Method and production plant for manufacturing a material plate, computer program product and use of the computer program product

Abstract

The invention relates to a method for producing material sheets (12) in a production plant (1), wherein at least one material (10) is formed into a nonwoven fabric (11) in devices (2, 3, 4, 5, 6, 7, 8, 9) of the production plant (1) and pressed into the material sheet (12) which has certain quality parameters, wherein preferably the material (10) is coated with adhesive before being formed into the nonwoven fabric (11), wherein the production plant (1) and / or the devices (2, 3, 4, 5, 6, 7, 8, 9) are controlled or regulated by means of at least one control unit (20), and wherein at least one quality value (24) of at least one quality parameter of the material sheet (12) to be produced is determined as a process parameter by means of an algorithm (32) based on artificial intelligence, and wherein the algorithm (32) was generated using a database (30).which data sets comprehensively include at least one quality parameter and input parameters (23) correlating with the quality parameter. The invention is characterized in that the data basis (30) is at least partially based on calculated data sets (30e) from at least one physical model (35), in particular a mathematical description of one or more processes occurring during the manufacture of material sheets (12). The invention further relates to a production plant (1) for the manufacture of material sheets (12) as well as a computer program product and the use of a computer program product.

Description

[0001] The invention relates to a method for producing material sheets according to the preamble of claim 1 and to a production plant for producing material sheets according to the preamble of claim 23. Furthermore, the invention relates to a computer program product according to claim 27 and to the use of a computer program product according to claim 28.

[0002] Continuous single- or double-belt presses are generally used to manufacture composite sheets from plastics and / or raw materials based on biological resources, such as wood or annual and perennial plants. For this purpose, the respective raw material is first prepared by cleaning and shredding, usually dried, and, if necessary, mixed with a binder in a predetermined ratio. The binder is typically an adhesive, which is applied to the material either as a liquid or dry sizing agent.The processed raw material is then laid or spread in a spreading device in predetermined layers and / or basis weight ratios and / or orientations to form a nonwoven fabric and fed into a press, where the nonwoven fabric is compressed into a sheet of material under pressure and heat. In particular, the production of these sheets takes place in a continuous operation, whereby the nonwoven fabric is formed on an endlessly circulating forming belt and compressed into a strand of material in a continuously operating press, which is then cut into individual sheets.

[0003] In manufacturing, to increase productivity and maintain quality, the aim is to control a large number of parameters, which include not only the settings for the plant or the individual machines arranged within it, but also the influence on the material quality and the physical properties of the material sheet to be produced.

[0004] In practice, a technologically experienced natural scientist, particularly a wood technologist, draws on their previous experience and predefines the values ​​for the production plant, such as the amount of binder per unit of material, the thickness and / or basis weight of the material spread to form a fleece, the pressure and heat input in the press during the pressing of the fleece into the material sheet or strand, as well as the production speed. The technologist then attempts to optimize these values ​​based on actual and target values ​​and their experience in order to improve the properties of the material sheet being produced and / or increase the quantity of material sheets produced per unit of time.

[0005] To determine the properties and quality data of a material sheet, it is known to take a laboratory section from the sheet or sheet strand and analyze it using sometimes complex procedures, either in a laboratory on the production site or externally. Particularly when determining the properties of material sheets, it can take several hours to several days to obtain the necessary quality data for the laboratory section, for example, because the section needs to cool down or because certain measurements require analysis of the section over an extended period. The downtime between creating the laboratory section and determining the properties and quality data of the material sheet is therefore considerable, which creates the risk of producing rejects over a prolonged period.

[0006] Furthermore, it is also known that, through the collection of empirical data and numerous previous manufacturing scenarios, which particularly represent data from measurements taken before and after pressing the material sheet, quality predictions can be made and the production plant settings can be optimized in advance using a physical model. The physical model used, or the use of several physical models, is based on characteristic values, curves, or maps determined scientifically or during operation. From these characteristic values, a physical model is described by mathematical formulas, taking into account machine-specific or process-specific characteristics. In addition to actual values, laboratory values ​​can also be retrieved, and target values ​​can be considered if necessary.

[0007] It is already known from WO 2021 / 165394 that, during the production of a material sheet, an artificial intelligence-based algorithm can be used to make a statement about at least one quality parameter of the manufactured material sheet based on input parameters from the production plant's equipment. This has generally proven successful and yields good results. The algorithm is formed using known data, which usually exhibits good or even very good quality parameters. In particular, the database for certain quality parameters and input parameters is underrepresented, so that the algorithm typically incorporates less data in this area compared to other areas. Furthermore, very similar sheets with only minor differences in parameters are often produced in a single plant.

[0008] The object of the present invention is to provide a method and a production plant for manufacturing material panels, as well as a computer program product and the use of a computer program product, which makes it possible to produce a wide variety of material panels in a broad range of applications with high quality values ​​in a production plant.

[0009] A further object of the invention is to specify a method and a production plant for the manufacture of material plates, as well as a computer program product and the use of a computer program product, in which a reliable quality statement is made possible over a wide range of input parameters by means of an algorithm based on artificial intelligence.

[0010] A further object of the invention is to specify a method and a production plant for the manufacture of material plates, as well as a computer program product and the use of a computer program product, by means of which a production plant can be operated essentially autonomously.

[0011] Finally, it is also a task of the invention to be able to react quickly and reliably to sudden changes in the process for manufacturing the material plate.

[0012] These and other problems are solved by a method for producing material sheets in a production plant according to claim 1 and a production plant for producing material sheets according to claim 23. In addition, the problems are also solved by a computer program product 27 and a use of a computer program product according to claim 28.

