one actual value, using a machine model F created with a machine learning method. The correction value is used to control the corrugated board machine, i.e., to adjust the at least one control variable by the at least one correction value.

[0180] The proposed method does not fundamentally restrict the way in which the machine model F, created by a learning process, can be integrated. At least three different integration variants are conceivable.

[0181] According to a first variant, the machine model F is used to optimize the control of at least one manipulated variable.

[0182] Figures 9A and 9B illustrate a partial aspect of an embodiment of the proposed method according to the first integration variant. The partial aspect shown in Figure 9A concerns the determination of the at least one correction value. Using a learning method, a machine model F is created in step 1, which maps at least one actual value of the at least one manipulated variable u as input value X to a predicted corrugated board curvature as output value Y. Subsequently, in step 2a, a PID control of the at least one manipulated variable u is optimized using the machine model F.

[0183] To optimize the process, the application of the PID controller is first simulated. This involves determining the correction value that the PID controller calculates based on a setpoint and a measured corrugated board curvature for a given configuration, i.e., PID coefficients. Using the machine model, the corrugated board curvature that results when the manipulated variable is adjusted by the correction value is then predicted. This process can be repeated until the control is complete, for example, until the corrugated board curvature converges to the setpoint. Subsequently, the PID controller configuration is varied, and the procedure is repeated. This allows the configuration that optimizes the control to be determined, for example, minimizing the settling time of the corrugated board curvature. The optimized configuration is then applied to the PID controller.The optimized PID control then determines at least one correction value during operation, with which the corrugated board plant is controlled.

[0184] In this variant, the machine model F is therefore not part of the control loop that regulates the at least one manipulated variable u as a function of the controlled variable corrugated board curvature. However, the machine model F is used to optimize the control in such a way that the at least one correction value determined by the control system reduces the corrugated board curvature. The correction value is thus determined indirectly using the machine model F and based on the at least one actual value of the at least one manipulated variable u and the measured corrugated board curvature.

[0185] Figure 9B is an alternative representation of the process shown in Figure 9A.

[0186] In a second integration variant, according to a further embodiment of the proposed method, corrugated board curvature can be predicted using the machine model F within the framework of a model predictive control.

[0187] Figures 10A and 10B illustrate the process of determining the at least one correction value according to the second integration variant. Using the machine model F created in section 1, a corrugated board curvature Y is predicted at a later time t+At for the at least one actual value of the at least one manipulated variable u within the framework of model predictive control. Subsequently, an optimizer determines the at least one correction value for the at least one manipulated variable u based on the predicted corrugated board curvature Y(t+At) and a setpoint f for the corrugated board curvature by minimizing a deviation e between the predicted corrugated board curvature Y(t+At) and the setpoint f. In the example shown in Figure 10A, time t is considered as an equidistant discrete quantity, so that different time points t can each be assigned an index k, k+N1, ...

[0188] In the second variant, the machine model F is itself part of the control loop. The output value Y of the machine model F is the corrugated board curvature. The corrugated board curvature is therefore not directly, but indirectly, through the use of the machine model F. Based on the predicted corrugated board curvature Y(t+At) and a target value f, the optimizer then determines at least one correction value.

[0189] Figure 10B is an alternative representation of the process shown in Figure 10A.

[0190] According to a third integration variant, in a further embodiment of the proposed method that differs from the one shown, the machine model F can determine at least one correction value for at least one manipulated variable u within the framework of model predictive control. The machine model F can thus directly map the at least one input value X to the at least one correction value.

[0191] According to a further alternative embodiment of the proposed method, the prediction of the at least one output value Y using the machine model F can take into account a time difference between a correction of the at least one manipulated variable u and its effect on the predicted output value Y. This can improve training / retraining, since in the training datasets the actual values ​​of the at least one manipulated variable and the measured corrugated board curvature are typically separated by a dead time At.