[0013] Advantageous embodiments of the method are set out in claims 2 to 22, and advantageous embodiments of the production plant are set out in claims 24 to 26.

[0014] The method according to the invention is particularly designed for implementation on the production plant according to the invention, especially on an advantageous embodiment thereof. The production plant according to the invention is particularly designed for carrying out the method according to the invention or an advantageous embodiment thereof. The method for producing material sheets is in particular a method for process optimization of a production plant for producing material sheets. The production plant for producing material sheets is in particular a device for process optimization of a production plant for producing material sheets.

[0015] In the process for manufacturing material sheets in a production plant, at least one material is formed into a nonwoven fabric and pressed into a material sheet with specific quality parameters in fixtures of the production plant. Preferably, the material is coated with adhesive before being formed into the nonwoven fabric. The production plant and / or the fixtures are controlled or regulated by means of at least one controller, which preferably includes a programmable logic controller (PLC). An artificial intelligence-based algorithm determines at least one quality value of a quality parameter of the material sheet to be manufactured based on input parameters as process parameters. The algorithm is based on a database comprising data sets containing at least one quality parameter and input parameters correlated with that quality parameter.

[0016] The method according to the invention is characterized in that the database is at least partially based on calculated data sets from at least one physical model, in particular a mathematical description of one or more processes occurring during the production of material sheets.

[0017] The invention is essentially based on the understanding that combining physical relationships, particularly in the data set, with an artificial intelligence-based algorithm improves the algorithm's accuracy, enabling even more precise predictions regarding quality parameters. A broader data set, especially one containing calculated data, enhances predictions and allows for faster responses to changes in the production process. Therefore, it is no longer necessary to rely solely on laboratory samples for the data set. Instead, the data set can be expanded through verified relationships within the physical model.The physical model can include one or more mathematical descriptions of one or more processes occurring during the production of material sheets, in particular during the pressing of the pressed material mat into a sheet.

[0018] A further advantage of a database containing at least partially calculated datasets from at least one physical model is that the algorithm can learn physical relationships more quickly with a significantly larger database, as the physical relationships are at least partially incorporated into the algorithm through the database. This allows quality values ​​to be determined more easily without requiring the algorithm to be adapted to plant-specific or process-specific conditions.

[0019] Furthermore, the inventive method allows a production plant to be operated within safe limits, since the data sets calculated using the database are based on established physical and / or mathematical relationships. Moreover, the algorithm makes it possible to generate physically based, explainable suggestions for recipe changes, as it is based on a broad database.

[0020] The calculated data sets are simulated data sets, which are determined or calculated using the physical model.

[0021] An artificial intelligence-based algorithm is understood to be an algorithm based on methods used in the development of artificial intelligence. Advantageously, an artificial intelligence-based algorithm includes or is based on at least one method or a modification thereof from the following groups: simple methods, such as linear regression, polynomial regression, and functional regression; and / or advanced methods, such as random forest regression, support vector regression, K nearest neighbors regression, neural networks, recurrent neural networks, convolutional neural networks, residual networks, and Bayesian networks; and / or classification-based methods, such as K nearest neighbors classification, decision trees, random forests, naive Bayes, and support vector machines.

[0022] Simple methods such as linear regression, polynomial regression, and functional regression are understood to be particularly widespread procedures that originate from the field of statistics and can be used for machine learning purposes.

[0023] Advanced methods, such as Random Forest Regression, Support Vector Regression, K Nearest Neighbors Regression, Neural Networks, Recurrent Neural Networks, Convolutional Neural Networks, Residual Networks, and Bayesian Networks, are understood to be methods that fall almost exclusively within the field of machine learning and have little relevance outside of this domain. Classification-based methods, such as K Nearest Neighbors Classification, Decision Trees, Random Forests, Naive Bayes, and Support Vector Machines, can also be considered a subset of advanced methods.

[0024] Preferably, in addition to a mathematical method, particularly with mapping functions and / or neighbor functions, the algorithm is given limits for certain parameters within which it can operate freely. For a method based on neural networks, for example, one or more mapping layers and weighting factors are specified, whereby the weights in particular are optimized by training with the database.

[0025] The artificial intelligence-based algorithm is not an algorithm in the strictest sense. Rather, it is a model or program that was created or trained using artificial intelligence based on the data. In particular, this model can be modified by retraining it with additional datasets that were not included in the original dataset used to develop the model.

[0026] Equipment of a production plant comprises components of the production plant used for the manufacture of material sheets. In particular, this includes equipment assigned a specific task within the production of material sheets, such as crushing, spreading, or pressing the material. Preferably, the equipment consists of individual machines or groups of individual machines.

[0027] The algorithm determines at least one quality value, preferably determining quality values ​​for several quality parameters in order to characterize the manufactured material sheet as completely as possible. Preferably, the database can also be designed to include one or more quality parameters for specific input parameters.

[0028] Quality parameters, in this context, are properties and / or parameters of the manufactured material sheet that characterize its quality. Examples of quality parameters for a material sheet include its strength along a major or minor axis, its transverse tensile strength along a major or minor axis, its modulus of elasticity along a major or minor axis, and its surface finish. Values ​​for these quality parameters can be determined through analyses, such as the analysis of a laboratory section. The algorithm aims to determine quality values ​​for at least one quality parameter, thus enabling more accurate assessments of the manufactured material sheet's quality with a reduced number of laboratory sections and, in particular, during ongoing operation under changing conditions.

[0029] Input parameters are defined as all process parameters that can influence the production of a material sheet. In particular, these can include parameters from the equipment of the production plant and / or from sensors in the production plant, or other measured values. Preferably, they are parameters that are incorporated into a physical model or determined using such a physical model.

[0030] The input parameters can include product parameters, plant parameters, material parameters, and environmental conditions. Preferably, the input parameters include only those relevant to the production of the material sheet and those related to the material sheet.

[0031] Product parameters are adjustable parameters of the material panel, which are specified or defined directly or indirectly. In particular, product parameters can include the type of material panel, for example, for wood-based panels, this can generally be particleboard, MDF, or OSB, or variations thereof, or the thickness of the material panel.

[0032] Plant parameters are defined here as parameters of the entire production plant or of the individual devices within the production plant. These parameters are specifically defined or predetermined and are entered into the control system or the physical model and / or determined by sensors. For example, plant parameters include the press length of the pressing device, the spreading width of the nonwoven fabric, the number of spreading heads for controlling the fabric, and the use and type of preheating for the nonwoven fabric. Plant parameters can also include the pressing pressure of individual cylinders in the press, a pressing profile (longitudinal and / or transverse to the conveying direction), the temperature of heating plates, the fill levels of intermediate storage tanks, inventory levels, or the temperature of a drying stage for drying the material before forming a nonwoven fabric.The plant parameters can include, in particular, setpoint values, which are preferably entered into the control system by an operator or a technical employee of the production plant, as well as actual values, which are measured, for example, by sensors. Consumption values ​​for the production plant or its equipment, such as power consumption and / or lubrication parameters, are also included in the plant parameters. Furthermore, specifications for a minimum production volume of material sheets or minimum quality standards for the material sheets are also considered plant parameters.

[0033] Environmental conditions include, for example, weather data, ambient temperature, humidity, air pressure, and the room temperature during the production of the pressed mat or material sheet. Furthermore, environmental conditions also encompass the availability of raw materials, electricity prices, and similar factors.

[0034] Furthermore, material parameters encompass all parameters of the material used that are known or can be determined using sensors. For the production of wood-based panels, material parameters include, for example, the type of wood, the density of the material, the thickness of the material, the width of the material, the length of the material, the moisture content of the material, the composition of the material used, in particular the ratio of virgin to recycled material, the binder used and the amount of binder used and / or the ratio of binder to wetted surface area, as well as the temperature of the material in an intermediate storage area or before entering the press. In particular, material parameters include parameters of the material that can be recorded at the individual processing stages, from delivery to the application of the fleece to the finished panel.The material parameters can also include both target values ​​and measured actual values.

[0035] A dataset is a group of data that includes at least one quality value and one or more input parameters that correlate with that quality parameter. In particular, a dataset differs from another dataset in at least one quality parameter and / or in at least one input parameter that correlates with that quality parameter.

[0036] An advantageous design is characterized by the fact that the database comprises only calculated data sets, in particular that the artificial intelligence-based algorithm is trained or formed exclusively with a database consisting solely of calculated data sets. This allows the physical relationships within the AI-based algorithm or model to be trained, enabling reliable determination of quality parameters even for unknown input parameters.

[0037] Alternatively, or preferably, the database should also include at least partially known, real-world datasets. This means that not only simulated datasets are used to create the artificial intelligence algorithm, allowing the algorithm or model to better reflect reality.

[0038] In particular, the data sets constitute a data space. A data space is understood to be, in particular, an area in which the input parameters are adjustable and / or selectable within their physical limits. Within the data space, the parameters for the data sets or the database can be varied as desired.

[0039] A preferred embodiment is characterized by the fact that the calculated data sets are determined by means of a simulation based on at least one physical model. The simulation can, in particular, be at least a two-dimensional simulation or, preferably, a three-dimensional simulation. In particular, the simulation is based on a finite element simulation and / or a finite volume simulation and / or a finite difference simulation. In particular, a finite element simulation, and especially a finite volume simulation, allows data sets to be determined using the physical model in a short computation time.

[0040] An advantageous design is characterized by the fact that at least one physical model relates to the processes occurring in a press for the production of material sheets and at least partially replicates one or more of these processes. In particular, the processes occurring within the press in the production plant influence the quality of the material sheet being produced. Furthermore, these processes, or individual aspects such as curing curves, can be represented mathematically very accurately, at least in part.

[0041] Furthermore, in at least one physical model, input parameters are preferably varied as process parameters. For example, the input parameters include different thicknesses of material sheets, different material characteristics, different binder characteristics, and / or different material moisture content. By varying the process parameters, a broad spectrum of material sheets should be covered as much as possible in order to obtain a large number of calculated data sets. Selecting data from a wide range allows the database to be expanded, thereby improving the algorithm or the artificial intelligence-based model.

[0042] Alternatively, or preferably additionally, the at least one physical model considers various plant parameters of the production plant as input parameters. These include, for example, production speeds, pressures, and / or temperature curves. This also expands the database for various plant parameters as input parameters and, in particular, allows parameters outside the usual production ranges to be considered.

[0043] Another preferred embodiment is characterized by the fact that the at least one physical model is embedded in the artificial intelligence-based algorithm, in particular that the embedding is such that the physical model is essentially mapped within the algorithm. The physical model thus forms a core of the algorithm or model based on artificial intelligence. This ensures that known physical relationships are correctly represented, and the algorithm or model can account for deviations between a state of the idealized process model and a real-world state. The production of material sheets can be carried out safely within physical limits with particular advantage.

[0044] Preferably, the control system is designed to autonomously make changes to the input parameters. In particular, through a physical model as the core of the algorithm or model based on artificial intelligence, the control system can autonomously make process changes within the physical limits and take the resulting changes in relation to the quality parameter into account.

[0045] Furthermore, a quality parameter is preferably determined within the algorithm based on the physical model, whereby the quality parameter essentially specifies at least one standard deviation of the determined quality value. The quality parameter can thus provide information about the quality and probability of the determined quality value. This allows for further improvements in statements regarding the quality value of a manufactured material sheet.