[0192] The actual measured values ​​Y'(t+At) acquired after the dead time At can thus be compared with predicted output values ​​Y(t+At) in order to retrain the machine model F.

[0193] Figure 11 shows a variant of the process depicted in Figures 10A and 10B. Here, too, the machine model F, within the framework of model predictive control, predicts a corrugated board curvature Y at a later time t+At for at least one actual value of at least one manipulated variable u. Subsequently, an optimizer determines at least one correction value for the at least one manipulated variable u based on the predicted corrugated board curvature Y(t+At) and a setpoint f for the corrugated board curvature by minimizing the deviation between the predicted corrugated board curvature Y(t+At) and the setpoint f. The future course of the manipulated variable u(t+At) determined by the optimizer is forwarded to a control device of the corrugated board machine WPA, which receives the future course of the manipulated variable u(t+At) and controls the manipulated variable accordingly.Accordingly, the dead time At is not waited for before the manipulated variable u is controlled with regard to the setpoint f. In contrast to the process shown in Figures 10A and 10B, the corrugated board machine WPA records an actual progression of the corrugated board curvature Y'(t+At), which is then used to retrain the machine model F and thus take into account the curvature values ​​Y'(t+At) actually measured after the dead time At in future control.

[0194] In principle, the dead time At can also be a multidimensional quantity for different machine elements and detection devices arranged along the conveyor belt FB.

[0195] Figure 12 shows a schematic representation of the time sequence of a section of an embodiment of the proposed method for predicting the corrugated board curvature Y. A corrugated board web, transported on a conveyor belt, passes through three temperature-setting devices at different times t_1, t_2, t_3. Each device provides at least one actual temperature value 0_1, 0_2, 0_3-1, 0_3-2. Based on these actual values ​​0_1, 0_2, 0_3-1, 0_3-2 and product metadata, the machine model F, created using a machine learning method, predicts the corrugated board web curvature Y(At) that the corrugated board is expected to exhibit when passing a sensor for measuring the corrugated board curvature Y.

[0196] Figure 13 also shows a schematic representation of the temporal sequence of a section of an embodiment of the proposed method for predicting the corrugated board curvature Y, wherein a corrugated board passes three adjusting devices at different times t_1, t_2, t_3 during transport on a conveyor belt. The first two adjusting devices each provide actual values ​​for a temperature 0_1, 0_2 and a humidity M_1, M_2, and a water spray profile HB_1, HB_2. The third actuator provides two temperatures 0_3-1 , 0_3-2, so that, based on the actual values ​​0_1 , 0_2, 0_3-1 , 0_3-2, M_1, M_2, HB_1, HB_2 and product metadata, the corrugated board web curvature Y is predicted using the machine model F created via a learning process, which the corrugated board is expected to have when passing a sensor for measuring the corrugated board curvature.The proposed solution is not limited to the specific embodiments discussed here. Rather, the proposed solution encompasses any combination of features from the discussed embodiments, provided that these can be combined in a feasible manner by those skilled in the art.

[0197] Reference symbol list

[0198] 101, 102, 103 Roll-off device

[0199] 11 Inner layer

[0200] 12 Outer layer

[0201] 13 Intermediate layer

[0202] 21, 22 Wave position

[0203] AP pressure device

[0204] AB working width

[0205] B Bridge

[0206] C1, C2, C3 Control element

[0207] CLOUD Cloud Storage

[0208] DR printer unit

[0209] E1 Single-sided machine

[0210] E2 Preheating unit

[0211] E3 Glue Unit (Glue Works)

[0212] E4 Pressure unit (traction section)