[0046] Preferably, at least one physical model, in particular the mathematical description of one or more processes occurring during the production of material sheets, is fixed and unchangeable within the algorithm. The physical model thus forms an unchanging core within the algorithm, whereby the algorithm or the model, based on artificial intelligence, determines the aspects outside this core.

[0047] Alternatively, or preferably additionally, at least one physical model is modified by the algorithm and thus adapted from the idealized process model to reality. In particular, this allows the physical model to be further developed, whereby previously unknown relationships can also be taken into account by the algorithm or the model based on artificial intelligence.

[0048] A preferred embodiment is characterized by the fact that the input parameters to be provided for determining the calculated data using the physical model are varied within a range around a known input parameter of a known data set. In particular, the input parameters can be varied around a known data set from a laboratory section. For example, the input parameters for determining the data sets can deviate from the known input parameters by up to 5%, preferably up to 10%, resulting in the variation. Preferably, only individual input parameters for determining the data sets deviate from the known input parameters.

[0049] Preferably, the physical model is verified using at least one known data set, wherein the known data set is preferably based on a laboratory section of a material plate. Thus, the physical model can be verified using known data, particularly before the model is used to determine data sets.

[0050] The calculated datasets advantageously achieve data compression for the artificial intelligence algorithm. By considering physical relationships and, in particular, their mathematical description, the number of input parameters can be significantly reduced and / or the algorithm's computation time can be considerably shortened. The physical relationships are already taken into account by the algorithm or the artificial intelligence model and do not need to be learned by the algorithm itself.

[0051] Preferably, the calculated data sets are determined in a region of a data space for which only a few known, real-world data sets exist. In particular, depending on the process parameters input into the physical model, data sets outside the otherwise optimal or known range can be calculated by the physical model with regard to quality values ​​and / or input parameters. Specifically, the calculated data sets can be determined outside of a standard configuration. The known data situation in real-world data sets usually includes those with good or very good quality parameters. In regions with few real-world data sets, the number of data sets in the data space is therefore quite limited. Without corresponding data sets, the algorithm or the artificial intelligence-based model cannot be properly trained, resulting in incorrect representations of relationships.This can lead to quality parameters not being correctly determined using the algorithm or model based on artificial intelligence. By identifying datasets specifically for this part of the data space, the data pool is expanded, thereby improving the determination of quality parameters across the entire data space.

[0052] A data basis via a data space is particularly advantageous if it has a balanced number of calculated data sets and / or known, especially real, data sets. In particular, the calculated data sets and / or known, especially real, data sets cover the entire data space.

[0053] Preferably, the data sets are normalized over time and / or the length of the production plant and / or a device. This reduces further plant-specific influences on the input parameters, allowing the data to be processed directly.

[0054] Preferably, the input parameters are normalized and / or aggregated. Normalizing the input parameters should enable their universal application across a wide range of systems, thereby eliminating system-specific characteristics. In particular, the input parameters are normalized to a statistical mean of zero and a standard deviation of one. Aggregating the input parameters significantly reduces the amount of data to be processed, thus enabling faster processing. Aggregation is particularly preferably performed by interpolation, extrapolation, and / or averaging.

[0055] Alternatively, or preferably additionally, the input parameters are aggregated by forming clusters, in particular by forming clusters of input parameters within a device. In a cluster, several input parameters are thus combined into a single area, further reducing the amount of data. Aggregation is particularly preferably carried out by interpolation, extrapolation, and / or averaging.

[0056] A particularly advantageous feature is that the artificial intelligence-based algorithm provides an absolute statement regarding the quality value. This means that not only relative values ​​or changes in the quality value can be indicated. This allows for a direct comparison, for example, using values ​​from a laboratory section.

[0057] Preferably, known datasets are checked using the physical model or at least with information from the physical model. This allows the datasets to be verified, thus avoiding training the AI-based algorithm with flawed data. This filters out implausible data that would otherwise degrade the performance of the AI-based algorithm or model. This further improves the quality and, in particular, the statements regarding the quality parameters of the AI-based algorithm.

[0058] In another advantageous embodiment, at least one non-measurable process variable is determined by the algorithm based on artificial intelligence and this non-measurable process variable is used as an input parameter for the physical model.

[0059] Particularly advantageous is the offline simulation of changes to the manufacturing process for the material sheet using the algorithm. This allows the effects of changes on quality parameters or other parameters to be determined in advance of the actual process, thereby further improving production quality and process control for the manufacture of material sheets. In particular, material sheet rejects due to incorrectly chosen parameters can be prevented by prior simulation using the algorithm or model based on artificial intelligence.

[0060] Advantageously, the calculated values ​​from both the physical model and the algorithm are used as parameters for data-driven control. The production plant control system can, in particular, provide suggestions for parameter changes to an operator or a technologist. However, the production process can also be autonomously modified and optimized by the control system based on the physical model and the artificial intelligence-based algorithm.

[0061] Simulated quality values ​​are particularly advantageous for use in a control loop.

[0062] In a preferred embodiment, the quality value is optimized in an optimization computer of the control system using an algorithm based on artificial intelligence.

[0063] Optimizing the quality value preferably refers to improving the quality value itself, combined with optimizing the quality of the material sheet. Furthermore, optimizing the quality value can preferably also consist of achieving a consistent quality value under modified input parameters, particularly through reduced use of raw materials, auxiliary materials, and operating supplies, and / or enabling higher production while maintaining the same quality value.

[0064] The optimization of the quality value is always carried out to ensure that the determined quality value lies within a predefined range or does not fall below a predefined minimum quality value. A lower limit for the quality value ensures that the manufactured material sheets meet a minimum quality standard, thus preventing the production of rejects. The minimum quality value can, for example, be specified by a standard for the material sheets.

[0065] In terms of quality, optimization can be achieved by improving the quality value, maintaining it by reducing material usage, or even reducing it to a minimum value. Furthermore, quality can also be optimized by increasing the quantity, volume, and / or surface area of ​​material sheets per unit of time, provided these material sheets meet at least the minimum quality standard.

[0066] In particular, the modified input parameters for optimizing the quality value can be manually entered by a production plant operator, so that ultimately the operator performs the optimization themselves, and the optimization computer only provides suggestions for modified input parameters to optimize the quality value. The modified input parameters are then output to the production plant and / or its devices via the control system and physically changed accordingly.

[0067] Alternatively, or preferably additionally, the modified input parameters determined by the optimization computer can be automatically changed in the control system and output to the production plant and / or equipment after approval by the operator. By authorizing a change to the input parameters in the control system, the operator can verify the plausibility of the determined modified input parameters and, if necessary, correct them based on their experience and the timing of the change, particularly before switching to a different material sheet or similar. With fully automatic modification of the input parameters determined by the optimization computer in the control system, operator intervention is no longer necessary for optimizing the material sheet production process.

[0068] In a preferred embodiment, the algorithm is verified using a test database, wherein the test database comprises test data sets with quality parameters as well as input parameters correlating to the quality parameters.

[0069] In particular, the test data sets differ from the data sets of the database. Preferably, the test database is based on calculated test data sets derived from at least the physical model, especially the mathematical description of one or more processes occurring during the production of material sheets, and / or from known, real-world test data sets. The calculated test data sets can be determined using the same procedure as the calculated data sets.

[0070] As a further solution, a production plant for manufacturing material sheets exhibiting specific quality parameters is described. The production plant includes at least devices for forming a nonwoven fabric from at least one material and pressing it into material sheets, and preferably for applying adhesive to the material before forming the nonwoven fabric. The production plant and / or its devices comprise a control system with a programmable logic controller (PLC) for controlling or regulating the production plant and / or its devices. Input parameters can be received, processed, and / or output via the control system. The control system is configured to determine at least one quality value for at least one quality parameter based on the input parameters using an artificial intelligence-based algorithm. The algorithm can be trained using a database.can be formed, which includes at least one quality parameter and input parameters correlating to the quality parameter.

[0071] The production plant according to the invention is characterized in that the database comprises at least partially calculated data sets from at least one physical model, in particular a mathematical description of one or more processes occurring during the manufacture of material sheets.

[0072] A preferred embodiment of the production plant is characterized by the fact that the database comprises only calculated data sets. Alternatively, the database can also include at least known, real data sets as well as calculated data sets.

[0073] In particular, the data sets form a data space.

[0074] An advantageous design of the production plant is characterized by the inclusion of a display on which the determined quality value can be shown. Alternatively, or preferably additionally, the production plant includes an interface for accessing the database and / or for transferring information with a user of the production plant, a manufacturer of the production plant, another production plant, and / or a cloud.

[0075] Preferably, the control system also includes an optimization computer, which, based on the artificial intelligence-based algorithm, allows at least one quality value to be optimized.

[0076] Preferably, the production plant includes devices for comminution, sorting, adjusting the mixing ratios of the material, providing a binder, applying glue, storing, conditioning the material, weighing, monitoring and / or testing the material or nonwoven, pre-compression, adjusting the moisture, temperature, width and / or height of the nonwoven, measuring product and material parameters and / or quality parameters of the material sheet, regulating the temperature of the material sheet, or the like.

[0077] As a further solution, a computer program product is specified with a computer-readable storage medium on which instructions are embedded which, when executed by a controller, cause the controller to be configured to carry out a method for manufacturing material plates as described above or an advantageous embodiment thereof, in particular in a production plant as described above or an advantageous embodiment of the production plant.

[0078] Furthermore, as a further solution, the use of a computer program product as outlined above in a production plant for the manufacture of material panels made from lignocellulosic material, in particular wood, and / or from recycled material and / or from plastic is proposed, preferably for the autonomous production of material panels.

[0079] Further advantageous embodiments of the process for manufacturing material panels as well as of the production plant for manufacturing material panels are shown in the following figures.

[0080] They show: Fig. 1. A basic representation of a production plant for the manufacture of material panels; Fig. 2. A flowchart for training an algorithm using a database; and Fig. 3. A flowchart for optimizing the quality value during the production of material panels.

[0081] In the following, identical reference symbols denote identical or at least equivalent parts.

[0082] In Fig. Figure 1 schematically depicts a production plant 1 according to the invention for manufacturing material panels 12. The production plant 1 comprises several devices 2, 3, 4, 5, 6, 7, 8, 9 through which the material 10 passes in order to ultimately be pressed into a material panel 12. The material 10 consists primarily of a plant-based raw material, in particular wood, which is delivered to the production plant 1. Besides wood as material 10, other lignocellulosic materials, recycled materials, for example recycled wood or recycled plastic, or even plastics directly can be processed in the plant as material 10. Furthermore, the material 10 can also consist of a mixture of different materials, for example, virgin wood and recycled wood and / or plastic.