[0213] E5 Longitudinal cutting unit / Grooving unit

[0214] E6 Cross-cutting unit

[0215] E7 Stacking tray

[0216] ECU control unit

[0217] F machine model

[0218] FB Conveyor belt

[0219] L Gluing device

[0220] M1, M2, M3 Material track

[0221] MEM storage

[0222] Meta data capture device

[0223] R Conveyor direction

[0224] RC Remote Computer

[0225] S Sensor

[0226] SV1, SV2 Actuator

[0227] T Transceiver

[0228] W1, W2 corrugated roller

[0229] WP, WP' Corrugated board / Corrugated board sheets

[0230] WPB corrugated board

[0231] WPA corrugated board plant e_i , e_2, e_3-i , e_3-2 Temperature M_1, M_2, Humidity HB_1, HB_2, Water spray t_1 , t_2, t_3 Time e Deviation

[0232] Dead time At

[0233] Y Output value

[0234] Y' measured value

[0235] X Input value f Setpoint u Control variable

Claims

1. Claims 1. Procedure for controlling a corrugated board plant, comprehensive: Capturing the curvature of corrugated board and at least one actual value of at least one control variable of the corrugated board machine, Determine at least one correction value relating to at least one control variable to reduce the corrugated board curvature based on the detected corrugated board curvature and the at least one actual value, wherein the correction value is determined using a machine model (F) created with a learning method, and Control of the corrugated board plant according to at least one correction value.

2. Method according to claim 1, characterized in that the machine model (F) is used to optimize the control of at least one manipulated variable.

3. Method according to claim 2, characterized in that, in order to optimize the control with the machine model (F), a time course of a predicted corrugated board curvature is simulated for at least one test configuration of the control.

4. Method according to claim 2 or 3, characterized in that coefficients of a P, PI, PD or PID control are adjusted during optimization.

5. Method according to one of the preceding claims, characterized in that the machine model (F) is used to predict corrugated board curvature within the framework of a model predictive control.

6. Method of claim 5, characterized in that, based on the predicted corrugated board curvature, the at least one correction value of the at least one control variable is determined.

7. Method according to claim 5 or 6, characterized in that, in order to determine the at least one correction value, a deviation (e) between the predicted corrugated board curvature and a target value (f) is minimized.

8. Method according to one of the preceding claims, characterized in that, within the framework of a model predictive control, the at least one correction value of the at least one manipulated variable is determined with the machine model (F).

9. Method according to one of the preceding claims, characterized in that the at least one control variable comprises a paper temperature and / or a steam temperature of a heated roller and / or a wrap angle of a preheater roller and / or a web speed and / or a moisture content and / or a water flow of a spraying and / or a water distribution of a spraying and / or a quantity of a glue application and / or a distribution of a glue application and / or a temperature within a double facer.

10. Method according to one of the preceding claims, characterized in that the corrugated board curvature is a curvature of exactly one corrugated board section from the plurality of corrugated board sections distributed over a working width (AB).

11. Method according to claim 10, characterized in that exactly one corrugated board section is selected from the plurality of corrugated board sections distributed over a working width (AB) based on a user selection and / or a predetermined characteristic.

12. System for controlling a corrugated board plant comprising: at least one sensor (S) for detecting corrugated board curvature, at least one electronically controllable actuator (SV1, SV2) for reading and setting at least one control variable of the corrugated board plant, at least one electronic control unit (ECU) connected to the at least one sensor (S) and the at least one actuator (SV1, SV2), which is configured as follows: • to detect corrugated board curvature via at least one sensor (S), • to read at least one actual value of at least one manipulated variable, and • based on the recorded corrugated board curvature and at least one actual value, one control variable to determine the relevant correction value for reducing the corrugated board curvature, wherein the correction value is determined using a machine model (F) created by a learning process, as well as • to control the corrugated board machine via at least one positioning device (SV1, SV2) according to at least one correction value.

13. System according to claim 12, configured to carry out a method according to any one of claims 1 to 11.

14. Corrugated board machine for producing corrugated board laminated on at least one side. (WP) with a system according to claim 12 or 13.

15. Computer program product comprising machine-readable instructions which, upon execution, cause the electronic control unit (ECU) of a system according to claim 12 or 13 to carry out a method according to any one of claims 1 to 11.

Images