[0083] In the illustrated production plant 1, the provided material 10 is first added to a device 5 for comminuting the material 10. For example, the comminuting device 5 could be a cutting device, a chipper, or a ring shredder. The comminuted material 10 is then fed to a drying device 6, in which it is dried to a predetermined residual moisture content for the subsequent process. Before entering the drying device 6, the material 10 is mixed with a binder in a gluing device 3. The use of a binder depends on the specific material 10. When using plastics as material 10, the addition of a binder may be omitted, thus eliminating the need for a gluing device 3. Alternatively or additionally, as shown in Fig. As shown in Figure 1, a device 3' for gluing may also be arranged downstream of the device 6 for drying. If necessary, the material 10 can be re-glued in the downstream device 3' for gluing after the device 6 for drying.

[0084] The adhesive-coated material 10 then passes through a sorting device 7, in which it is separated into several fractions. In this case, the material 10 is fractionated based on the size of the comminuted material 10, whereby material 10 with a smaller particle size is separated from material 10 with a larger particle size and fed to separate processing lines or intermediate storage areas. Fractionation of the material 10 in the sorting device 7 aims to improve the production of the material sheet 12, for example, by arranging the material 10 with a smaller particle size on the outer surfaces of the material sheet 12, while coarser material 10 forms a middle layer of the material sheet 12. However, fractionation of the material is not strictly necessary, especially if a homogeneous material sheet is desired.

[0085] The material 10 is then fed to a device 2 for spreading the material 10. In this device 2, the material 10 is spread onto a forming belt to form a nonwoven fabric 11 or a mat, with the fractions of the material 10 being spread in layers so that the nonwoven fabric 11 has a layered structure. Alternatively, the material 10 can also be spread onto a forming belt in a single layer. The nonwoven fabric 11 thus spread then passes through, as described in the Fig. Figure 1 shows a pre-compression device 8 in which the nonwoven fabric 11 is pre-compressed. In the device 8, the nonwoven fabric 11 is thus compressed on the one hand and de-aerated on the other, thereby enabling improved compression of the nonwoven fabric 11 against the material plate 12 in the pressing device 4. Depending on the material 10 used, for example, the use of plastic as material 10, a pre-compression device 8 may not be necessary.

[0086] In the pressing device 4, which is designed as a continuous press, the nonwoven fabric 11 is subjected to pressure and heat, causing the binder in the nonwoven fabric 11 to cure and forming the material sheet 12 or a continuous strand of material sheets at the end of the pressing device 4. Optionally, a further device for preheating the nonwoven fabric 11, for example by means of steam or microwave radiation, can be arranged upstream of the pressing device 4. The continuous strand of material sheets exiting the pressing device 4 is then cut in a separating device 9 using diagonal saws, producing material sheets 12 of the desired length. The material sheets 12 thus formed are then cooled, stacked, and sent to storage or for immediate further processing.

[0087] Production plant 1 further includes a control unit 20, which is directly connected to the individual devices 2 to 9 of production plant 1 as well as other equipment, such as measuring devices or temperature sensors. The control unit 20 comprises a programmable logic controller (PLC) 21, by means of which production plant 1 and devices 2 to 9 are controlled or regulated. Through the functional connection between the control unit 20 and the individual devices 2 to 9 of production plant 1, input parameters 23 are acquired by the control unit 20, and modified input parameters 28 are transmitted to production plant 1 and the individual devices 2 to 9. The input parameters 23 are setpoints and actual values ​​of plant parameters of production plant 1 and devices 2 to 9.These include, for example, the temperature in the drying device 6, the fill levels of intermediate storage tanks, the press speed at which the nonwoven fabric 11 is compressed in the device 4, and the temperatures and pressures in the press in device 4. Some of the system parameters are recorded by sensors on the individual devices 2 to 9, while others are defined by fixed parameters that characterize the individual devices 2 to 9 as well as the production plant 1. Fixed system parameters include, for example, the press length of the continuously operating press (device 4), the number of spreading units in device 2, and the number of fractions in device 7 for sorting.

[0088] In addition to the system parameters, the material parameters also constitute further input parameters 23, which are fed into the control system 20. These material parameters include the type of material 10 used, the residual moisture content of the material 10 after drying in the device 6, the binder used in the gluing device 3, and the height and width of the spread nonwoven fabric 11 in and / or after spreading in the device 2. Ambient conditions such as outside temperature or humidity can also be fed into the control system 20 as input parameters 23.

[0089] Finally, further input parameters 23 can also be specified by an operator. For example, product parameters that characterize the material sheet 12 are also input parameters 23. These product parameters are predefined parameters of the material sheet 12, which is to be manufactured in the production plant 1. Thus, certain minimum requirements for the material sheet 12 are defined by the product parameters, which are also incorporated into the control system 20. The product parameters include the type of material sheet to be manufactured and its thickness.

[0090] The controller 20 is designed to receive and process these input parameters 23 and to output modified input parameters 28 to the devices 2 to 9 of the production plant 1. Furthermore, the controller 20 is designed to determine a quality value 24 based on the input parameters 23 and to display this value on a screen 26 for an operator or technologist of the production plant 1. The controller 20 also includes an optimization computer 22, which optimizes the determined quality value 24. The parameters recorded in the controller 20 are transmitted via an interface 25 to other devices, such as a memory 27, additional displays, or a cloud.Preferably, interface 25 can also be used to establish a connection to a system of the manufacturer of production plant 1 and / or of devices 2 to 9 of production plant 1, via which data from production plant 1 and / or devices 2 to 9 are transmitted and data is transferred to the controller 20. The controller 20 can also receive data, in particular data sets or test data sets, information and / or parameters from other production plants 1 via interface 25, especially if the manufacturer of the material plates 12 operates several production plants 1.

[0091] In Fig. Figure 2 schematically illustrates the process for creating and training an algorithm 32 or model based on artificial intelligence to determine at least one quality value 24 of a quality parameter. To train the algorithm 32, a method on which the algorithm 32 is based is first defined in the controller 20 or selected from a list. In this case, the method of neural networks is used as the basis for the algorithm 32. In order to determine or predict a quality value 24 for a quality parameter using the algorithm 32, it is necessary that the algorithm 32 is trained.

[0092] For training algorithm 32, a database 30 is provided, which comprises a large number of data records. The individual data records of database 30 include information on quality parameters as well as on the input parameters that correlate with the quality parameters.

[0093] The data basis 30 is based on known, real data sets 30b as well as on calculated data sets 30e. Alternatively, the data basis 30 can also consist solely of calculated data sets 30e. The known, real data sets 30b originate primarily from known laboratory sections or from other production facilities.

[0094] The calculated data sets 30e are those determined by means of a physical model 35, in particular a mathematical description of one or more processes during the production of material sheets 13. The physical model 35 is based on one or more simulations, by means of which quality values ​​are determined from input parameters and corresponding calculated data sets are generated. In this case, the simulation is a finite volume simulation, which can quickly and reliably determine the corresponding calculated data sets 30e. Furthermore, the simulation is at least two-dimensional. Alternatively, a three-dimensional simulation of the processes can also be used mathematically, whereby, for example, the width of the press mat or the material sheet is considered as the third dimension.The simulation describes, in particular, the idealized process conditions and process sequences within the press-type device 4, for example, curing curves or compression curves. However, other processes in other devices or idealized processes in the entire production plant 1 can also be simulated using the physical model 35.

[0095] To generate calculated data sets 30e, input parameters in the physical model 35 are varied. These can be product parameters, plant parameters, material parameters, or environmental conditions. In particular, plant parameters of the production facility, such as production speeds, pressures, and temperature curves, are varied, and further calculated data sets 30e are generated from this. The variation of the parameters is primarily based on a known data set 30b. For example, the input parameters for determining the calculated data sets 30e can deviate from the known input parameters by up to 5%, preferably up to 10%, resulting in the variation.

[0096] The calculated data sets 30e allow for the creation of a broad database 30 based on known physical relationships, independent of known, real-world data sets 30b. This broadens the overall database 30, enabling the algorithm 32, trained on the database 30 and based on artificial intelligence, to determine and provide more reliable values ​​across the entire range of a data space in which the input parameters are adjustable and / or selectable within their physical limits.

[0097] The database 30 should be designed such that it comprises a balanced number of calculated data sets 30e and, if applicable, also known, real data sets 30b across the entire data space. Since known data sets 30b are often only available for good or very good quality parameters, the physical model 35 is used to determine calculated data sets 30e, particularly in a region of the data space where there are few known, real data sets 30b. This ensures a balanced number of data sets across the entire data space. In particular, the physical model can also calculate data sets 30e outside the otherwise optimal or known data with respect to the quality values ​​24 and / or input parameters 23, which are then incorporated into the algorithm 32 based on artificial intelligence. Specifically, the calculated data sets 30e may lie outside certain standard configurations.This allows production plant 1 to be operated more stably across the entire data space.

[0098] The numerous data records 30b, 30e in database 30 are structured such that they differ in at least one parameter, a quality parameter, or an input parameter correlated with the quality parameter. Furthermore, data records 30b, 30e in database 30 are structured to cover the broadest possible range of product parameters with respect to data on the product parameters, which form a subset of the input parameters. For material plates 12, the product parameters intended to cover the broadest possible range are defined by the plate thickness and the type of material plate 12. This ensures that the algorithm 32 trained by database 30 can reliably determine a quality value 24 for a quality parameter across the broadest possible range of product parameters, and moreover,

[0099] The physical model 35 can be embedded within the controller 20, as shown in Fig. Figure 2 is shown. Alternatively, the physical model 35 can also exist outside the controller 20, with the calculated data sets 30e then being transferred to the database 30 via an interface 25. Similarly, the real data sets 30b can also be provided via the interface 25. In this process, known data sets 30b from other production plants can also be taken into account, for example, from other production plants of the same manufacturer.

[0100] Before the algorithm 32 is trained with the database 30, the data of the database 30 can be normalized and aggregated in a processing step 31, as schematically indicated by the dashed arrow in Fig. 2 is shown. However, this processing is only optional.

[0101] After completing the training of algorithm 32 with data set 30, the resulting algorithm 32 can be verified in a further step using a test data set 33. Test data set 33 again comprises test data sets, which are structured analogously to the data sets in data set 30. The test data sets in test data set 33 may also have undergone processing before use, for example, by normalizing and aggregating them. In particular, the test data sets in test data set 33 may also have been calculated using the physical model 35. This allows for particularly effective verification of algorithm 32.

[0102] The input parameters stored in the test data sets of test database 33, which correlate with the quality parameters, are provided to algorithm 32 as input parameters. Based on these input parameters, algorithm 32 determines a quality value 24 for at least one quality parameter. Specifically, algorithm 32 determines quality values ​​24 for multiple quality parameters. During the verification of algorithm 32, quality values ​​24 are determined for quality parameters stored in the test data sets of test database 33.

[0103] The quality values ​​24 determined using algorithm 32 are then compared with the data on the quality parameters stored in the test database 33 for the corresponding test data set. This ensures that the trained algorithm 32 can also determine reliable values ​​for the quality parameters.

[0104] If, using the test data base 33, there are larger deviations between the data on the quality parameters stored in the test data base 33 or the test data sets and the quality values ​​24 determined by the algorithm, algorithm 32 can be retrained using an extended data base 30 or even fundamentally rebuilt, whereby in the case of a new construction of algorithm 32 it can also be based on a different method, for example K Nearest Neighbor Regression.

[0105] To further improve the quality of the artificial intelligence-based algorithm 32, the physical model 35 is verified using at least one known dataset before it is used to determine the calculated datasets 30e. This at least one known dataset should be based on data from a laboratory section. Furthermore, known datasets 30b can also be checked for plausibility using the physical model before being used in the database 30. This allows, in particular, erroneous known datasets 30b to be filtered out, thereby improving the database 30 overall, which in turn improves the artificial intelligence-based algorithm 32.

[0106] In a preferred embodiment not shown, the physical model 35 is embedded in the algorithm 32 based on artificial intelligence. This is intended to ensure that the physical relationships within the algorithm are represented substantially correctly. This can also be achieved, for example, by training or forming the algorithm solely with calculated data sets 30e. The physical model within the algorithm can then be adapted, in particular to the actual conditions compared to the assumed idealized process model underlying the physical model 35.

[0107] By combining physical model 35 and artificial intelligence-based algorithm 32, it is possible to specify quality parameters that allow for an assessment of the quality of the determined quality values ​​of algorithm 32. In particular, at least one standard deviation can be determined. Furthermore, compared to the physical model, the combination also allows for absolute statements regarding the quality parameter.

[0108] In Fig. Figure 3 schematically illustrates the optimization of a determined quality value 24 in the controller 20. Input parameters 23 are fed into the algorithm 32, based on which the algorithm 32 determines one or more quality values ​​24. The input parameters 23 provided for determining the quality values ​​24, as well as the determined quality values ​​24, can be stored as a production data set in a memory 27. This makes it possible to document the determined quality values ​​24 for the input parameters and, if necessary, to verify them against subsequently measured values ​​for the quality parameters.

[0109] The determined quality values ​​24 are now transferred to an optimization computer 22, which optimizes the determined quality value 24. The optimization within the optimization computer 22 is based on an artificial intelligence algorithm, which in this case is identical to the artificial intelligence algorithm 32. In particular, it is the same algorithm 32. Alternatively, the algorithm of the optimization computer 22 can be different from algorithm 32.

[0110] The optimization of the quality value 24 in the optimization computer 22 is subject to certain boundary conditions, which are specified by an operator or which are limited by physical limits or the algorithm.

[0111] Optimizing the quality value 24 can be based on a variety of aspects. The optimization computer 22 is designed to optimize and improve the quality value 24 itself. The goal is to produce the highest possible quality material plate 12.

[0112] Optimization of the quality value 24 can also be achieved by maintaining the quality value 24, but by changing the input parameters 23 to achieve the quality value 24, thereby requiring less raw material, auxiliary, and operating material or enabling a higher output of material sheets 12 per unit of time. Thus, while maintaining the same quality (given a roughly constant quality parameter 24), resource consumption will be reduced, making the production of the material sheets 12 more cost-effective.

[0113] Furthermore, the quality value 24 in the optimization calculator 22 can also be optimized by reducing it to a minimum value, thereby further reducing the use of raw materials, auxiliary materials, and operating supplies. The material panels 12 thus meet minimum requirements, while simultaneously reducing resource consumption and making the material panels 12 more cost-effective to manufacture.

[0114] To optimize the quality value 24, modified input parameters 28 are determined in the optimization computer 22, based on which the desired quality value is to be achieved. The determined modified input parameters 28 are then displayed on the display 26 to an operator of production plant 1 by means of a message on the display 26. The operator will then check the modified input parameters 28, correct them if necessary, and transmit them to the programmable logic controller (PLC) 21 or enter them there manually. From the PLC 21, the determined modified input parameters 28 are transferred to production plant 1 or the devices 2 to 9 of production plant 1 and set accordingly. A transfer of the corresponding modified input parameters 28 is in Fig.3 is represented by dashed lines. However, the operator can also refrain from optimizing the quality value 24 using the determined changed input parameters 28, for example, if circumstances such as upcoming maintenance or a product change are known to him and optimization would therefore no longer be productive.

[0115] Alternatively, the determined modified input parameters 28 can be changed directly in the programmable logic controller 21 and then transferred to the production plant 1 or the devices 2 to 9 of the production plant 1 for changing the respective settings. The production of the material sheets 12 in the production plant 1 is thus continuously and fully automatically optimized via the modified input parameters 28 determined in the optimization computer 22. The production plant can therefore be operated autonomously. By integrating the physical model 35 into the algorithm 32 or by creating the database 30 from at least partially calculated data sets 30e, it can be ensured that the production plant can be operated safely and optimally within certain physical limits, especially without human intervention.In particular, modified input parameters 28 can be determined and transferred directly to the devices 2, 3, 4, 5, 6, 7, 8, 9 in order to improve the quality parameters or the economic efficiency of the production plant 1.

[0116] Before implementing changes, in particular the transmission of modified input parameters 28 to the devices 2, 3, 4, 5, 6, 7, 8, 9 of production plant 1, the effects of the modified input parameters 28 within production plant 1 can be simulated offline using algorithm 32 based on artificial intelligence. This allows the modified input parameters 28 and their effects to be checked again, ensuring a smooth process flow in production plant 1. Reference symbol list 1 production plant 2. Device for spreading 3, 3' Gluing device 4 Pressing device 5 Device for shredding 6. Drying device 7 Sorting device 8 Device for pre-pressing 9 Device for separating 10 Material 11 Fleece 12 Material board 20 Control 21 programmable logic controllers 22 optimization calculators 23 Input parameters 24 Quality value 25 Interface 26 ads 27 storage 28 Changed input parameters 30 database 30b known data sets 30 calculated data sets 31 Processing 32 Algorithm 33 Test database QUOTES INCLUDED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature

[0000] WO 2021 / 165394

[0007]

Images